Meshy Release Notes

Meshy 7.1 generates the richest geometric detail of any image-to-3D model. Ultra 4K scales geometry to 4096³, measured by our new Detail Richness benchmark.

TL;DR

Meshy 7.1 targets one thing: geometric detail. A new Ultra 4K mode scales geometry generation from 2048³ to 4096³, as of this release the only 3D generative model to reach it, with raw meshes up to 80 million triangles before simplification. On our new Detail Richness benchmark, Meshy 7.1 ranks first at 1K, 2K, and 4K render resolutions. Rolling out from September 10, 2026.

We release Meshy 7.1 today as an upgrade to Meshy 7. We built the model around one goal: generating the richest geometric detail of any available image-to-3D model. As alignment and structural correctness approach production quality across 3D generative AI models, the surface itself becomes the frontier. Armor should carry its engraving, a dragon should carry its scales, and a carved pillar should carry its relief, all within the geometry.

With Meshy 7.1, we deliver the generation mode with the finest detail, Ultra 4K. Our internal evaluations show that it generates the richest surface detail among all available models on the market at 1K, 2K, and 4K render resolutions.

Detail Richness scores at each render resolution. All results were generated using the maximum quality settings available for each generative model.

Generate Better to Look Closer

As 3D generative models improve, people look closer and closer at the assets the models produce, and how close is really enough? This is the question we would like to answer with Meshy 7.1.

Inferring a 3D shape from a single 2D image is the core capability that sets a 3D foundation model apart from other generative models. The generated shape should not only reproduce what the image shows faithfully with every part in its correct place and proportion, but it also should reproduce that shape in full detail down to the finest structure on the surface. Both parts matter to the quality of the generation result, because an accurate generation with low precision gets the overall form right but misses the details that carry the asset.

Therefore, Meshy 7.1 specifically targets the generation of geometry details. Meshy 7.1 substantially improved its generation stage dedicated to detail. The model inherits the strong image understanding capability from Meshy 7, and further scales the generation resolution from 2048³ to 4096³ with detail enhancements. According to our comprehensive tests, Meshy 7.1 is the only 3D generative model on the market that is able to obtain a 4096³ geometry resolution to date.

To achieve the increase in voxel resolution and enable the model to generate in a massive space with billions of voxels, we have scaled up our geometry representation and generation backbone, with efficient high-precision mesh extraction and simplification algorithms to support ultra-high face count. A raw mesh directly from the model, without any simplification, can contain as many as 80 million triangles.

How We Measure Detail

Along with the release of Meshy 7.1 and Ultra 4K geometry quality, we developed a Detail Richness benchmark to measure how much detail a generated 3D model carries. The benchmark is done by rendering each generated model from multiple viewing directions at a fixed screen resolution, then measuring the fraction of its surface whose orientation varies significantly inside a single pixel.

Surface detail may seem like a subjective metric, because you recognize it the instant you see it. However, the computer graphics industry has spent decades on making that judgment precise. If you play games and have tweaked graphics settings, you have probably run into "anti-aliasing". You switch it off, and the details across the scene turn boxy and jagged, even though the screen resolution has not changed at all. You switch it back on and the details turn smooth again.

That is exactly the foundation of our metric. A single pixel on the screen usually covers an area of the 3D surface, and anti-aliasing processes surface variation inside one pixel. Where the area is flat, the pixel shows one clean shade and no anti-aliasing is needed. Where there are finer details than the pixel, anti-aliasing looks in a finer scale than the pixel and ensures that the fine details correctly contribute to the pixel. Our Detail Richness benchmark simply measures: when rendered at multiple directions, how much of the model needs to be anti-aliased at a given render resolution. The higher the render resolution, the less anti-aliasing is necessary, and the finer the details the metric measures.

To keep the comparison objective, we specifically constructed the benchmark input set using images with very rich structural details but very simple textures, similar to a high-precision single-color 3D print model. This allows us to rule out the biggest ambiguity in detail comparisons. When an input image shows a rich colored texture on a smooth surface, the generative model must decide whether to sculpt it into geometry or assign the pattern to texture. Using such images as input could easily make a model that correctly recognizes texture patterns look like one that fails to generate geometry details.

Therefore, the input image of this benchmark ensures that every detail visible in the input represents a geometric fact. The task for the generative model is therefore unambiguous: reproduce every detail as geometry.

Richest Details at Every Resolution

Meshy 7.1 ranks first across every render resolution we measure, and the ranking holds from the coarsest scale to the finest.

At render resolution 1024, Meshy 7.1 keeps 53.5% of the visible surface detailed, 2.1 points ahead of Rodin 2.5, the next model. At 2048 render resolution, it keeps 37.4%, again 2.4 points ahead of Rodin 2.5. At 4096 render resolution, structure at engraving scale comes into view and every model loses most of its detailed surface, but Meshy 7.1 holds 23.1% while no other model reaches 22%, the closest being Hi3D 3.0 at 21.9%. Meshy 7.1 achieves a consistent lead in Detail Richness when compared with other generative models on the same object. When compared object by object at the finest render resolution, Meshy 7.1's win rate on Detail Richness ranges from 56% to 98%.

In the score distributions per object, the spread of scores across objects is wide for every model because some objects carry more detail than others. The Meshy 7.1 distribution sits to the right of every other generative model at every render resolution. Its median object is richer than the median object of any other model, and its top quarter reaches higher than the top quarter of any other model.

Besides the benchmark scores, we also show three examples of Ultra 4K generation. Meshy 7.1 Ultra 4K generated each model below from a single image. We rendered them with a clay material, so every ridge, stitch, and scale you see is real geometry built by the model.

The orc warrior shows this detail in cloth and rope. Every twist of the cord around the forearm guard is clearly visible under the embossed diamond, and the skirt shows its woven threads all the way down to the torn, frayed edge.

On the wooden hand cannon, the wood grain forms raised rings around each knot on the barrel, and two clear rows of stitches run along the full length of the leather wrist strap. A texture could only paint these features. Ultra 4K builds them into the mesh.

The armored war beast combines different kinds of surface detail in one model. Scales cover the skin and become smaller from the body toward the legs, a face emblem stands out from the saddle plate inside a ring of studs, and the rope around the back leg keeps every knot while it holds a tied bone in place.

A generative model that only gets the overall shape right would still produce the cannon, the orc, and the beast. But it would not produce the stitching, the weave, or the scales, and those details are what turn a plain shape into a production asset.

More Detail Without More Noise

Our Detail Richness score is based on tracking surface variation. However, not all surface variation originates from true geometric details in real-world 3D assets. For example, a 3D asset may score very high on Detail Richness simply because its surface contains too much high-frequency random noise. To ensure that our improvement in Detail Richness is due to true geometric details instead of noise, we also measured how organized the surface variations are as an indicator of detail quality. Organized structure, including creases, ridges, or engraved lines, bends a surface consistently and scores high. Disorganized noise bends the surface randomly and scores low.

We found that Meshy 7.1 successfully increases details without introducing more noise. Across three generation settings, Standard, Ultra 2K, and Ultra 4K, we did not observe a significant decrease in detail quality as Detail Richness increased. The figure below displays the same orc character input image generated at three different geometry resolution settings. In close-up views, details like fabric weaves become much better formed and sharpened in Ultra 4K, while no visible noise or bumpiness can be seen on flat surfaces.

Increasing resolution means the model spends more computation effort, and in Meshy 7.1 we ensure that the effort is correctly spent on the true details.

We also discovered that further scaling geometric resolution beyond 4096³ will likely provide little to no additional benefit to generation quality perceivable by our users. At a screen resolution of 4096 × 4096, the benchmark detects surface variation as fine as the human eye can perceive. When calibrating our benchmark, we found that the 3D model compression format commonly used for Web3D streaming created compression noise that caused a significant false increase in Ultra 4K Detail Richness scores, even though its visual loss was barely visible upon close inspection. Therefore, we have to evaluate all generative models on uncompressed geometry, while our previous benchmarks did not have to enforce this constraint. If we pushed geometry resolution further, we would only produce details too small to be distinguished by the human eye or preserved in the saved 3D model file.

Availability

Starting September 10, 2026, Meshy 7.1 rolls out gradually in the Meshy workspace and will reach all users soon after.

Meshy 7.1 is open to all registered users, while only subscribed users are able to access Ultra 2K and Ultra 4K resolutions as well as download models generated by Meshy 7.1. You can compare tiers on the pricing page.

Ultra 2K is available for both single-view and multi-view generation, and Ultra 4K is available for single-view generation only.

The Detail Richness benchmark introduced in this article will be released separately as a standalone tool.

Detail forms the difference between a generated asset and a finished asset. Meshy 7.1 puts the detail to the eye's limit.

Original source

Meshy 7 is our new image-to-3D model, tuned for geometry alignment: how closely the generated 3D shape matches your input image. Benchmark and results inside.

TL;DR

Meshy 7 is our new image-to-3D model, tuned for geometry alignment: how faithfully the generated 3D shape matches the input image. On our new automatic 3D geometry alignment benchmark — scored directly from geometry against held-out reference models along overall proportion, spatial distribution, and surface details — Meshy 7 leads all tested models under single-view input.

Today we release Meshy 7, our new image-to-3D generative model and the successor to Meshy 6. Meshy 7 sets a new state of the art in 3D geometry alignment: how faithfully the shape and structure of the generated 3D model match the input image.

As output quality of 3D generative models approaches production-ready and severe geometry errors become increasingly rare, what sets models apart is no longer just whether they can generate a valid 3D asset. It is whether that asset actually matches the object the user asked for. Our evaluations show that Meshy 7 achieves leading geometry alignment in single-view generation, leading all tested models on all three metrics, and remains among the leading models when four views are available.

To measure this capability, we developed our first 3D geometry alignment benchmark. This article introduces the benchmark methodology and reports its results; the benchmark itself will be made available separately after the Meshy 7 release.

[Table of benchmark results omitted for brevity]

Geometry alignment scores across two input conditions. Tested using reference 3D models excluded from training. 100% = the reference model measured against itself.

Why Geometry Alignment Matters Now

Image-to-3D generation has improved rapidly. Two years ago, severe failures were common: broken geometry, nonsensical shapes, missing limbs, or objects that were simply unusable. Today, such failures have become rare enough in leading 3D generative models that we treat every one we find as an error case worth investigating.

That changes what matters.

A model can generate a clean and plausible 3D asset and still generate the wrong asset. The torso can be too wide. A handle can sit too far to the left. A ridge visible in the concept can disappear. The result may look perfectly reasonable on its own while failing to reproduce the user's original idea.

That difference is geometry alignment.

For many 3D generation workflows, the user's idea begins in 2D: a concept image, illustration, photograph, or other visual reference. Users do not merely want a plausible object from the same category. They want the resulting 3D geometry to agree with that specific image as closely as possible.

Geometry is only one part of image-to-3D alignment. Texture presents its own alignment problem: whether colors, materials, patterns, and appearance agree with the reference image. In this article, we focus specifically on geometry alignment. We will introduce a separate benchmark for texture alignment later.

As baseline geometry quality approaches production-ready, geometry alignment increasingly separates a strong 3D model from one that merely avoids obvious failures.

With Meshy 7, we therefore made geometry alignment a primary target. We improved it in three places at once.

First, Meshy 7 looks at the input image more carefully. We improved the image encoder in the Meshy 7 architecture to read features at multiple scales and accept higher-resolution images. This allows finer shape information in the input image to survive into the generative model.

Second, Meshy 7 learns from better-aligned data. We rebuilt the training data around a stricter quality standard: every training sample corresponds precisely to its target geometry, while the effects of style, lighting, and background are systematically removed.

Third, geometry alignment is now part of our training evaluation loop. Throughout Meshy 7 development, we use geometry alignment as a key evaluation signal across training cycles. The capability we want to improve is therefore also one of the quantities we directly measure while developing the model.

But making geometry alignment a training target creates another problem: how do we measure it rigorously?

How to Measure 3D Geometry Alignment

With the release of Meshy 7, we are also introducing our first 3D geometry alignment benchmark. The benchmark is scored automatically from geometry, rather than by human preference or a VLM judge. We start from reference 3D models specifically held out from our training set, render images from them using known cameras, and use those images as inputs to each 3D generative model. We then compare the generated geometry directly against the original reference geometry. Because we know the source 3D model, every test case has a known correct 3D answer.

Our benchmark design avoids a fundamental ambiguity in image-space evaluation: a single 3D object can produce very different images depending on viewpoint, focal length, distance, lighting, and stylization. If we had to estimate the camera before comparing a generated model with its input image, camera-estimation errors would become mixed with geometry-generation errors. Starting from a known 3D reference and known camera lets us isolate the geometry alignment we actually want to measure.

For each test case, we first line up the generated model with the reference through translation, rotation, and uniform scaling. We never stretch or squash individual axes, so incorrect proportions cannot be hidden during alignment. We then measure agreement along three dimensions: overall proportion, spatial distribution, and surface details, moving from the object's broad structure to increasingly fine geometric differences.

[Illustration of the three geometry alignment metrics]

Overall proportion measures whether the generated model occupies the same region of space as the reference at the level of major parts. This metric sees the object at its coarsest level: the overall silhouette, major limbs, and broad mass distribution. If the reference fills a short and wide region of the grid and the generation fills a tall and narrow one, the overall proportion metric catches the mismatch.

Spatial distribution compares on a finer scale how much the geometry of the generated model agrees with the reference. It measures how far the generated model's material sits from where the reference places it, tallying the total effort needed to move every piece of the generated geometry onto the reference's placement. Small displacements cost little, while material that lands far from where it belongs costs more in proportion to the distance. A perfect score means nothing needs to move at all. A score of zero means the material is displaced as far as if it belonged to a different object entirely. Two models can fill the same large-scale region while distributing geometry differently within it: a torso too thin on one side or a handle mounted slightly too far left. By capturing these placement errors as a cost that grows with distance, spatial distribution catches shifts that proportion's coarser view would miss.

Surface details measure how close the surface of the generated model is to the reference. It catches any disagreement in details, such as a hallucinated ridge or bump the reference does not have, or a carved pattern smoothed away or a bolt that is missed in the generation.

Building the Benchmark

We selected reference 3D models across a wide range of assets, including characters, weapons, vehicles, sculptures, and props with thin structures. These models are excluded from our own training.

We test the same objects under three input conditions, increasing the amount of visual information available to the generative model.

Front view provides a single rendering taken directly in front of the object. This is the strictest condition. The model sees one angle and must infer the entire 3D object, including every side it cannot see.

Top-quarter view provides a single rendering from an elevated angle, revealing more of the object's top and side surfaces. The model still receives only one image, but more of the underlying 3D structure is visible.

Four-view input provides front, left, back, and right renders. The model sees the object from every horizontal direction, substantially reducing the amount of geometry it has to infer.

These conditions let us measure not only how well each model aligns with the reference, but also how effectively it uses additional visual information.

State-of-the-Art Geometry Alignment

Under the top-quarter view, the single-view condition that reveals the most of the object, Meshy 7 leads the tested models across all three geometry-alignment metrics.

The differences are largest on the harder parts of the benchmark. Overall proportion is already relatively strong across leading models, so the scores are compressed. Surface detail remains much harder, and it is also where Meshy 7 shows the largest advantage over previous Meshy models and the other models we tested.

That pattern reflects how the problem itself is changing. As 3D generation improves, progress increasingly moves from getting the broad object right toward preserving the geometric details that make one specific object different from another.

[Score distributions under the top-quarter single-view condition]

The front-view condition is the strictest single-view test. The model sees one side and must infer everything hidden behind it. Meshy 7 leads here across the overall statistics and across each of the three geometry-alignment metrics. The same pattern therefore holds across both of our single-view conditions.

[Score distributions under the front-view condition]

When all models receive four renders instead of one, alignment improves across the board, but the gains differ considerably.

Meshy 7 improves substantially, especially in spatial distribution and surface details. These are the two dimensions most affected by geometry hidden from a single viewpoint. Once additional views reveal that geometry, the scores respond accordingly.

Other models also benefit strongly from multi-view input. Tripo 3.1 in particular closes much of its single-view gap and approaches Meshy 7 under the four-view condition. The remaining models fall further behind, while the previous Meshy model benefits relatively little from the additional views.

With four views, the amount of geometry a model must infer drops sharply, and the leading models land within a narrow band. Meshy 7 posts the top score on spatial distribution, the metric most sensitive to whether individual parts sit where they belong.

The benchmark therefore tells us more than which model ranks first. It shows where geometry disagrees with the input and how those errors change as the model receives more information.

[Score distributions under four-view input]

Where Meshy 7 Shines

The improvements measured by the geometry alignment benchmark are also visible directly in the generated assets. The examples below highlight three strengths in particular: preserving facial expression, placing many parts correctly within a complex object, and carrying dense surface relief into clean 3D geometry.

Facial Expression and Character

Faces are one of the hardest tests of geometry alignment, because even small geometric changes can alter the expression immediately. Everyone notices a wrong brow or jaw at a glance.

In the bust examples below, Meshy 7 does more than reproduce a generic elderly face. It carries the specific expression of each reference image into the generated geometry. In the smiling portrait, the cheeks lift, the eyes narrow, and the smile lines deepen. In the stern portrait, the brow tightens, the mouth turns downward, and the folds around the nose and lips become sharper. In the eyes-closed portrait, the eyelids settle naturally and the face relaxes into a quieter expression. In the laughing portrait, the mouth opens wide, the cheeks rise, and the whole face compresses upward in the way laughter does.

[Four Meshy 7 bust generations matching four reference expressions]

What makes these examples strong is not only the likeness of the face, but the way the expression is preserved through geometry. Across all four generations, the headscarf, tied bow, and blouse remain stable, while the facial structure changes in the same direction as the source image changes.

Placing Mechanical Components

The mechanical owl shows a different alignment challenge. Here, the difficulty is not facial expression, but organizing many distinct parts into the right structure.

[Meshy 7 mechanical owl generation compared with its concept art]

Concept art (inset), the generated geometry, and the textured result. The gear-filled chest cavity, layered wing plates, thin legs, and sculpted base all survive as separate components in the geometry.

The design is built from layered wing plates, a circular eye housing, decorative chest panels, partially exposed gears, thin legs, and a sculpted base. Meshy 7 preserves the object as a structured assembly rather than reducing it to a simplified bird-shaped form. The major components remain clearly separated, and their relative positions stay close to the concept image: the gear-filled cavity sits in the center of the torso, the eye housing stays prominent, the wing panels wrap the body cleanly, and the feet and base retain their designed shape.

This is the kind of example where geometry alignment depends heavily on spatial distribution. The object only looks right when each part sits where it belongs.

Preserving Dense Surface Relief

The jade coin is an extreme test of fine geometric detail. Its shape is defined almost entirely by shallow relief: a coiling dragon around the rim, dense cloud forms, and a raised central character, all packed closely onto a nearly flat circular surface.

[Meshy 7 jade dragon coin generation compared with its reference]

Reference image (inset), the generated geometry, and the textured result. The dragon stays continuous around the rim, the cloud forms stay separated rather than collapsing into noise, and the central character remains legible as raised relief.

Meshy 7 preserves this structure clearly in 3D. The dragon remains continuous around the coin rather than breaking into disconnected fragments. The cloud forms stay distinct instead of collapsing into noise. The central character remains legible as an embossed geometric feature, and the outer ring keeps its circular composition intact.

This is the regime where surface details matter most. The coin does not depend on large-scale shape differences to look correct. It depends on whether many small geometric features survive together with the right structure and relative placement.

Together, these examples show the same pattern reflected in the benchmark: Meshy 7 aligns well at multiple scales, from broad facial and object proportions, to the placement of individual components, to dense local surface detail.

Where Quality Is Still Limited

What still goes wrong depends less on what the object is than on what kind of geometry it contains. Two limits show up again and again in otherwise strong generations.

[Meshy 7 limitations on loose hair strands and exactly repeating floors]

Two recurring limits, marked with dashed circles. Left: loose hair strands fuse into thicker clumps and ribbons. Right: the floor bands of a repeating tower wobble and drift out of alignment.

The first limit is hair, the hardest kind of thin, tangled structure. In the illustration above, Meshy 7 rebuilds the entire scene from a loose watercolor: the pose, the denim jacket and its folds, the apple and its leaf, even the splash at the base. The loose strands framing the face are the exception. Where the drawing scatters single wisps, the model fuses them into thicker clumps and thin ribbons. Anything thinner than the model can capture merges with whatever sits next to it, and loose hair is made almost entirely of such strands.

The second limit is exact repetition. The tower shown above is a stack of identical floors wrapped in a regular window grid, and the whole point of such a building is that every floor repeats exactly. Meshy 7 gets the overall shape, the rounded floor plan, the rooftop equipment, and the base right. The pattern itself drifts: the floor bands wobble, their spacing varies, and the windows slip out of line from floor to floor. Nothing is missing, but a building defined by exact repetition comes back only roughly right.

Availability

Meshy 7 becomes generally available on August 10, 2026 in the Meshy workspace. Meshy 7 Ultra Mode is available for single-view generation only, and multi-view support will be available soon.

Meshy 7 is open to all registered users, while only Pro tier and above are able to download models generated by Meshy 7. You can compare tiers on the pricing page.

The 3D geometry alignment benchmark introduced in this article will be made available separately after the Meshy 7 release, and we will introduce a dedicated texture alignment benchmark in future work.

Better Geometry Alignment, and a Way to Measure It

A generated 3D model can be clean, plausible, and technically valid while still disagreeing with the image that inspired it. As baseline generation quality improves, that gap becomes increasingly important.

Meshy 7 is our biggest step yet toward closing the geometry part of that gap.

At the same time, the 3D geometry alignment benchmark gives us a more rigorous way to measure where that gap remains: from broad proportions, to the placement of individual parts, to fine surface structure.

Geometry is only one part of image-to-3D alignment. Texture comes next. But for geometry, Meshy 7 gets the generated 3D model closer to the object the user actually asked for.

FAQs

What is Meshy 7?

Meshy 7 is Meshy's image-to-3D generative model, released on August 10, 2026 as the successor to Meshy 6. It converts a reference image into a 3D model and is built specifically around geometry alignment, meaning how closely the generated shape and structure match the input image.

What is 3D geometry alignment?

3D geometry alignment measures how faithfully the shape and structure of a generated 3D model match the image it was generated from. It is distinct from raw output quality: a model can produce clean, valid geometry that still describes a different object than the one in the reference image. Meshy measures it along three dimensions: overall proportion, spatial distribution, and surface details.

How is Meshy 7 different from Meshy 6?

Meshy 7 improves geometry alignment through three changes: an image encoder that reads features at multiple scales and accepts higher-resolution input, training data rebuilt so every sample corresponds precisely to its target geometry, and geometry alignment used as a direct evaluation signal throughout training. The largest measured gains over Meshy 6 are in surface details.

Can free users generate and download Meshy 7 models?

Meshy 7 is open to all registered users, so anyone with a Meshy account can generate with it. Downloading models generated by Meshy 7 requires the Pro tier or above. Current tier details are on the pricing page.

Does Meshy 7 support multi-view input?

Yes. Meshy 7 accepts four-view input in addition to single images, and the benchmark reports results for front view, top-quarter view, and four-view conditions. Meshy 7 Ultra Mode supports single-view generation at launch, with multi-view support coming soon.

When will the geometry alignment benchmark be released?

The 3D geometry alignment benchmark will be released separately after the Meshy 7 launch. A dedicated texture alignment benchmark is planned as future work.

Try Meshy 7 on Your Own Reference Image

Meshy 7 is available now to every registered user. Bring a concept image and see how closely the geometry comes back.

Original source

Meshy's Multiview Upgrade adds Multiview Texture and improves Multiview Geometry, generating from up to 4 views with native 2K/4K/8K base color texture.

TL;DR

Meshy's Multiview Upgrade is live for 100% of users. Multiview Geometry, an existing capability, has been enhanced, and Multiview Texture is brand new: texturing now supports multi-view generation for the first time. Both read a Main View plus up to three auxiliary views, Left, Back, and Right, so the model has more of the object to go on than a single photo ever provides. Multiview Texture generates native 2K, 4K, or 8K base color textures (no PBR yet), with stronger style fidelity on stylized and anime assets and more accurate text and patterns. Enable Multi-view in Image to 3D for geometry, then in the texturing flow for texture.

What Is Multi-View Image to 3D?

Most image-to-3D tools work from a single photo or render. That photo only captures whatever the camera happened to be pointed at, so anything the camera couldn't see, the back of a head, the underside of a shoe, the far side of a prop, simply isn't in the image. The model has no choice but to infer it.

Multi-view image to 3D changes the input, not just the output. Instead of one photo, Multiview Geometry and Multiview Texture read a Main View plus up to three auxiliary views, Left, Back, and Right, so more of the object's actual surface is visible before generation even starts. Any auxiliary view a creator doesn't have is generated automatically, so the workflow still runs on a single image; it just has more to work with the more views it's given.

What's New

Higher Multi-View Consistency, From Structure to Texture

Cross-view geometry and texture now align more accurately across the whole model. When the Left, Back, and Right views are read alongside the Main View, structure and surface detail stay consistent from one side of the model to the other, rather than only matching well on the face the single input image showed.

Native 2K/4K/8K Across All Views

Multiview Texture generates at native 2K, 4K, or 8K resolution, and that resolution is consistent across every view, not just the one matching the primary reference. A character's face, back, and profile all hold the same level of texture detail.

Stronger Style Fidelity on Stylized, Anime, and Hand-Drawn Assets

Multiview Texture preserves the art style of the input images more accurately on stylized, anime, and hand-drawn characters and props, where flat cel-shading, line art, and painterly rendering are easy for a model to drift away from between views.

Accurate Text and Patterns

Logos, labels, and repeating patterns render with more detail and a more natural look, because the model has the pattern's actual geometry and orientation from multiple angles to work from instead of extrapolating it.

Single Image vs. Multi-View: What Changes

• Hidden-side geometry: Single image only - Inferred; Multi-view - Read directly from the matching view
• Style across the model: Single image only - Set on one side, extrapolated to the rest; Multi-view - Anchored by an actual reference on each side
• Text and pattern orientation: Single image only - Extrapolated from one angle; Multi-view - Taken from the pattern's real geometry in each view
• Views you don't have: Single image only - N/A; Multi-view - Filled in automatically, still inferred (completion mode)

Multiview Texture in this release generates base color textures only. PBR (metallic, roughness, normal maps) is not included yet.

As a secondary improvement, de-lighting is cleaner across multi-view generations: less baked-in shadow, and richer, more balanced color in dark areas.

Best For

Stylized, Anime, Chibi, and Hand-Drawn Characters and Props

Flat color and clean linework have nowhere to hide inconsistency, whether the concept art came from an AI character generator or a sketchbook. On a realistic asset, shading and surface noise absorb a certain amount of drift; on a chibi character, a hair silhouette that changes shape between the front and the back of the head, or a jacket stripe that shifts position on the reverse, reads immediately. Multiview Texture removes the guess: the back and side faces come from references a creator supplies instead of from inference. Base color is also the layer that defines a toon or cel-shaded look, so this release's base-color-only output matches how stylized materials are already built, with lighting and outlines handled in the shader rather than baked in.

Creators Who Already Have Multiple Reference Images

A turnaround sheet, a concept sheet, or a folder of front, back, and side shots of the same subject is work that single-image generation throws away. Every view it doesn't read is a view a creator ends up chasing through repeated Retexture passes, swapping the reference image and re-running until the back and sides agree with the front. Multiview Geometry and Multiview Texture take the Main View plus up to three auxiliary views in one pass, so existing reference material does the work it was drawn to do.

Assets With Logos, Text, or Repeating Patterns

A chest logo from a 3D logo generator on the front and a different graphic on the back are two separate design problems, and one image only answers one of them. Left, Back, and Right exist to cover the other faces. Completion mode will fill in any view a creator skips, but for a face where the mark has to be exact, supplying the real reference gives Multiview Texture something to read instead of something to infer.

How to Use

Multiview Geometry and Multiview Texture are two separate stages.

Stage 1: Generate Geometry with Multiview

Geometry comes first, since Stage 2 needs a model to add texture onto.

1. Go to Image to 3D in the Meshy workspace.
2. Enable the Multi-view toggle.
3. Upload your Main View, plus any of Left, Back, and Right that you have. Missing auxiliary views are generated automatically.
4. Generate. The output is an untextured model.

Stage 2: Generate Texture with Multiview

Stage 2 isn't limited to models fresh out of Stage 1: any untextured model, including one imported from elsewhere, can go through Multiview Texture using the same Main View, Left, Back, and Right principle.

1. Take an untextured model, whether just generated in Stage 1 or imported from elsewhere, into the texturing flow.
2. Enable multi-view there.
3. Choose a resolution (2K, 4K, or 8K).
4. Generate.

Supported inputs for both stages: a single image, multiple images, or a pre-composited reference sheet (also known as a character turnaround sheet), all through the same model. Accepted formats are .png, .jpg, .jpeg, and .webp, up to 20MB per image.

Multi-view references carry over automatically: the views you used for geometry are pre-filled in the texture stage, so there is no need to upload them again.

Why It Matters

From a single image, a model has to guess at the sides and back it never sees. That guesswork is where problems show up:

• Structure warps on the hidden side, since there's no reference for what it actually looks like.
• Style drifts between the front and back of a stylized character, since nothing anchors the back to match the front.
• Patterns break at the seam where the model ran out of reference to extrapolate from.
• Text garbles on faces the model never had a clear view of.

The usual fix has been manual rework after the fact: fixing topology by hand, repainting texture seams, or redrawing a logo that came out wrong.

Why This Happens

A photograph is a record of light bouncing off whatever faced the camera at the moment it was taken. Everything outside that field of view, geometry, color, pattern, simply isn't in the data. A single-image 3D model doesn't have access to it either, so it fills the gap with a learned guess rather than an observation of the actual object in front of it. That guess tends to be least reliable exactly where it's hardest to check: the back, the underside, the far side.

Multiview Geometry and Multiview Texture remove that tradeoff at the source. When the model reads every view a creator actually provides, Left, Back, and Right included, it has less to guess and more to go on, so the result needs less cleanup after generation.

FAQs

What is Multiview Image to 3D?

Multiview Image to 3D is Meshy's Image to 3D workflow with the Multi-view toggle enabled. Instead of reading a single photo, it reads a Main View plus up to three auxiliary views, Left, Back, and Right, so the model has more of the object's actual geometry to work from instead of guessing at the sides it can't see.

How many reference images can I upload for a Multiview generation?

Up to four: one Main View plus Left, Back, and Right. You can also upload a single image, multiple images, or a pre-composited reference sheet, all through the same model. Any auxiliary view you don't provide is generated automatically.

Can I use a character reference sheet or turnaround sheet?

Yes. A pre-composited reference sheet, the kind often called a character turnaround sheet, is one of the accepted input forms for both the geometry and texture stages.

Does Multiview Texture support PBR textures?

Not in this release. Multiview Texture generates base color textures only; PBR maps (metallic, roughness, normal) aren't included yet.

What resolutions and file formats does Multiview Texture support?

Multiview Texture generates natively at 2K, 4K, or 8K, consistent across every view. Accepted input image formats are .png, .jpg, .jpeg, and .webp, up to 20MB per image.

Are Multiview Geometry and Multiview Texture generated in the same step?

No, they're two separate stages. Multiview Geometry runs first in Image to 3D and outputs an untextured model. Multiview Texture then runs afterward in the texturing flow.

Can I use Multiview Texture on a model I didn't generate in Stage 1?

Yes. Multiview Texture isn't limited to models fresh out of Stage 1: any untextured model, including one imported from elsewhere, can go through the texturing flow with multi-view enabled.

Why can single-image 3D generation warp or drift on the back of a model?

A single photo only captures light bouncing off the surfaces that faced the camera. Everything else, the back, the underside, the far side, isn't in the image, so the model has to infer it instead of reading it. Multi-view generation closes that gap by reading a Main View plus Left, Back, and Right directly.

Try the Multiview Upgrade on Meshy

Generate 3D models and textures from a Main View plus Left, Back, and Right references, with native 2K/4K/8K texture output.

Try Multiview Upgrade for Free

Original source

Meshy now generates a real 8K base color texture: sharper pore and fiber detail, cleaner de-lighting, and a photoreal look, without any upscaling.

TL;DR

Meshy 8K Texture is live now in the Texture panel, generating a real 8K base color map built on Multiview Diffusion, Meshy's new texturing algorithm. Pores, scratches, and fiber-level detail hold up under a close crop, de-lighting is cleaner, and the overall look reads as photoreal. A full 8K generation takes about 2 minutes and works across nearly every model type in Meshy. Available today for Game Dev, Film, and Large-Format 3D Printing workflows.

What 8K Texture Actually Changes

Zoom into the cheek of a character textured with Meshy's new 8K pipeline, and you can pick out individual pores. Crop into a rusted panel and the corrosion breaks up unevenly instead of smearing into a flat grime layer. That's what real 8K resolution gets you over 4K, once a camera actually gets close.

8K Texture is powered by Multiview Diffusion, a new texturing algorithm built specifically to hold up high-frequency surface detail (pores, scratches, individual fabric fibers) at a true 8K base color resolution, not a resolution slider turned up on the old pipeline. Alongside the resolution jump, it ships two other changes: cleaner de-lighting, and a material response tuned to read as photoreal rather than just sharp. All three are live now in the Texture panel, and a full 8K generation takes about 2 minutes.

What's New in 8K Texture

Real 8K Base Color, Not Upscaled

A lot of tools in this space advertise 8K textures. Most of them get there the same way: generate at a lower resolution, then run the result through an upscaler that interpolates new pixels between the ones that already exist. It looks sharper in a thumbnail. Zoom into a wrinkle, a weld seam, or a strand of hair, and the upscaler starts inventing texture that was never there: smeared micro-detail, soft edges where a pore or scratch should have a hard boundary, and a faint plastic sheen that gives away the trick.

Meshy's 8K Texture doesn't upscale. The base color map is generated natively at true 8K resolution by Multiview Diffusion, so fine surface detail (skin pores, leather grain, brushed-metal scratches, individual fabric fibers) stays distinct and legible instead of blurring together once you're in close on a hero asset. That's the difference between a texture that holds up when a camera gets close and one that only looks good until it does.

[Table comparing Native 8K and Upscaled ("fake") 8K]

A few tells give away an upscaled texture once you crop in: edges around pores or scratches that look softened rather than cut, a faint waxy or plastic sheen on skin and fabric, and detail that stops getting sharper no matter how far past a certain point you zoom, since an upscaler can only invent so much before it runs out of real information to work from.

Cleaner De-Lighting

Baked-in shadows and highlights from the source reference are one of the most common tells in AI-generated textures: a material that looks fine from a distance but has a phantom light source burned into the base color map. Multiview Diffusion de-lights more aggressively, stripping out those baked lighting artifacts so the base color stays neutral and relights correctly under any engine or renderer's own lighting.

A Look That Reads as Real, Not Just Sharp

Resolution and realism aren't the same thing. A sharpened texture can still look synthetic if the surface behaves uniformly, the way upscaled textures often do. Meshy's 8K output is tuned for photoreal material response: the irregularity in how leather creases, how rust spreads unevenly across metal, how fabric catches light differently at the weave level. That variation is what makes a surface read as a real material instead of a high-resolution pattern.

Seeing the Difference: 4K vs. 8K

At a distance, 4K and 8K can look close to identical. The gap shows up the moment you crop in. On the same model and the same camera distance, the 4K version starts to lose pore and fiber definition while the 8K version keeps texture legible.

Same story on hard-surface materials: scratches, edge wear, and rust bloom stay sharp and irregular in 8K instead of softening into a uniform grime layer.

Woven and layered materials show the same pattern from a different angle. Fabric weave, leather grain, and overlapping scales hold their individual threads and edges in 8K; at 4K those same details blur into a flat repeating pattern, which is often the first thing that gives an AI-generated texture away.

The gap comes down to pixel budget: 8K packs four times the pixel data of 4K into the same UV space, so fine, high-frequency detail that would get averaged away at 4K has enough room to actually exist.

Where This Matters: Game Dev, Film, and Large-Format 3D Printing

Game Development: Hero Assets That Survive a Close Camera

A first-person weapon, a cutscene close-up, or a character select screen all put a camera closer to a model than gameplay ever does. That's exactly where 4K textures run out of detail. 8K Texture gives hero and cinematic-facing assets enough resolution headroom to hold up under that scrutiny, without needing a separate high-res texture pass done by hand.

Film and Animation: Materials That Don't Need a Touch-Up Pass

Render-ready materials for film and previz work usually go through a manual detail and de-lighting pass before they're camera-ready. With cleaner de-lighting and real 8K base color detail out of the gate, that manual cleanup step shrinks, which matters most on tight production timelines where every asset touched by hand is time spent somewhere else.

Large-Format 3D Printing: Detail That Holds Up at Full Scale

A large-format print enlarges every surface flaw right along with the model itself: a texture that reads fine on a monitor can turn blurry or pixelated once it's scaled up and printed at size. 8K Texture gives large-format prints enough resolution headroom that surface detail (grain, wear, fine surface texture) stays legible at print scale, not just in a viewport.

How to Generate an 8K Texture in Meshy

1. Open the Meshy Workspace and generate or import a model.
2. Open the Texture panel and set the resolution to 8K (2K and 4K are also available).
3. Generate. A full 8K texture takes about 2 minutes.
4. Export your model to any major 3D tool.

8K Texture is available on paid plans; free-tier generations default to lower resolutions.

FAQs

What is 8K texture size?

An 8K texture is an 8192 × 8192-pixel image used in 3D design, game development, and film to capture extreme, photorealistic surface detail. Because a texture at that resolution takes up a large amount of VRAM, it's typically reserved for hero assets and close-up cameras rather than general background use, which is why Meshy still defaults new generations to 2K or 4K.

What is the difference between 4K and 8K texture?

4K is 4096 × 4096 pixels; 8K is 8192 × 8192, four times the total pixel count. That extra resolution matters most at close range: fine detail like pores, fabric weave, or scratch texture holds up under a close crop or zoomed-in camera at 8K, where a 4K texture starts to blur or pixelate.

Are 8K textures overkill?

For most assets, yes. Background props, distant environment pieces, and anything a camera won't get close to render just as well at 2K or 4K, at a fraction of the file size and VRAM cost. 8K earns its cost on hero assets: characters, weapons, or any surface the camera or player will scrutinize up close.

Which model types work with 8K Texture?

Nearly all of them: imported models, standard Meshy generations, topology mesh, remesh models, segmented multi-part models, and models that already have a texture applied.

Do I need a specific plan to use 8K Texture?

8K Texture is available to paid subscribers. Free-tier generations use lower default resolutions; upgrading unlocks the 8K option in the Texture panel.

Try 8K Texture on Meshy

Generate a real 8K base color texture with pore-level detail, cleaner de-lighting, and a photoreal material look.

Original source

Split 3D models into printable parts in 40 seconds with Meshy Auto Split

Watertight cuts, auto-arranged build plate, one-click export to your slicer.

TL;DR

Generating a 3D model was never the real challenge. Turning it into parts that can actually print (and fit back together) was. Auto Split handles that in one click. The full split previews in about 40 seconds, every cut is auto-capped watertight, and parts arrive on the build plate already arranged. No round-trip to Blender, no manual hole-capping, no re-alignment by hand.

Why You Need to Split a 3D Model Before Printing

Most models can't come off the printer in one piece, not because of a design flaw, but because real printers have real constraints. Splitting a model before printing is what makes those constraints manageable, and it unlocks a few things at once.

The model is bigger than the build plate

If your model exceeds the printer's Z height or footprint, it can't print in one piece. The standard fix is to cut it into plate-sized sections, print each separately, and join them. The hard part is deciding where to cut without landing the seam at a stress point or somewhere visible.

Multi-color FDM burns filament on every color swap

Multi-color FDM printers (Bambu Lab, Prusa MMU, and similar) switch filaments by purging the previous color before loading the next. Every swap wastes material. The community has a name for the blocky blob that accumulates: the "poop tower." On a complex figure with many color zones, that tower can outweigh the model itself.

Splitting the model by color (so each part prints in a single filament) cuts most of those swaps. Fewer swaps means less waste and a shorter print.

Hand-painted figures need parts that come apart

Miniature and figure painters know the frustration of trying to reach inside a fully assembled model with a brush. Joints, recessed folds, the inside of a cape: they're unreachable once the model is together. The professional answer is to split first, paint each part separately, then assemble. The problem has always been doing that split without breaking the geometry.

How People Split Models Today (and What Usually Goes Wrong)

What the manual workflow looks like

The current standard for splitting an STL file for printing is a manual workflow in Blender, Meshmixer, or a slicer's built-in cutting tools. The steps look roughly like this:

1. Pick a cut plane by eye
2. Execute the boolean cut
3. Add alignment pins or dovetails so pieces fit back together
4. Check for non-manifold edges created by the cut
5. Repair the mesh before it will slice cleanly

It works, but it's slow and error-prone.

Four failure points that keep coming up

Common failure | What goes wrong | How Auto Split handles it

Cut placed on thin wall or joint | Part snaps under stress after printing | Cuts follow the model's natural semantic boundaries, not arbitrary planes

No alignment feature added | Parts gap or shift at the seam | Every contact surface is auto-capped and closes flush

Boolean cut creates non-manifold edges | Model errors out in slicer, can't print | Watertight caps are generated automatically on every cut

Seam lands on a face or visible curve | Seam shows after sanding and painting | Boundaries follow structural seams, not surface contours

Other 3D generation tools have introduced splitting features: Rodin, Tripo, and others. These tools are generally positioned around creative editing, decomposing a model into segments like head and torso for rigging or retexturing. Tripo's own tutorial content notes that their segmentation feature may not produce geometry ready to slice directly; additional surface repair in Blender is often still needed. Auto Split is built for a different goal: parts that go straight from the split into the slicer without any detour.

How Auto Split Works

Fast Preview: See the Full Split in About 40 Seconds

Boolean operations in Blender can run from seconds to hours depending on mesh complexity. Auto Split returns a full, exportable preview of the split in roughly 40 seconds. That preview is the real result (same geometry you'll print), so you can evaluate it, decide if you want a different split, and re-run without having committed to anything.

Smart Cuts: Boundaries That Follow the Model

Auto Split reads the model's geometry to find natural split lines, like where a character's neck meets the torso or where a wing root meets the body. Cuts land at those structural boundaries rather than across arbitrary planes. Parts are also sized to fit a target build plate, so you don't have to check dimensions afterward.

Built to Assemble: Watertight, Arranged, Slicer-Ready

Every cut surface is automatically capped into a closed, watertight mesh. Parts are laid flat and arranged on a virtual build plate. One click sends the result directly to Bambu Studio, OrcaSlicer, or ElegooSlicer. You're done before you would have finished roughing out the first cut manually.

From Generation to Print: The Full Workflow

Auto Split connects the end of the Meshy generation workflow to your slicer without any extra steps in between:

Generate → Auto Split → Review → Export → Slice → Print

1. Generate a model in Meshy 6 (untextured / draft)
2. Open Auto Split and click to split. Preview arrives in ~40 seconds.
3. Re-split with different settings if the first result isn't what you wanted
4. Export parts directly to your slicer
5. Print each part, then assemble

What assembly actually looks like

In practice, the cuts are clean enough that assembly is straightforward. Auto-capped surfaces sit flush; parts fit together without filling gaps or filing down ridges.

FAQs

Which models does Auto Split currently support?

Auto Split currently works with untextured (draft) models generated by Meshy 6. Textured models and models imported from outside Meshy are not supported in this release. Support for additional model types is on the roadmap.

Do the split parts need any post-processing before slicing?

No. Every cut surface is automatically capped to produce a watertight, manifold mesh. You can send the exported parts directly to Bambu Studio, OrcaSlicer, or ElegooSlicer without repairing geometry in Blender or Meshmixer first.

Which slicer software does Auto Split export to?

Auto Split supports one-click export to Bambu Studio, OrcaSlicer, and ElegooSlicer. You can also download parts as individual STL or 3MF files and open them in any slicer you prefer.

How does Auto Split compare to splitting manually in Blender?

The manual process (cutting, adding alignment pins, checking for non-manifold edges, repairing the mesh) typically takes 30 minutes to a few hours. Auto Split returns a complete result in about 40 seconds, and watertight caps are added automatically, so there's no repair step.

Can I re-split if I don't like the first result?

Yes. Auto Split is non-destructive. If the first split isn't what you wanted, adjust the settings and re-run. Each attempt takes about 40 seconds, so iterating is fast.

What's Next for Auto Split

Auto Split is the first step toward a fully automated path from idea to printed object.

Conversational splitting is coming soon. Instead of accepting what the algorithm decides, you'll be able to describe the split in plain language. Tell Meshy things like "separate the wings from the body" or "cut horizontally at the waist," and it will reshape the split to match. An AI-native way to get exactly the parts you had in mind.

Try Auto Split on Meshy

Generate a model and split it into printable, assemblable parts in one click. Preview in 40 seconds, export directly to your slicer.

Original source

TL;DR: Meshy 3D Agent is the world's first conversational AI agent for 3D creation, an AI 3D model generator you talk to. From a text description, a photo, or a sketch, it brainstorms directions, generates concepts in batches, and turns your favorite into a downloadable 3D model, ready for 3D printing or your game pipeline. No Blender or Maya required: easier to start, smarter to use, fuller in one chat. Built for 3D printing makers, indie game developers, 3D artists, and designers. Try it at meshy.ai.

AI Agents. Now for 3D.

No learning curve. Just chat, and your idea takes shape in 3D.

For years, 3D has lived behind a wall of software. Blender, Maya, ZBrush, and hours of tutorials before you can place a single cube. Meanwhile, AI agents started writing code, drafting decks, and editing video. We kept asking the same question: why not 3D?

Today, that wall comes down. Meet Meshy 3D Agent, the AI agent paradigm, now for 3D.

Meshy 3D Agent at a glance

• Inputs: text description, reference photo, sketch, or a rough idea
• Outputs: downloadable 3D models in FBX, OBJ, GLB, USDZ, STL, BLEND, 3MF, DXF (GLB by default), ready for your slicer, Blender, or game engine
• Best for: 3D printing, indie game assets, prototyping, concept art
• What's in the chat: brainstorming, batch concept generation, 3D conversion, built-in 3D Q&A
• Availability: Public Beta

What is Meshy 3D Agent?

Meshy 3D Agent is the world's first conversational AI agent built for 3D creation. It's an AI 3D model generator you talk to: you start with a thought, a photo, a sketch, a text description, or a vibe, and the agent takes it from there.

• Brainstorms with you: pitches creative directions before generating anything, so you start with momentum instead of a blank page.
• Generates visuals in batches: produces multiple concepts in one go, then refines them through chat until the look is right.
• Turns it all into 3D: converts your favorite concept into a model that's yours to download, 3D print, or take into your own workflow.
• Answers your 3D questions: a built-in knowledge base, right in the chat, so you never have to step away to learn how.

You can make game props and characters, 3D-printable figurines and toys, product and concept prototypes, and art assets, anything you can describe or sketch. As an indie game asset generator, it also handles style-consistent 3D assets: tell the agent the look you're after and it generates batch 3D assets that hold together. As of its Beta launch, no other product offers a fully conversational, multi-step agent workflow purpose-built for 3D creation — making Meshy 3D Agent the first of its kind in the AI 3D generation space.

What makes Meshy 3D Agent different: Easier, Smarter, Fuller

What sets 3D Agent apart isn't a single feature. It's that the Agent thinks ahead: pitching next steps, answering the questions you'd otherwise have to go look up, suggesting the right settings, so you don't have to. The result is 3D creation that's easier to start, smarter to use, and bigger in what one chat can deliver.

Easier to start

Traditional 3D asks you to learn the tool before you can make anything: software to install, an interface to master, tutorials to sit through. 3D Agent removes that entry cost. You describe what you want the way you'd explain it to a friend, and the agent meets you there. The skill barrier that keeps most people out of 3D is no longer the price of entry.

Smarter to use

A single-shot text-to-3D generator waits for a perfect prompt and returns exactly what you typed. 3D Agent works differently: it pitches a few creative directions before generating, suggests the right settings instead of expecting you to know them, and carries the context of your whole conversation so each step builds on the last. You bring the idea, and the agent handles the craft like an AI 3D art director at your side.

Fuller in one chat

In most pipelines, brainstorming, generating, converting to 3D, and looking up answers each live in a separate tool. 3D Agent folds them into one conversation, end to end: ideate, generate visuals in batches, turn a favorite into a model, and export it in any of 8 standard formats for 3D printing, game engines, or your own workflow, without switching windows.

How does Meshy 3D Agent work?

In one chat, Meshy 3D Agent takes you from idea to model in a few steps:

1. Describe or show your idea. Type what you want in plain language (text-to-3D), or drop in a reference photo or sketch (image-to-3D). A rough idea is enough to start.
2. Get creative directions. Before generating, the agent pitches a few directions, so you choose between options instead of facing a blank page.
3. Generate concepts in batches. It produces multiple visual concepts at once, and you refine them through chat until the look is right.
4. Convert to 3D. Pick your favorite and the agent turns it into a model: GLB by default, exportable to FBX, OBJ, STL, and more.
5. Download and use. Take it into your slicer to 3D print, or into Blender, Maya, or your game engine.

Ask 3D questions at any step and the agent answers in the same chat, carrying your full context so each step builds on the last. This is conversational 3D modeling: chat to 3D, idea in, model out.

Who is Meshy 3D Agent for?

Picture a father at the kitchen table watching his daughter sketch a monster for her birthday cake. He has a 3D printer but not ten hours of tutorials to spare. He describes the monster to the agent, and a few exchanges later, it's sitting on his desk. This is 3D printing AI in one conversation, no tutorials required.

Or picture an indie developer racing toward Friday with twelve cyberpunk props to build. Single-shot generators keep giving her twelve different art directions. She tells the agent the look she's after, and the whole alley takes shape in one conversation, style intact.

Meshy 3D Agent is built for 3D printing makers, indie game developers, 3D artists, designers, and anyone with an idea waiting to take shape in 3D.

What's Coming Next

Beta is just the beginning. In the releases ahead, you'll ask the agent to bring your characters to life, with rigging and animation built right into the chat. Your work will flow into more of the DCC and slicer tools you already use, all without leaving the conversation. And you'll find references and inspiration in the chat itself, never needing to step away. All to keep you in the flow, where creating in 3D feels like one continuous conversation.

FAQ

How is this different from a regular text-to-3D tool?

A text-to-3D tool turns one prompt into one output. Meshy 3D Agent is a conversation. It brainstorms directions, generates concepts in batches, refines through chat, and answers your questions along the way, taking you from a single line to a finished model.

Do I need to know Blender or Maya to use Meshy 3D Agent?

No. There's no learning curve. You describe what you want in plain language, and the agent handles the rest.

Does Meshy 3D Agent work with Blender? Is it a plugin?

Meshy 3D Agent is a standalone conversational tool, not a Blender plugin or software you install. You build your model in chat, then export it as BLEND (or FBX, OBJ, and more) to open directly in Blender, Maya, or your game engine. There's nothing to download to get started.

What inputs does Meshy 3D Agent accept?

Text descriptions, reference photos, sketches, or just a rough idea. It's conversational 3D modeling, chat to 3D, in one place.

Can I download or 3D print the results?

Yes. Finished models are yours to download, 3D print, or bring into your own workflow.

What file formats can I export?

Models export in FBX, OBJ, GLB, USDZ, STL, BLEND, 3MF, and DXF (GLB by default), so they drop straight into your slicer (STL/3MF for 3D printing), Blender (BLEND), AR (USDZ), or game engines like Unity and Unreal (FBX/OBJ).

Is Meshy 3D Agent free to use?

You can create a free Meshy account and start exploring at no cost. Meshy 3D Agent is currently in public Beta for Meshy Pro members and above, and a free trial is on the way, so more creators can try the agent in chat.

Who is it best suited for?

3D printing makers, indie game developers, 3D artists, and designers. Anyone who wants to create in 3D without the traditional software barrier.

What is the best AI tool for 3D modeling?

For conversational, end-to-end 3D creation, Meshy 3D Agent is the first AI agent purpose-built for the job. Unlike single-shot generators that require a perfect prompt, the agent brainstorms with you, generates in batches, and converts to 3D, all in one chat. It's designed for anyone who wants to go from idea to downloadable model without learning traditional 3D software.

Can AI create 3D models from text?

Yes. Meshy 3D Agent turns a plain-language text description into a downloadable 3D model. Describe what you want, and the agent generates visual concepts, refines them through conversation, and outputs a model in FBX, OBJ, GLB, STL, or other standard formats, ready for 3D printing, game engines, or your own pipeline.

Original source

Meshy Workspace 3.0 rebuilds AI 3D creation with a new Discover creation bar, task-oriented modules for Image, 3D Model, 3D Printing, and Animate, and unified asset management.

TL;DR Meshy Workspace 3.0 is a completely rebuilt AI 3D creation environment designed to reduce friction between idea and finished 3D asset. It introduces three core upgrades: an integrated creation bar on the Discover page, task-oriented workflow modules for Image, 3D Model, 3D Printing, and Animate, and a unified Assets tab for managing all generated content in one place.

From Legacy Workspace to Workspace 3.0: What Changed

The legacy Meshy workspace was organized around individual tools — users navigated to separate pages for Image to 3D, Text to 3D, AI Texturing, and Text to Image. Remesh and Texture were standalone steps in the sidebar. My Assets was a separate top-level navigation item. Every task required switching between disconnected pages and tools. Workspace 3.0 replaces this tool-oriented, fragmented experience with a unified, task-oriented workspace.

What's New in Meshy Workspace 3.0

Feature - What It Does

• Integrated Creation Bar - Generate 3D models or images directly from the Discover page with prompt presets
• Image Module - AI image generation with Stylization, Aspect Ratio, Image References, Multi-view, and Pose controls
• 3D Model Module - End-to-end model workflow — generate, texture, remesh, rig, and animate without leaving the page
• 3D Printing Module - Print preparation with Multi-Color Print, Add Base, and direct export to 6 slicer applications
• Animate Module - Dedicated space for animating generated 3D assets
• Scene Module (New) - New module for 3D scene creation and composition
• Video Module (New) - New module for video generation from 3D assets
• Unified Assets - My Assets and My Projects accessible directly inside Workspace
• Prompt Presets - Curated starting prompts for strong first results, even on your first try
• Projects (Coming Soon) - Organize assets into folders by game, print collection, or any project

Instant AI 3D Generation — Start Creating the Moment You Arrive

Meshy Workspace 3.0 puts the creation bar front and center on the Discover page, so users can start generating without any extra navigation. Your new starting line is front and center — choose your mode, type a prompt, and go.

Key capabilities:

• Switch between 3D Model and Image generation in one click
• Type a custom prompt or select from curated preset prompts
• Preset prompts are designed to deliver strong first results — even on your very first try
• Not sure how to write a great prompt? Check out our guide: How to Write Perfect Text Prompts for Meshy's 3D Model Generation
• Zero extra navigation steps required to begin generating

Task-Oriented 3D Workflows — Dedicated Spaces for Every Creative Mode

The legacy workspace required users to jump between separate tools — Text to 3D, Image to 3D, AI Texturing, and Remesh were all disconnected steps. Workspace 3.0 reorganizes the entire creation process into dedicated task-oriented modules, each containing everything you need for that workflow in one place.

The modules in Workspace 3.0:

🖼️ Image — A complete AI image generation suite. Want to go straight from image to 3D? Learn how in our guide: Instantly Transform Text into AI Images for 3D Creation

• Choose your AI model (e.g. Nano Banana Pro)
• Set Aspect Ratio (1:1 / 16:9 / 9:16 / 4:3 / 3:4)
• Select Number of images (1–4)
• Apply Stylization for different visual styles
• Enable Multi-view for consistent multi-angle outputs — see how it works: How to Convert Images to 3D Model with AI (Multi-View Support)
• Set Pose (None / A-Pose / T-Pose)
• Upload up to 5 Image References for guided generation

🧊 3D Model — The complete end-to-end model creation workflow, without leaving the page. New to Image to 3D? Start here: How to Convert 2D Images to 3D Models Using AI

• Generate from text prompt or reference image
• Choose Model Type: Standard or Low Poly (Beta)
• Select AI Model version (Meshy 6)
• Toggle Image Enhancement, Multi-view (Beta), and Pose (A-Pose / T-Pose / Custom)
• Apply Texture directly from the bottom toolbar — no separate page needed
• Remesh, Rig, Download, and send to Animate from the same toolbar
• View Topology (faces, vertices) and Printability check inline

🖨️ 3D Printing (New) — A complete print preparation workflow, from model to printer. For a deeper dive, read: How to Create 3D Models for 3D Printing Using Meshy

• Post-Processing:
  • Multi-Color Print — Preview and prepare your model for multi-color printing
  • Add Base — Add a base to your model, with multiple base shape options
• Send to Printer — Direct one-click export to 6 major slicer applications: Bambu Studio, OrcaSlicer, Creality Print, Elegoo Slicer, LycheeSlicer, Ultimaker Cura
• Want to print from a photo? See our guide: How to Make a 3D Print from A Photo
• Working with Blender? Check out: Blender for 3D Printing with Meshy

🎬 Animate — Animation tools for bringing your generated 3D assets to life. Learn more: AI-Powered 3D Modeling and Animation

🌐 Scene (New) — Create and compose 3D scenes

📹 Video (New) — Generate videos from your 3D assets

Who it's built for:

• Game developers assembling 3D asset batches for their projects
• 3D printing hobbyists prepping and optimizing models for print
• Digital artists exploring AI-assisted generative workflows

Unified 3D Asset Management — My Assets and Projects, All in One Place

In the legacy workspace, My Assets was a separate top-level navigation item, disconnected from the creation tools. In Workspace 3.0, both My Assets and My Projects are accessible directly inside the Workspace dropdown — always one click away from wherever you're creating.

What's changed:

• My Assets and My Projects live inside Workspace — no more separate top-level navigation
• Images and Models consolidated under one unified Assets view
• Search and filter across your entire asset library without leaving your workflow
• Assets are always one click away from your active module

Coming soon — Projects:

• Organize your assets into folders by game, print collection, client, or any project you're working on
• A cleaner way to manage growing libraries as your output scales

Frequently Asked Questions

What is Meshy Workspace 3.0? Meshy Workspace 3.0 is a rebuilt AI 3D creation workspace that reorganizes all creation tools into task-oriented modules: Image, 3D Model, 3D Printing, Animate, Scene, and Video. It integrates My Assets and My Projects directly into the Workspace, and adds a creation bar on the Discover page for instant generation.

What's the difference between the legacy Meshy workspace and Workspace 3.0? The legacy workspace was tool-oriented — users navigated between separate pages for Text to 3D, Image to 3D, AI Texturing, Remesh, and Animation, with My Assets as a separate top-level page. Workspace 3.0 replaces this with task-oriented modules where every step of a workflow lives in one place. AI Texturing and Remesh are now integrated into the 3D Model module, and three new modules have been added: 3D Printing, Scene, and Video.

How do I start generating 3D assets in Workspace 3.0? Open Meshy and go to the Discover page. Use the integrated creation bar to select 3D Model or Image generation, enter a prompt or choose a preset, and click generate. For tips on writing better prompts, see: How to Write Perfect Text Prompts for Meshy's 3D Model Generation

What are the workflow modules in Workspace 3.0? Workspace 3.0 includes six modules: Image (AI image generation with stylization and references), 3D Model (end-to-end model creation with integrated texturing and rigging), 3D Printing (print preparation with post-processing and slicer integrations), Animate (3D animation), Scene (scene creation), and Video (video generation).

Is AI Texturing still available in Workspace 3.0? Yes. AI Texturing is no longer a standalone page — it is fully integrated into the 3D Model workflow. After generating a model, apply textures directly from the bottom toolbar without navigating away.

Is Remesh still available in Workspace 3.0? Yes. Remesh is no longer a standalone sidebar module — it is integrated directly into the 3D Model workflow toolbar, alongside Texture, Rig, Download, and Animate.

What are prompt presets in Meshy? Prompt presets are curated example prompts built into the creation bar on the Discover page. They help users — especially first-timers — get high-quality 3D generation results without crafting a prompt from scratch. Want to go deeper? Read: How to Write Perfect Text Prompts for Meshy's 3D Model Generation

What is the 3D Printing module in Meshy Workspace 3.0? The 3D Printing module is a brand-new addition in Workspace 3.0. It provides a complete print preparation workflow: post-processing tools including Multi-Color Print and Add Base, followed by direct one-click export to six major slicer applications — Bambu Studio, OrcaSlicer, Creality Print, Elegoo Slicer, LycheeSlicer, and Ultimaker Cura. Learn more: How to Create 3D Models for 3D Printing Using Meshy

What is Multi-Color Print in Meshy? Multi-Color Print is a post-processing feature inside the 3D Printing module. It allows users to preview and prepare their 3D model for multi-color printing before exporting to a slicer.

What is the Add Base feature in Meshy? Add Base is a post-processing feature inside the 3D Printing module. It lets users add a base to their 3D model before printing, with multiple base shape options to choose from.

What slicer software does Meshy Workspace 3.0 support? The 3D Printing module supports direct export to six slicer applications: Bambu Studio, OrcaSlicer, Creality Print, Elegoo Slicer, LycheeSlicer, and Ultimaker Cura.

What are the new Scene and Video modules? Scene and Video are new modules introduced in Workspace 3.0. Scene enables 3D scene creation and composition, while Video supports video generation from your 3D assets.

What is the unified Assets tab in Meshy Workspace 3.0? My Assets and My Projects are now accessible directly inside the Workspace dropdown, replacing the separate top-level My Assets navigation. All generated Images and Models are managed in one consolidated view.

What is the Projects feature and when is it coming? Projects is an upcoming feature that will let users organize assets into custom folders — by game, print collection, client, or any category. It is currently listed as coming soon in Workspace 3.0.

Is Workspace 3.0 available to all Meshy users? Yes. Workspace 3.0 is available to all Meshy users starting March 17, 2026.

This Is Just the Beginning

Workspace 3.0 is the foundation for everything we're building next — faster creation, smarter organization, and workflows designed around how 3D artists actually work.

Ready to see it for yourself? Explore Workspace 3.0 →

Original source

Meshy-6 introduces cleaner geometry, sharper hard-surface details, and powerful new workflows featuring Low Poly Mode, multi-color 3D printing, and upgraded APIs to help you create production-ready 3D assets faster and with greater control.

We are excited to introduce Meshy-6, our most refined and versatile generation model to date. Building on the evolution from Meshy 5, this release focuses on higher-quality geometry, better structural consistency, and powerful new workflow features. From more anatomically accurate characters to cleaner hard-surface models, and from low-poly optimization to multi-color 3D printing, Meshy-6 is built to give you more control at every stage of creation.

Key Upgrades: Beyond the Limits

Meshy-6 delivers meaningful improvements at the core of 3D generation quality, ensuring models are more usable right out of the box.

Refined Geometry

Meshy-6 generates smoother, more anatomically correct geometry for characters and organic models, drastically reducing the time spent on manual cleanup and adjustment.

Enhanced Hard Surface

For mechanical and geometric models, Meshy-6 offers sharper edges, clearer silhouettes, and a cleaner overall structure.

Next-Level Features: Boost Your 3D Workflow

Meshy-6 also introduces a set of new features designed to support real-world use cases across games, 3D printing, and developer integrations.

Low Poly Mode

A dedicated Low Poly Mode built specifically for game developers. This mode produces efficient wireframes optimized for real-time performance, making assets easier to integrate into game engines without extensive manual retopology.

Multi-Color 3D Printing

The new multi-color printing option automatically simplifies complex textures into clean color blocks optimized for FDM hardware. It exports slicer-friendly formats like 3MF, ensuring a seamless transition from digital model to physical print.

API Upgrades

We've made several important upgrades to the Meshy API to give developers more flexibility and a better testing experience:

• API Playground: A new interactive environment on the Meshy website where developers can explore API parameters and test requests directly.
• Text to Image API & Image to Image API: New AI-powered image generation and editing APIs that support both nano-banana and nano-banana-pro models. Optional multi-view generation is available, enabling more consistent and production-ready visual outputs.

Create with Confidence in Meshy-6

"When Meshy 6 dropped, everything changed. It literally became 'download, slice, and print.' What used to take days of cleanup now takes minutes — the workflow improvement is night and day."

Chad Hunter
3d printing artist

With improved geometry, smarter structure, and workflow-focused features, Meshy-6 represents a major step forward in usability and control. Whether you're building real-time game assets, preparing models for 3D printing, or developing with our APIs, Meshy-6 helps you move faster with fewer compromises.

Welcome to the next generation of Meshy.

Ready to Experience the Power of Meshy 6?

Meshy 6 powers your workflow with precision and speed. Bring your ideas to life effortlessly!

Try Meshy 6 for Free

Original source

Meshy has become the first-ever 3D AI tool to officially sponsor the Blender Foundation. With a custom-built plugin, Meshy enables seamless one-click model imports into Blender.

Blender x Meshy: Shaping the Future of 3D Creation

Blender has long been the go-to platform for 3D creators worldwide. Now, Meshy marks a major milestone as the first and only AI-powered 3D tool to officially sponsor Blender. Through this collaboration, Meshy introduces a dedicated plugin that enables one-click model import into Blender—eliminating traditional download steps and streamlining the workflow.

Exclusive Collaboration: Meshy Becomes Official Blender Foundation Sponsor, Revolutionizing 3D AI Integration

Meshy has become the first officially certified 3D AI sponsor of the Blender Foundation, marking a milestone in AI technology's integration with 3D creation. This partnership introduces a custom-designed Blender plugin that streamlines the 3D workflow, bringing AI-powered tools directly into the creative process.

Traditionally, using external models in Blender was time-consuming—searching for assets, downloading files, and manually importing them. The Meshy plugin eliminates these steps, allowing creators to instantly access and import AI-generated models from Meshy's library without delays. This integration enhances efficiency and makes AI creation accessible to 3D artists, designers, and developers.

Meshy's AI technology can generate a wide range of 3D assets, including buildings, characters, and props, while Blender offers powerful tools for professional-level editing. Together, they create a seamless workflow of "AI-powered rapid generation + human-led refinement," empowering creators to overcome bottlenecks and unlock limitless possibilities.

Hands-On Guide: How to Use the Meshy Plugin in 6 Simple Steps

Step 1: Log in to your Meshy account and download the plugin

First, visit Meshy AI and log in to your account.

After logging in, navigate to the Assets section and click on Download Blender Plugin. This will download a .zip file containing the official Meshy Blender plugin.

Tip: Do not unzip the file. Blender installs the plugin directly from the .zip file.

Step 2: Install the Plugin in Blender

1. Open Blender (version 3.0 or higher is recommended).
2. Drag and drop the downloaded Meshy plugin .zip file directly into the Blender window.

Alternatively, you can go to the top menu: Edit → Preferences → Add-ons, then click Install and select the .zip file you just downloaded.

1. Enable the plugin by checking the box next to Meshy Plugin.
2. You should now see the Meshy panel in the right-hand toolbar (press N if it's hidden).

Step 3: Get Your Meshy API Key

To access Meshy's cloud assets in Blender, you'll need to link your account:

1. Go back to Meshy AI.
2. Click on API in the top navigation bar.
3. Click Generate API Key, then copy the key provided.

Step 4: Authenticate the Plugin in Blender

Go back to Blender and:

1. Open the Meshy panel from the right-hand side.
2. Paste your API key into the input field.
3. Click Login.

You're now connected to your Meshy account within Blender. You can browse community assets or import your own.

Step 5: Search and Import 3D Models from the Meshy Community

In the Meshy plugin:

1. Use the Search bar to browse Meshy's public model library.
2. Enter a keyword such as "cyberpunk," "robot," or "fantasy castle."
3. Preview the models directly within the plugin interface.
4. Click Import Model to bring the desired model into your scene. It will be automatically downloaded and placed in your Blender viewport.

Step 6: Import Your AI-Generated Models from the Meshy Plugin

If you've already generated 3D models using Meshy, you can import them directly into Blender—no need to return to the website.

1. In the Meshy plugin, navigate to the My Assets section.
2. If your models don't appear right away, click Refresh Assets.
3. Browse your generated models (you can preview them via thumbnails).
4. When you find the model you want, simply click Import Model.

The process is seamless—your model will instantly appear in the Blender scene without any manual downloading or file management.

Conclusion

The collaboration between Meshy and Blender marks a deeper integration of AI technology into 3D creation. Through this partnership, creators can not only integrate AI tools more efficiently but also break through the limitations of traditional workflows.

The plugin provided by Meshy simplifies and accelerates asset imports, ensuring a seamless creative process. As this collaboration progresses, Meshy and Blender will continue to open up new possibilities for global creators, driving 3D creation towards a more intelligent and efficient future.

Original source

The wait is finally over! We are excited to introduce our newly launched workspace UI, designed to offer a more streamlined and flexible workflow experience. In this post, we will guide you through some of the exciting features in our revamped workspace UI.

Meshy creators, we've got something big for you! Our new workspace is here to elevate your 3D AI creation experience—making it smoother, faster, and more flexible. Ready to dive in? Let's explore the key features and steps to get started!

Model Generation

Transform your ideas into 3D models using the following methods:

• Text to 3D: Enter a detailed description of your model. Be specific—describe objects, shapes, styles, and unique characteristics to guide our AI.
• Image to 3D: Upload a clear, front-view image with a plain background for the best results. In just 60 seconds, Meshy generates four unique interpretations of your input. Pick your favorite and bring it to the next stage.

Texture Generation

Elevate your creation to new heights with texture generation:

• After generating a detailed mesh, use a text prompt to generate textures for the mesh.
• Stuck on ideas? Try the Prompt Helper for inspiration and get stunning results faster.

Remesh Your Models

Optimize your mesh to better suit a specific use case:

• Target Polycount: Adjust the polycount of the mesh to achieve an optimal balance between detail and performance. For intricate details, such as those required for 3d printing, use a higher polygon count. For performance optimization in game engines, opt for a lower polygon count.
• Topology: Select either Triangle or Quad topology based on the specific requirements of your project. Use Triangle topology for straightforward applications like 3D printing and opt for Quad topology if you anticipate making further edits in other 3D modeling software like Blender.

Stylize Your Models

Elevate your models with artistic styles:

• Sculpture Style: Generate high-poly models with integrated baked PBR textures, including displacement and ambient occlusion maps. Ideal for projects demanding photogrammetry-level quality and intricate detail.
• PBR Style: Create Physically-Based Rendering (PBR) maps to enhance realism and add intricate surface details to your assets.

Animate Your Models

Bring your creations to life with animations:

• Open the Animation Panel, choose a character type, and place rigging points.
• Preview animation presets, then download your fully rigged and animated model—ready to integrate into your 3D environment or refine further.

Tips for New Workspace

• Try a Non-Linear Workflow: The new UI gives you complete freedom to generate, texture, and optimize in any order. Customize your workflow to suit your project needs, such as trying a model → remesh → texture approach for maximum flexibility.
• View Your Assets: If you generate multiple assets and need to manage them quickly, click the Related Assets button. All related assets will appear in the left panel, streamlining your workflow.

Special Reward: Unlock Discounts & Badges!

From now to January 1, 2025, enjoy:

• 50% off generation tasks.
• An exclusive badge for trying all features in the right panel.

With Meshy's redesigned workspace, your creative journey has never been smoother, faster, or more powerful. Start creating today, and welcome to the future of 3d art with Meshy!

Original source

Introducing Meshy-4, a groundbreaking update in 3D generative AI, enhancing geometry quality and user workflow.

The art of structure has shaped human creativity for millennia, from the pyramids of Egypt, to the delicate Renaissance sculptures, and now soaring skyscrapers of today. However, building such structures in the virtual world presents unique challenges. While iconic buildings like St. Peter's Basilica took over a century to complete, merely five years is considered long for developing a game today. Expectations are just as high: travellers in the virtual realm are demanding environments with both stunning visuals and engaging narratives.

These challenges in virtual world-building call for revolutionary technologies, where creating highly detailed geometry plays a crucial role. At Meshy, we believe that generative AI may serve as a viable path towards this goal. Previous versions of Meshy were capable of rapid 3D model generation in minutes, but despite the results were visually appealing from a distance, they lacked the detail and accuracy needed for professional projects.

Introducing Meshy-4 —today, we break ground towards truly elevating creative world-building through generative technologies.

Remastered Geometry

With Meshy-4, we have pushed the boundaries of what's possible in 3D generative AI. Whether you're using Text to 3D or Image to 3D, you will now experience a dramatic increase in geometry quality of the generated models.

With a greatly improved generation algorithm, Meshy-4 produces models with extremely clean hard surfaces, free from the bumps and dents plaguing generation results of previous versions.

Additionally, geometry details have been substantially enhanced, allowing for the creation of highly intricate models, capturing subtleties previously beyond the reach of AI-generated models.

We've also updated the Discover page to pair with Meshy-4's model quality enhancements. Now, when you hover over a model generated by Meshy-4, you can instantly reveal its untextured version. This allows you to closely examine the underlying geometry, showcasing the clean hard surfaces and intricate details that Meshy-4's improved algorithms deliver.

New Text to 3D Workflow

The largest update in Meshy-4's user interface is a new Text to 3D workflow. In previous versions, the Text to 3D process was divided into a Coarse stage and a Refine stage. While this approach aimed to incrementally improve model quality, it often left users feeling uncertain about the final outcome as we have received numerous user feedbacks on the refined model deviating from the coarse result.

Therefore, we now split the Text to 3D process in Meshy-4 in a more distinct and focused way: modeling and texturing.

• Modeling Stage: In the first stage, 4 untextured models are generated from the text prompt. The modeling stage generates clean, highly detailed meshes that provides a robust foundation for the final asset.

• Texturing Stage: After the modeling stage is completed, you can select a generated mesh and move on to generate textures. The textures are directly generated in the quality of the Refine stage of previous versions, resulting in a visually stunning and coherent final product.

Our new Text to 3D workflow also paves the way for exciting new features specifically focused on modeling or texturing. This separation allows for introducing even greater control over the final result, such as editing the mesh before generating the texture or customizing multiple color variants for the same model. Stay tuned for Meshy's future releases on these new capabilities!

Try Again?

People say that now generative AI is like a gacha game—you never quite know what you're going to get with each pull. Sometimes results can be perfect, and other times, they might miss the mark. That's why we've introduced a new Retry feature in Meshy-4's Image to 3D tool, designed to give you more flexibility over the outcome.

This Retry feature lets you quickly regenerate your model without consuming more Credits if the initial result isn't what you were hoping for. More importantly, you don't have to wait until generation has completed. Within 15 seconds, you'll get a quick look at your model through a preview slideshow, allowing you to assess whether it meets your expectations. If it doesn't, simply hit the Retry button, and a new version will be generated instead.

After your textured model is ready, Retry is still there for you in case you are not satisfied with the final result. You might want to be more careful here, as retrying the generation will discard your current result.

Since Retry directly involves more computational resources, the feature is an exclusive benefit available to our subscribed users:

• Pro users have 4 retries available per model.
• Max users have 8 retries available per model.
• Max Unlimited users, as the name suggests, also enjoy unlimited retries.

Pick Your AI

Previously with the release of Meshy-3 Turbo, we introduced a model selector for text to 3D generation which allows you to switch between Meshy-3 and Meshy-3 Turbo algorithms. Now we would like to expand this functionality further with the introduction of Meshy-4, where you can select the generative model in both Text to 3D and Image to 3D tasks among the options below.

• Meshy-4: Our latest and most advanced model, capable of generating flat geometry, sharp corners, and complex details.
• Meshy-3 Turbo: Our fastest model with less details. Suitable for organic modeling or heavily artistic styles.
• Meshy-3: Our legacy model, available for Text to 3D only.

You may also notice that the "Mode" selection in Image to 3D has been replaced. This is because it has an exact correspondence to the underlying model: the Organic Mode was powered by Meshy-3 Turbo and the Hard Surface Mode was powered by Meshy-4.

Above and Beyond

Rome was not built in one day, and neither is great art. We never envision a future where an once-and-for-all AI completely replaces the work of artists and designers. Instead, the creative process is as much about the journey as it is about the destination, and technologies should serve as a vehicle that enhances creative abilities, allowing you to push the boundaries of imagination while still retaining full control over your vision.

We are here to support that process, providing the tools helping you to focus on what you do best—it is your creativity, intuition, and personal touch that truly bring a virtual world to life.

Original source

Update (September 2026)

The Meshy visionOS viewer described in this post has been retired and is no longer available on the App Store. Meshy now offers an official mobile app for iPhone and Android. Get it from the Meshy app page, the App Store, or Google Play.

Unleashing 3D art creativity for everyone, from hobbyists to professionals, has always been our mission at Meshy. Today, we are thrilled to introduce a new way to experience your 3D ideas with the power of Apple Vision Pro - Meshy visionOS!

Dive into a vibrant world of creativity where the boundaries between imagination and reality blur. Imagine thousands of stunning AI generated 3d models by Meshy's innovative community, all brought to life right in front of your eyes using Apple Vision Pro. With Meshy visionOS, you can customize these dynamic models and transform your room into a personalized 3D canvas!

Immersive 3D Viewing

One of the most captivating features of Meshy on Apple Vision Pro is the ability to experience a variety of 3D models generated by Meshy in different sizes and orientations. From intricate details to full-scale structures, every creation is vividly brought to life in stunning clarity. Apple's Vision Pro provides a transformative visual experience that makes you feel as if you can reach out and touch your creations. It is so much fun to explore the models from every angle, zoom in to see the details, and interact with them in a way that feels natural and intuitive. You can truly feel like as if these 3D models are part of your real world.

Transform Your Space

Imagine being able to transform any space into your personal 3D wonderland. With Meshy on Apple Vision Pro, you can virtually place, rotate, and scale 3D models with ease.

Our intuitive interaction support allows you to create captivating scenes that reflect your artistic vision, making it simpler than ever to reinvent your space. Whether you're creating a small animal town, setting up an tea party, or just playing around with 3D decorations, the possibilities are endless.

Hands-On Fun

What sets Meshy on Apple Vision Pro apart is the unprecedented level of interactive engagement it offers. You can literally put your creativity at your fingertips by manipulating models directly with your hands. Apple's advanced hand-tracking technology provides precise control, allowing you to toss, rotate, resize, and experiment with 3D models seamlessly.

Get Started with Meshy visionOS

Ready to discover the power of Meshy visionOS? Download Meshy visionOS on the App Store!

Meshy's dedicated 3D model viewer for Apple Vision Pro highlights the extraordinary potential of AI in 3D creativity. Our community thrives on innovation, and we are eager for you to join us in pushing the boundaries of what's possible. Generate and customize your own models at meshy.ai, then watch them come to life within the app.

Your feedback is invaluable to us. Meshy visionOS was released in its beta stage, so early users may have encountered features that were still in development or areas where improvements could be made.

For suggestions, comments, or bug reports, reach out to us on X at @MeshyAI or our Discord channel. We are always listening and constantly striving to improve your experience.

Original source

We are thrilled to announce the release of Meshy-3 Turbo, the latest, fastest, and most powerful version. Building on the strong foundation of Meshy-3, this release brings two major enhancements that will accelerate the way you create and animate 3D models. Dive into the future with us as we explore the incredible new features of Meshy-3 Turbo.

Animation: Bring Your Models to Life

Meshy-3 Turbo introduces an all-new animation feature that allows you to compose rigged character animations directly from models generated in Meshy. Whether it's a smooth walk cycle or an energetic run, your characters are now ready to be integrated into 3D environments right out of the box. This new capability saves hours of rigging and animating, streamlining your workflow and bringing your characters to life faster than ever before.

With Meshy-3 Turbo, creating animated characters is as simple as a few clicks. Simply open your generated 3D model, click on the "Animation" icon on the right sidebar, and select your character type. Adjust the character orientation, place rigging points on the model, then begin generating the animations. After the animations are ready, you can preview pre-defined animations like walking or running. Once you're satisfied with the preview, download your animated, rigged character to use immediately or create more diverse animations on your own.

Turbocharged Speed: 3x Faster Generations

In addition to the all-new animation capabilities, Meshy-3 Turbo delivers substantial speed improvements. We have developed and optimized a new diffusion-based generative model so that generations are up to three times faster, with record-level generation speed of Image to 3D at less than 1.5 minutes. We are also working on accelerating Text to 3D generation to below 2 minutes with our latest model. Thanks to the updated generation algorithm, you will be able to preview the generation result within only 30 seconds into Image to 3D generation, without having to wait for the finalized model to be ready. This means you can iterate more quickly, experiment more freely, and bring your projects to completion faster than ever.

Experience blazing-fast generation speeds that let you focus more on creativity and less on waiting. Meshy-3 Turbo ensures your models are ready in a fraction of the time, allowing you to rapidly explore ideas for your 3D environments with ease.

What's More

• Community challenge: Introducing a new weekly community challenge for all Meshy users! Each week, we will announce a unique theme and corresponding hashtag, which you can include in your prompt text of Text to 3D generations to participate. We hope that this event offers a fun and engaging way to showcase your work and explore different styles with Meshy. By using the designated hashtag in your prompts, you'll have the chance to earn special badges, additional Meshy credits, and more. Join us in this exciting new tradition and let's create together!

• Multi-currency support: Meshy-3 Turbo will soon offer multi-currency support, enhancing convenience and accessibility for our global community. With this feature, you will be able to transact in various major currencies, such as Euros, British Pounds, and Japanese Yen. This update eliminates the hassle of currency conversion for a more seamless and transparent purchasing experience.

Together, we're pushing the boundaries of what's possible in 3D content creation, and we hope you enjoy exploring the new features of Meshy-3 Turbo. As always, your feedback is invaluable to us. Let us know what you think on Discord, and stay tuned for more exciting updates from Meshy!

Original source

Today marks the first anniversary of Meshy, and what a journey it’s been! We're thrilled to unveil Meshy-3, our latest leap in 3D generative AI, enhancing Text to 3D and introducing a new Image to 3D pipeline.

Today, as we celebrate the first anniversary of Meshy, we're excited to unveil Meshy-3, our latest advancement in 3d generative AI. Meshy-3 brings groundbreaking updates to Text to 3D and introduces a new Image to 3D pipeline. Dive into the future with us:

Text to 3D: Ultra Realism

Experience unparalleled detail with our new high-poly model generation, suited for next-gen games and epic film scenes. The sculpture style in Meshy-3 Text to 3D allows users to create high-poly models using Meshy-generated displacement maps. Our vision with this feature is to craft models with photogrammetry quality, complete with crisp PBR maps for stunning realism:

You can use this feature in Text to 3D by picking the "Sculpture" style. It takes Meshy a few minutes to generate the high-poly model, and then you can download the baked model with PBR textures (displacement, normal, curvature, and AO maps included) for your DCC or game engine.

Revolutionary PBR Texture Generation

It has always been a challenge to use AI-generated models under dynamic lighting, that's why Meshy-3 Text to 3D allows you to generate PBR (Physically-based rendering) maps. These maps not only enhance the realism of multi-material assets but also boost the level of detail on all surfaces.

Simply select the "PBR" style in Text to 3D to get started.

Superior Mesh and Texture

With Meshy-3, the base mesh and texture quality is further enhanced, pushing the boundaries of 3D AI modeling.

Image to 3D: Vision to Creation

Meshy-3 brings you a brand-new Image to 3D system, with more natural shapes and clearer textures. This system allows for seamless image to 3d model conversion, enhancing creative possibilities.

What's more...

Smart healing of textures.
Correct AI-induced anomalies like extra eyes or misplaced texture features with our new smart healing tool.

Prompt helper.
Struggling with the right words for your 3D prompts? Use our prompt library to compose prompts effortlessly, streamlining your creative process.

Community Badges.
Join our vibrant community of creators and earn badges as you create and share.

As we reflect on this past year since Meshy’s inception in April 2023, we're amazed by the progress and community growth. Stay tuned for more updates as we continue to push the boundaries of what generative AI can achieve in 3D modeling. A huge shoutout to the Meshy team, and a big thank you to you, our fellow users, for being part of the Meshy community!

Get Involved

Ready to explore the capabilities of Meshy-3? Register now and start creating! Your feedback drives our innovation, and we can't wait to see what you'll generate next!

Original source

Hot on the heels of last month's Meshy-2 launch, we're proud to unveil Meshy-2.5. This upgrade offers a few handy tools that streamline your creation, and boosts the quality of Text to 3D.

Text to 3D: new tools and enhancements

AI Texture Editing. To enable better control and address occasional quirks of AI, we're introducing a feature that allows for regional, AI-assisted texture editing on your models. You can now easily correct or enhance specific areas for a more polished texturing.

AI texture editing is a game-changer for precision and creativity, and it enhances Meshy in several key ways:

• Enables face repainting to correct AI inaccuracies, and improve face resolution.
• Simplifies texture by removing extraneous details, like unwanted spots, and resolving issues like multiple faces.
• Provides control over specific texture areas, allowing for easy modifications, such as changing logos on Meshy-generated footwear.

Image to Prompt. Ever struggled to find the right words for your Text to 3D prompts? Our new Image to Prompt feature converts images into descriptive prompts, simplifying the process of accurately conveying your vision. This streamlines your creative process, transforming visual inspiration directly into actionable prompts.

Improved Geometry and Texture Quality. Enjoy significantly improved geometry and texture with the new Text to 3D upgrade, ensuring your creations are more detailed and realistic than ever. Compared to Meshy-2, meshes generated by Meshy-2.5 have more elegant shapes, clearer structures, and more geometry details. The texture is cleaner and more accurate, with fewer artifacts.

We've optimized the geometry and texture quality of preview models to assist you in making more informed decisions during the preview stage.

Pricing, model downloads & accessibility

We've eliminated the 20-credit fee for downloading both coarse and refined meshes in Text to 3D across all of our users. This change aims to make creative experimentation more accessible.

Pro & Max users can now download the CC0 community models for free. This saves the time you remix and regenerate the model.

Most importantly, we're excited to offer a 7-day free trial of our Pro plan to all users. Start your free trial today and explore the endless possibilities with Meshy 2.5!

Our future focus prioritizes 1) quality, 2) control, and 3) speed, in that order, to best meet the needs of 3D content creators, especially CG artists. Developing Meshy and engaging with our user community has been an incredible journey. Your ongoing support and feedback are invaluable to us. What features are you most excited about? Let us know your thoughts on Discord!

Original source