Resemble Ai Release Notes

Resemble AI logo

While others need constant retraining, DETECT-World uses physical reality to train faster and catch real-time threats with benchmark-leading accuracy

What sets Resemble AI apart is speed. They track where deepfake detection is heading and have solutions ready before we need them.”
— Tomaz Levak, CEO of Umanitek

MOUNTAIN VIEW, CA, UNITED STATES, August 12, 2026 / EINPresswire.com / -- Resemble AI, the leading deepfake detection model company, today launched DETECT-World, built on the same technology at the center of AI's biggest research race this year: world models.

Resemble DETECT-World doesn't just look for familiar signs of a fake, it checks whether what's on screen could actually happen in the real world, much like the models built to mimic it. AI's most advanced labs are racing to build these systems, which understand physical reality well enough to simulate it perfectly. Resemble AI is the first company to apply that technology to detection of everyday communications like phone calls and video meetings, enterprise use cases like ID verification and claims reimbursements and impact-critical content like political disinformation campaigns.

Deepfake Detectors Must Understand Physical Reality to Keep Up

Hugging Face’s model count has grown from 2 million to nearly 3 million in the past 9 months, adding thousands of new AI models every day, a growing share of them built to generate images, audio, and video. Humans alone miss roughly 7 in 10 deepfakes unaided, and while pattern-based detectors perform better, they still need to be retrained on each new generator to keep up. DETECT-World builds on that same pattern-based foundation, then adds a check for physical reality. In a video of a person speaking, the model may look for things like whether lighting changes naturally across the scene, or whether a shadow matches its source. A fake that passes every known pattern check can still be caught by Resemble DETECT-World if it breaks the laws of physics.

"Frontier labs are moving fast to build the next generation of generative AI. We're moving just as fast to build the detection that keeps pace with them. Our mission is to be the most reliable line of deepfake defense in a world full of companies whose goal is producing content indistinguishable from reality," said Zohaib Ahmed, CEO of Resemble AI. “We help everyone from government agencies to gig platforms to global telcos to detect that content and operationalize the results into fraud prevention.”

Deepfakes are Now the Top Critical Cybersecurity Threat

Gartner recently named deepfakes one of four critical cybersecurity threats where attackers currently hold the advantage, warning that generative AI has made deepfake creation more realistic, accessible, and increasingly real-time, expanding attackers' ability to impersonate identities in fraud, social engineering, and recruitment scams. Without a detector that keeps pace with new generators the moment they ship, those attack paths stay open until someone notices and retrains the detector.

The Challenge and Reward of Building a World Detection Model

Resemble AI built DETECT-World on world architecture, the same class of technology researchers across the AI industry are betting billions of dollars on this year as the next step beyond language models. Combining that architecture with pattern-based detection requires a foundation most detection companies don't have: a detection model already trained at scale, plus world architecture itself. Resemble AI already had the first piece. Building the second on top of it is what lets DETECT-World train faster against new generators and reach high-confidence day-zero coverage sooner than detection built on patterns alone.

“What sets Resemble apart is speed. They track where deepfake detection is heading and have solutions ready before we need them,” said Tomaz Levak, CEO of Umanitek.

What It Catches

• Executive impersonation on live calls. Real-time face-swaps on Zoom or Teams, discovered by looking at facial and or body continuity across frames, minor distortions in features while speaking or moving, and probable lighting and reflections.
• Synthetic identity fraud with official documents. Deepfake selfies and synthetic ID photos, caught by inspecting liveness cues like blink patterns and depth inconsistencies, and by detecting injection-attack signatures where a live camera feed has been spoofed or swapped out.
• Claims and evidence review. Videos and images of damage for reimbursement or claims processing inspected for lighting and shadow inconsistencies, unnatural continuity between frames, and signs of digital splicing or manipulation.

No detector will ever be 100% accurate, but across every modality, DETECT-World is now getting as close as it gets: audio accuracy reaches 99.5% as benchmarked by Podonos and now covers 54 languages, and detection overall has been benchmarked across more than 250 generation models.

When It Comes To Detection More Information Is Better

No single control stops deepfake attacks, which is why DETECT-World isn't meant to work alone. It is the detection layer of Resemble AI's security stack. Explainability, powered by Resemble Intelligence, turns a DETECT-World score into evidence an analyst can act on immediately. Identity Management enrolls a voice or likeness for protection, while Resemble Signal checks callers and content against known fraud patterns across audio, image, video, and text. Watermarking declares origin for content an organization creates. Together, these layers, available as composable APIs, protect enterprise decisions from start to finish.

Availability

DETECT-World, the deepfake detection model powering Resemble Detect, is now available via real-time streaming and batch API at resemble.ai. Deployment options include cloud, VPC, on-premises, and air-gapped environments, with SIP, and SIPREC integration for live call and telephony use cases. Enterprise licensing is available through Resemble AI's sales team.

About Resemble AI

Resemble AI is the deepfake detection models company that delivers real-time, accurate detection results combined with intelligence and next-action recommendations. It's what enterprise teams need to detect a threat, protect their content, and make informed decisions quickly. Our models work across every modality, integrate with your workflows, and adapt to your deployment needs to make detection comprehensive, seamless, and secure. Learn more at resemble.ai.

Original source

The field of deepfake detection has a problem

Detectors are much less effective on deepfakes created by generators newer than they are, as well as unfamiliar models. Resemble AI says by rethinking what is being detected, it has moved beyond this performance limitation with its new deepfake detection model.

Researchers with the Vector Institute have shown “performance drops sharply” if the deepfake detector has not been trained on content from the generator that created a given deepfake, according to a Resemble AI post explaining its new DETECT-World software.

Instead of training a deepfake detection model to recognize artifacts found in the creations of known generators, the company says it built its model to consider “does this content violate my model of how physical reality works?”

The company claims 99.47 percent audio deepfake detection accuracy, 98.2 percent detection accuracy for video and 95.8 percent for images.

DETECT-World is the third generation deepfake detection model developed by Resemble AI, following DETECT-2B and DETECT-3B Omni, the latter of which added image and video capabilities. It is now operating within Resemble Detect, the company’s real-time deepfake detection API.

The change is in implementing the World-Vision Hybrid Encoder to look for inconsistency with the physical world, and return a manipulation probability score, per-frame scores and a spatial heatmap.

An example of its effectiveness, according to the post, is seen in DETECT-World’s 95 percent accuracy catching Haotian-style real-time face-swap attacks. These attacks were misclassified by leading detectors at rates near 100 percent in an investigation published by 404 Media in May. The new model had not been exposed to this type of attack before, Resemble says.

Mountain View, California-based Resemble AI was sitting second on Hugging Face’s Speech Deepfake Arena Leaderboard when the benchmark was paused for its next stage.

Original source

Zohaib Ahmed, CEO and co-founder of Resemble AI, answered Lets Data Science's questions at the launch of the company's Deepfake Threat Report and DETECT-World detector, and spent most of the exchange explaining what the headline numbers do not say. A single domain seizure carries roughly 14,000 of the 15,736 documented victims, the $2.24 billion figure is statutory exposure rather than losses, and 97.7 percent of verified attacks disclose no loss figure at all. He also gave externally validated false positive rates for the new detector and said which numbers remain internal.

Resemble AI published its Deepfake Threat Report today, alongside DETECT-World, a detector the company says judges whether audio, images and video behave like the physical world instead of only hunting generator artifacts. The report's headline is $2.24 billion in liability exposure across 15,736 documented deepfake victims.

Before publishing, Lets Data Science put seven questions about the methodology to Zohaib Ahmed, the company's CEO and co-founder, and he answered all of them in writing. What makes the exchange worth reading is that he spent most of it explaining what his own numbers cannot say.

One seizure carries most of the victim count

Of the 15,736 documented victims, roughly 14,000 come from a single row: the June 11 seizure of CFAKE.com and SOCFAKE.com by the Department of Justice and Homeland Security, announced on June 12 as the first domain seizures under the TAKE IT DOWN Act. The sites published nonconsensual AI-generated sexual imagery of adult public figures, and the 14,000 figure is the enforcement agencies' own count, drawn from French investigators who identified about 300,000 images and 7,000 videos depicting roughly 14,000 people.

Ahmed volunteered the sensitivity rather than defending the headline. Without that row, the report is left with 1,736 documented victims, and statutory exposure of $137.2 million instead of $2.24 billion. "On every consequence axis in this dataset, a single incident carries most of the total," he told Lets Data Science. "One incident is 46% of documented losses, one is 87% of documented files, one is 89% of documented victims. That is not a quirk of our methodology, it is what disclosure-based data looks like."

The seizure also cuts against the report's own central finding. Excluding CFake, 61% of the remaining victims are private individuals rather than public figures. "Incidents cluster on famous people, but the harm mostly lands on ordinary ones," Ahmed said. "CFake cuts against that finding and we report it anyway."

A ceiling, not a damage estimate

The $2.24 billion is 14,915 documented NCII and CSAM victims multiplied by the $150,000 statutory election a victim can make under 15 U.S.C. 6851. Asked who actually carries that exposure, Ahmed declined to say. "We don't take a position on who carries it. That question is being litigated right now," he said, adding that the report labels the figure "exposure, never losses" and never adds the two together.

He did note which direction the statutes are moving. Under the DEFIANCE Act's $250,000 ceiling, passed by the Senate in January and pending in the House, the same victim count implies $3.73 billion. Courts and legislatures are already attaching real numbers: an Amsterdam court issued an injunction against an AI image generator at 100,000 euros per day per defendant, and Minnesota's nudification statute carries $500,000 per incident.

What the database structurally cannot see

The database is built from public disclosure, and Ahmed was blunt about the blind spots: completed frauds that companies never announce, sectors like healthcare that surface through enforcement rather than news, and abuse in closed channels that may never surface at all. In the small set of fraud cases where coverage states an outcome, he said, most of the frauds completed, which is why the company refuses to publish an interception rate.

The starkest gap is financial. 97.7% of verified attacks carry no stated loss figure, so the report documents $6.95 million in losses while the FBI logged $893 million in AI-enabled fraud complaints in 2025, a figure the bureau itself considers understated. Resemble publishes no multiplier of its own; where it extrapolates, it uses published reporting rates of 14.3% to 26% and labels the result implied rather than documented.

The same discipline ran through the Grok analysis: the company reproduced the widely cited figure of 23,000 images appearing to depict children as CCDH's modeled estimate, carrying two layers of sampling uncertainty, and rejected an "80 million Grok images" number it could not trace to a primary source.

The detector's numbers, labeled unevenly on purpose

Asked for the false positive rate behind DETECT-World's 99.5% audio accuracy claim, Ahmed gave it: 0.66%, externally validated on a Podonos benchmark published today, against 2.50% for the company's previous detector. Video and image are a different story, at 2.01% and 4.31% respectively, and both remain internal measurements. "I will label them that way rather than present them as equivalent," he said.

On day-zero coverage, the company ran a fixed evaluation across more than 260 generator models it had not trained on, with at least 50 samples each. Detection on deepspeak2 moved from 32.0% under pattern-matching to 96.0% with the world-model approach, with similar jumps on sadtalker and wav2lip.

Asked whether world models simply arm both sides, Ahmed did not argue. "We presume detection and generation will keep improving against each other indefinitely," he said. The number that matters, in his view, is how much performance survives in the minutes and hours after a new generator ships.

What practitioners should change on Monday

Ahmed's advice for engineers building identity verification, independent of buying anything: prioritize image and multimodal detection, because image-only attacks now make up 37.1% of incidents against 28.6% for video-only. Stop relying on repeat-target signals for intimate-imagery abuse, since NCII in the data shows zero repeat targets, meaning hash-matching catches recirculation but not the first-generation attacks NCII now mostly is.

And put verification friction in-line, before money moves: in every reported fraud case with a stated save, the interception happened at the transfer approval, the callback or the verification step. "Detection after the fact gives you evidence rather than your money back."

Correction — September 14, 2026

The domains were seized on June 11, 2026, and the Department of Justice announced the action on June 12. An earlier version used the announcement date as the seizure date.

Key Points

1. Resemble AI CEO Zohaib Ahmed told Lets Data Science that one seizure row carries 89% of documented victims, and that the report publishes every headline number with and without it: $2.24B in statutory exposure becomes $137.2M.

2. The company documents only $6.95M in losses because 97.7% of verified attacks disclose no figure, against $893M in AI-enabled fraud complaints the FBI logged in 2025, a gap Ahmed says is what disclosure-based data looks like.

3. DETECT-World's audio false positive rate is 0.66%, externally validated by Podonos; video and image FPRs are internal numbers, and Ahmed labels them that way rather than presenting them as equivalent.

Original source

Resemble Watermarker now embeds imperceptible, tamper-resistant marks across every content modality - giving AI output a verifiable chain of custody

MOUNTAIN VIEW, CA, UNITED STATES, June 24, 2026 / EINPresswire.com / -- Resemble AI, the generative AI security platform, today expanded Resemble Watermarker from audio-only watermarking to full multimodal coverage, adding image, video, and text. The expansion is powered by PerTh Multimodal, a ground-up rebuild of the model behind the company's existing audio watermarker, which is also available as open source. Now with PerTh Multimodal, enterprises and creators can include imperceptible, tamper-resistant marks across every content type they generate. The update arrives as synthetic media threats reach an inflection point and organizations face mounting pressure to prove the origin and authenticity of the content they create and distribute. In a June 2026 report, Gartner named deepfakes one of four critical threats where enterprise defenses are most overmatched, the cases where the attacker now holds the advantage.

Impersonation and fraud are now a core business risk

The risk is what bad actors do with synthetic media: deepfaking your assets and impersonating your brand, your executives, and your talent to commit fraud and damage reputation. As found in The 2025 Resemble AI Deepfake Threat Report, brand and reputation attacks generated 156.3 billion impressions last year, more than all other attack categories combined. Across every category, synthetic media drove 296.4 billion impressions, the equivalent of roughly 53 for every internet user on earth, which means the damage a fraudster can do is far-reaching and as fast as their access to social media.

What makes it hard to fight is a combination of proof of ownership and the ability to remove the unauthorized content: when a fake of your brand or an executive spreads, most organizations have no way to show which content is authentically theirs and which is not. Research presented at ACM in 2025 found that only about 38% of AI content generators watermark their outputs adequately, so most AI-generated assets carry no verifiable chain of custody. While deepfake detection catches the fakes other people make, watermarking proves the authenticity of the content you make, so a genuine asset can be told apart from an impersonation, and the question of who made something, and whether it has changed, is answerable by the content itself.

Resemble Watermarker: the authentication layer for AI-generated content

Resemble Watermarker is a declaration of ownership placed inside the content itself. It proved the approach in audio first, and this update extends the same imperceptible, in-signal mark to image, video, and text, so one product now covers every content type an enterprise generates. Where metadata-based approaches can be stripped the moment a file is re-uploaded, compressed, or re-encoded, Resemble Watermarker embeds the mark directly into the content.

A properly watermarked asset can answer four questions at any point in its lifecycle:

• Who issued this content?
• Was it AI-generated, AI-altered, or authentic?
• Has it been tampered with since creation?
• Are other provenance marks present, including SynthID or C2PA credentials?

Resemble Watermarker reads SynthID and C2PA markers, and it writes and signs C2PA credentials, so it delivers both layers regulators are converging on: a provenance manifest any compliant tool can read, and an in-signal watermark that survives when that metadata is stripped. Assets can also carry an explicit attribution mark that encodes issuer identity, content date, and authenticity status. Support for on-device, VPC, and air-gapped deployment lets enterprises run it inside their own environments.

“A watermark is a declaration of ownership placed inside the content, ideally at the moment of creation, so the question of who made something and whether it changed can be answered by the file itself, not the channel or handle it came from,” said Zohaib Ahmed, CEO and Co-Founder of Resemble AI. “Deepfake detection answers whether content is AI-generated or altered, a watermark proves who it belongs to, and that is a foundational authentication layer every organization needs."

Built for the regulation that is arriving, in the EU and beyond

Resemble Watermarker is designed to map directly to the obligations enterprises face as AI content rules move from rule-making to enforcement. The EU AI Act's Article 50 transparency obligations take effect August 2, 2026, requiring AI-generated and manipulated content to carry machine-readable marking, with penalties up to 15 million euros or 3% of worldwide annual turnover.

The EU is not moving alone. China's mandatory labeling rules took effect September 1, 2025, requiring all AI-generated content to carry both a visible label and an embedded watermark or metadata identifier, the same two-layer pattern the EU is now adopting. More jurisdictions are expected to follow as the approach becomes the global default. Resemble Watermarker produces the machine-readable marking these rules call for across all four modalities through a single API, giving compliance, legal, and trust and safety teams an authentication layer that travels with the content itself.

Availability

Resemble Watermarker powered by PerTh Multimodal is available today through a self-serve API at resemble.ai, and audio watermarking remains available as open source. Enterprise licensing, including VPC and on-premises deployment, is available through Resemble AI's enterprise sales team. Resemble Watermarker integrates with Resemble AI's broader generative AI security platform, including Resemble Detect for deepfake detection.

About Resemble AI

Resemble AI is generative AI security that gives organizations the tools to verify, detect, and act on synthetic media attacks and fraud attempts — anywhere they appear. Built on our foundational models, Resemble provides a layered trust stack that turns model output into decisions companies can understand, audit, and defend. Founded in 2019, Resemble AI is trusted by global enterprises and government agencies to secure the full lifecycle of synthetic media across audio, video, image, and text. Learn more at resemble.ai.

Original source

RESEMBLE VOICE CREATION

Clone any voice or design one that doesn't exist yet.

Two paths to a production-ready voice, powered by the open-source model that outperformed ElevenLabs in blind evaluation.

MODEL

Powered by Chatterbox

Chatterbox is MIT-licensed and open-source. Clone any voice from 10 seconds of audio, or generate one from a text description. PerTh watermarking available on every output.

~63.75%

Of evaluators preferred Chatterbox over ElevenLabs in an independent blind study.

Benchmarks available upon request.

RESEMBLE AI VOICE CREATION PATHS

Choose based on whether you're working from existing audio or building from scratch.

• Rapid Clone — 10 seconds of audio
  READY IN <1 MIN

• Pre-built professional voices
  DEPLOY INSTANTLY

• Voice Design from a text description
  3 OPTIONS RETURNED

• On-prem self-hosting via Docker/Kubernetes
  MIT LICENSED

• Professional Clone — 10–25+ min of audio
  TRAINED IN ~40 MIN

• PerTh watermark on every output
  AVAILABLE

OUR CAPABILITIES

Everything a production voice needs, built in.

Designed to keep your voice consistent across languages, use cases, and API calls.

Custom pronounciation

Generates multiple pronunciation variants for any term. The approved pronunciation applies across every voice, session, and API call automatically.

Multilingual cloning

Clones a voice once and generates speech across 23 languages, with accent and vocal character retained without separate training runs.

Per-use-case variants

Prompt one clone for unlimited variants: conversational, commercial, phone agent, and more, each in natural language.

<1 min

Time to a functional Rapid Clone from 10 seconds of audio

23

Languages supported with zero-shot voice cloning

10M+

Hugging Face downloads, 24,000+ GitHub stars

RESEMBLE VOICE CREATION USE CASES

Built for teams where the voice has to be consistent, accurate, and yours.

Every use case below starts with a problem a generic voice couldn't solve.

Voice agents

The problem:
Generic voices don't fit the brand. Cloning takes too long to iterate on.

Resemble AI solution:
Rapid Clone in under 1 minute with per-use-case variants available instantly.

Games

The problem:
Casting and recording 40 characters is slow, expensive, and hard to iterate on.

Resemble AI solution:
Voice Design generates distinct characters from a text description. No casting, no studio time.

Audiobooks & long-form

The problem:
Long-form narration demands a consistent voice with full emotional range across hours of content.

Resemble AI solution:
Professional Clone trains on 10 to 25+ minutes of audio and delivers consistent, expressive output at any length.

Media & Podcasts

The problem:
Localizing content means re-recording the host in every language.

Resemble AI solution:
Clone the host voice once and generate in 23 languages with accent and emotion retained.

Healthcare & HealthTech

The problem:
Drug names, legal terms, and brand vocabulary mispronounce on every generic engine.

Resemble AI solution:
Lock the correct pronunciation once. Applies across every voice and API call.

Regulated industries

The problem:
Sensitive voice data cannot leave the organization's infrastructure.

Resemble AI solution:
MIT-licensed. Self-host via pip install. On-premise via Docker/Kubernetes. No forced cloud dependency.

SOC2 Type II

Enterprise plans

EU AI Act ready

Mandatory Aug 2026

GDPR Compatible

On-prem available

HIPAA Compatible

Air-gapped deployment

SSO / SAML

Enterprise identity

C2PA Standard

Content provenance

TRUSTED IN PRODUCTION

Trusted where it matters most

When the stakes are real, rely on Resemble AI

AI Voice Reconstruction

AI narration for The Andy Warhol Diaries, generated from three minutes of source audio.

Personalized TTS

354,000 personalized audio messages generated. 7x revenue impact on fan engagement.

Multilingual Audio

Production-grade voice generation at scale for educational content across multiple languages.

Consumer Voice Cloning

Parents record 25 sentences. Resemble clones the voice. Bedtime stories narrated in the parent's own voice. 4.8 App Store rating.

INTEGRATIONS AND DEPLOYMENTS

Go live in hours, not sprints.

• Rest API
• Python SDK
• Node.js SDK
• WebSocket streaming
• ZIP or WAV dataset upload for cloning
• On-prem available
• Webhook callback on training completion

Frequently asked questions

What is the difference between Rapid Clone and Professional Clone?

Rapid Clone needs 10 seconds, delivers in under 1 minute. Professional Clone needs 10 to 25+ minutes, trains in ~40 minutes, and produces a voice with full emotional range nearly indistinguishable from the source.

How does Voice Design work?

Describe the voice you want in natural language: age, accent, tone, style. The API returns 3 distinct candidates. Choose one and generate speech immediately, even while the voice finishes building in the background.

Can a cloned voice speak multiple languages?

Yes. Chatterbox Multilingual supports zero-shot cloning across 23 languages. Clone once in any language and generate speech across all 23 without separate training runs. Accent and vocal character are retained.

Is consent required for voice cloning?

Yes for Professional Clone. Explicit verifiable consent from the voice talent is required before training data is uploaded. Consent workflows are built into the platform.

Can I create multiple variants of the same cloned voice?

Yes. Once a voice is cloned, prompt it in natural language for different use cases: conversational, commercial, phone agent, and more. Each variant is saved and reusable via API.

What are the deployment options?

Cloud API, open-source self-hosting via pip install, and on-premise via Docker/Kubernetes. Business plan or higher required for the Voice Cloning API. SLA documentation available on request.

How quickly can we integrate?

Most teams are live within a day using the REST API or SDK. Open-source deployment via Hugging Face requires no API key.

Original source

This report documents an alarming surge in AI-generated deepfake incidents, revealing over $1.28 billion in financial losses and a massive expansion of attacks targeting brand reputation, political stability, and individual safety.

Deepfakes Were an Enterprise Tech Problem; Now They Are Everyone's Problem

As synthetic content and deepfakes explode online, multimodal AI security isn’t optional; it’s essential for enterprises and everyday trust.”
— Zohaib Ahmed, CEO and Co-Founder of Resemble AI

MOUNTAIN VIEW, CA, UNITED STATES, March 26, 2026 / EINPresswire.com / -- Resemble AI, the leading generative AI security platform, on Thursday announced a new Deepfake Threat Report, free Google Chrome Detection Extension and Deepfake Detection Bot for X and new enterprise capabilities designed to help organizations and individuals identify and respond to synthetic media threats.

The announcements come as AI-generated content becomes more common across social platforms, news environments and digital communications, raising global concerns about impersonation, fraud and public trust. According to Europol, as much as 90% of online content could contain some form of AI-generated material by the end of 2026.

“For years, the industry focused on making AI-generated voice, image and video more realistic,” said Zohaib Ahmed, CEO and Co-Founder of Resemble AI. “We started by building voice AI models, so we understand how these systems work and how they can be weaponized. Multimodal generative AI security is now foundational for enterprises, employees and everyday people trying to navigate a world where more content is now synthetic.”

Deepfakes are driving fraud, abuse, and mistrust

The company’s new Deepfake Threat Report is based on Resemble AI’s proprietary database of verified deepfake incidents, continuously maintained and deduplicated from global media coverage. The report includes 1,567 unique incidents from 2025 drawn from 3,253 news stories. Financial losses are included, though most incidents disclosed no loss figure, suggesting the true economic impact is likely much higher. Each incident is classified by attack type and target category. Among the report’s findings:

• Nonconsensual intimate imagery and child sexual abuse material accounted for 20% of verified incidents.
• Nearly $1.3 billion in confirmed fraud losses were attributed to generative AI deepfakes, though roughly 80% of incidents disclosed no damages.
• The average corporate deepfake incident remained in the news cycle for 3.5 years, suggesting reputational harm can persist long after the original event.

New detection tools put real-time media verification in the hands of millions

Resemble AI also introduced two free tools designed to help users quickly assess the authenticity of digital content across platforms. The Deepfake Detector for Google Chrome is a browser extension that enables one-click scanning of image, video and audio content across media-heavy websites, displaying results as a color-coded badge: green (authentic), red (AI-generated) and yellow (uncertain). It also provides frame-by-frame analysis for video and segment-by-segment scoring for audio, and works across a wide range of traditional and social platforms, including X, Reddit, Instagram, TikTok, Facebook, LinkedIn, Vimeo and Twitch.

Complementing the extension, the Deepfake Detection Bot for X allows users to analyze suspicious image and video content directly within the platform. By tagging the @resemble_detect bot in a post with “is this fake?”, users can trigger an automated scan, with results returned in-thread to indicate whether the media is likely authentic or AI-generated. Designed to meet users where misinformation spreads most rapidly, the bot offers a fast, accessible way for journalists, researchers and the public to verify content in real time without leaving X.

Together these tools provide a simple and scalable way to evaluate suspicious media at a time when synthetic content is increasingly used to impersonate individuals, fabricate events and commit fraud. In addition to these free public tools, Resemble AI also announced three expanded enterprise capabilities:

• Multimodal watermarking lets organizations sign every piece of content at the moment of creation, establishing a tamper-resistant chain of custody across audio, image, and video. By automatically identifying file types to embed invisible signatures at the point of generation, the system provides a scalable provenance layer for high-volume production environments.
• Zero Retention Mode addresses the primary barrier to deepfake detection in regulated industries like finance and healthcare: the legal risk of cloud-based media storage. This configuration ensures submitted media is analyzed and immediately purged. Once processed, all data is rendered inaccessible to Resemble staff and systems, transforming a complex procurement hurdle into a compliant workflow.
• Reverse image search identifies "zero-day" synthetic media that lacks a prior digital footprint. While statistical models identify known patterns, this feature searches the web for matching images, cross-references against known debunked content, and traces back to original source material to detect novel fakes. By combining technical analysis with historical context in a single request, the system surfaces how widely a piece has spread and identifies synthetic media that statistical models alone may miss.

Resemble AI said these new releases are aimed at helping organizations establish authentication, detection, provenance and intelligence workflows as synthetic media becomes more widespread.

About Resemble AI

Resemble AI is the only complete generative AI security platform for creating, verifying, and detecting synthetic media across audio, video, and image. Founded in 2019, the company builds its own foundational models for both generation and detection, enabling a powerful advantage in identifying AI-generated content. Its open-source Chatterbox TTS model has surpassed 5 million downloads on Hugging Face, while its PerTh Watermarker helps secure content provenance from creation through compression, re-encoding, and format conversion. Its DETECT-3B Omni model is independently benchmarked for deepfake detection across audio, video, and image, trained against more than 160 AI models. Trusted by global organizations, Resemble AI is defining security for the age of synthetic media. Learn more at resemble.ai.

Original source

Rapid Voice Cloning

Resemble AI is launching Rapid Voice Cloning, a new feature of its platform that significantly expedites the process of generating voice clones. The company works in the elusive AI voice category focused on enterprise users.

Available today, Rapid Voice Cloning can duplicate voices from relatively short datasets and produce an output in just about a minute. The move, Resemble says, marks a significant development and will make voice cloning technology more accessible, empowering more users to create custom voices for their applications. The company believes it will make an impact across fields such as content creation, personalization and accessibility.

Resemble published multiple voice clone samples showcasing the prowess of the new technology. VentureBeat also tested the feature to see how it really works.

How does the new AI voice Cloning feature work?

When using Resemble’s web platform, users can create a digital replica of their voice by uploading an audio sample or recording a series of sentences. The company has been offering this feature for a while, but the process took time. Users had to record around 25 sentences or upload at least three minutes of voice content to set up the system, which would then take another hour or so to provide a clone.

Now, with the launch of Rapid Voice Cloning, users can get started with the technology more easily. All they have to do is give a clear audio sample of the target voice, lasting anywhere from 10 seconds to 1 minute. The company’s model under the hood instantly captures all the parameters, including accents, from the sample and gives the result for downstream use cases in a minute.

“While other state-of-the-art models often struggle to replicate the nuances and subtleties of different accents, Resemble AI’s advanced machine learning algorithms excel in this area. By analyzing and learning from just a 10-second voice sample, our Rapid Voice Cloning can create an AI-generated voice that faithfully mimics the unique intonations, pronunciations, and cadences of the original speaker’s accent,” the company noted in a blog post announcing the feature.

The company published a bunch of samples comparing its offering with Microsoft’s VALL-E and XTTS-v2 voice cloning models, complete with the input voice sample and the text used for the clone. The results were pretty impressive. However, when we created a free test account to see how the tech works for real, there were some clear gaps.

In our tests, the system mandated recording at least three long sentences, with no option to record a smaller 10-second sample. The processing was swift but it couldn’t recognize the speaker’s Indian accent and took the input by default as a voice sample in American English. This affected the accent of the output voice. However, it is expected to be fixed, since according to the company Rapid Voice Cloning will support most English accents.

Notably, the company will continue to provide the original cloning feature under the name of professional voice cloning. This option, with lengthy input requirements, will take time but support all English accents with support for text-to-speech and speech-to-speech use cases. Rapid cloning will only support text-to-speech generation.

Use across different categories

With Rapid Voice Cloning’s speed and dramatically reduced sample requirements, Resemble AI expects to see more users using the technology with faster iterations and deployments. The biggest adoption is expected from content creators who may use the tech to generate voiceovers, dubbing, narration and dialogue for their podcasts, videos, audiobooks or e-learning materials. The company also says businesses can create enhanced accessibility and personalization experiences with the technology.

“For example, a fitness app could use Rapid Voice Cloning to create a personalized AI coach that speaks to each user in a familiar voice, providing encouragement and guidance. Similarly, a virtual assistant could adapt its voice to match the user’s preferences, creating a more intimate and tailored interaction,” the company stated.

While it remains to be seen how the tech gets adopted, it is important to note that Resemble is not the only player cutting down the time to generate voice clones. ElevenLabs, another major player in the category, offers a feature called Instant Voice Cloning that needs at least a minute of clear audio to generate a clone almost instantly. Like Resemble, ElevenLabs also offers a professional version of the tool, which covers more languages and accents.

As of now, Resemble AI allows users to create one free voice clone. For more, users would have to take up a paid plan from the company, which starts from $29/month and goes up to $499/month. There is also the option of a pay-as-you-go personal plan or a bigger enterprise plan with custom pricing.

Original source

Rapid Voice Cloning

We’re excited to announce the launch of our groundbreaking new feature: Rapid Voice Cloning. This innovative technology allows you to create high-quality voice clones faster and easier than ever before, unlocking new possibilities for your voice-enabled projects.

Rapid Voice Cloning is a game-changing feature that streamlines the voice cloning process by enabling the creation of voice clones from remarkably short audio samples. With Rapid Voice Cloning, you can now generate a voice clone using as little as 10 seconds of audio data.

This revolutionary approach to voice cloning makes the technology more accessible and efficient than traditional methods that require lengthy audio recordings. By reducing the barrier to entry, Rapid Voice Cloning empowers more users to leverage the power of custom voices in their applications.

How Does Rapid Voice Cloning Work?

Under the hood, Rapid Voice Cloning employs cutting-edge algorithms to analyze and replicate the unique characteristics of a source voice from just a brief audio sample. Our advanced machine learning models are able to capture the essence of a voice and generate a high-fidelity clone in a matter of moments.

The process is designed with simplicity in mind. All you need to do is provide a clear audio sample of the target voice, lasting anywhere from 10 seconds to 1 minute. Our intelligent system takes care of the rest, delivering a fully-functional voice clone that’s immediately ready to use.

Rapid Voice Cloning offers a range of compelling benefits:

• Speed: Generate voice clones in seconds, enabling rapid iteration and deployment in your projects.
• Accessibility: With dramatically reduced audio requirements, voice cloning is now within reach for a wider user base.
• Seamless Integration: Rapid Voice Clones work flawlessly with our Web UI and API, allowing for frictionless use across your applications.
• Efficiency: Save valuable time and resources by eliminating the need to record and process lengthy voice samples.

Comparison to other State of the Art Models

We compare our results to other state of the art AI Voice cloning models including Microsoft’s VALL-E and XTTS-v2. Like VALL-E and XTTS-v2, our model has never seen any of the Voice Prompt speakers during training.

Breakthrough in Accent Retention

Resemble AI’s Rapid Voice Cloning technology is a true game-changer when it comes to capturing and reproducing accents with unparalleled accuracy. While other state-of-the-art models often struggle to replicate the nuances and subtleties of different accents, Resemble AI’s advanced machine learning algorithms excel in this area. By analyzing and learning from just a 10-second voice sample, our Rapid Voice Cloning can create an AI-generated voice that faithfully mimics the unique intonations, pronunciations, and cadences of the original speaker’s accent.

Applications of Rapid Voice Cloning

Rapid Voice Cloning is a versatile technology with a wide array of potential applications across industries. Its ability to create high-quality voice clones from short audio samples opens up exciting possibilities for content creation, personalization, accessibility, and more. Let’s dive deeper into some of the key use cases.

Content Creation

Rapid Voice Cloning can be a game-changer for content creators, allowing them to efficiently generate voiceovers, narration, and dialogue. Whether you’re producing podcasts, videos, audiobooks, or e-learning materials, this technology enables you to quickly create compelling voice content without the need for extensive recording sessions.

Imagine being able to clone the voices of multiple characters for an animated series, or generate localized versions of a product demo in various languages – all with just a few clicks. Rapid Voice Cloning makes these scenarios possible, empowering content creators to scale their productions and explore new creative avenues.

Personalized User Experiences

In today’s digital landscape, personalization is key to engaging and retaining users. Rapid Voice Cloning allows businesses to take this to the next level by integrating custom voices into their applications and services.

For example, a fitness app could use Rapid Voice Cloning to create a personalized AI coach that speaks to each user in a familiar voice, providing encouragement and guidance. Similarly, a virtual assistant could adapt its voice to match the user’s preferences, creating a more intimate and tailored interaction.

By leveraging the power of voice cloning, businesses can create more immersive and memorable user experiences that foster deeper connections with their audience.

Accessibility and Assistive Technologies

Rapid Voice Cloning has the potential to make a significant impact in the realm of accessibility and assistive technologies. For individuals with speech impairments or those unable to speak, voice cloning can provide a means of communication that preserves their unique vocal identity.

By cloning a person’s voice from a short sample recorded before the onset of their condition, assistive devices can generate speech that sounds authentically like the individual. This can be incredibly empowering, allowing people to express themselves in a way that feels true to who they are.

Moreover, Rapid Voice Cloning can be used to create more engaging and inclusive educational materials. By cloning the voices of educators or subject matter experts, learning content can be made more accessible to students with diverse needs, such as those with visual impairments or language barriers.

Prototyping and Product Development

Rapid Voice Cloning can significantly accelerate the prototyping and development process for voice-enabled products and features. By quickly generating voice clones, teams can test and iterate on their ideas faster than ever before.

For instance, a company developing a voice-controlled smart home device can use Rapid Voice Cloning to experiment with different voice personas and interaction styles, gathering user feedback early in the design process. This agile approach allows for more informed decision-making and can ultimately lead to better, more user-centric products.

Creative Industries

The creative industries are another domain where Rapid Voice Cloning can make a significant impact. In film and television production, voice cloning can be used to create dubbed versions of content for international markets, or to generate dialogue for computer-generated characters.

In the music industry, Rapid Voice Cloning opens up new possibilities for artistic expression. Musicians can experiment with incorporating cloned voices into their compositions, creating unique vocal textures and harmonies. DJs and producers can use voice cloning to generate bespoke vocal samples and hooks, adding a new dimension to their tracks.

Ethics and Responsible Use

At Resemble AI, we’re committed to developing voice cloning technology that is not only powerful and accessible but also grounded in strong ethical principles. We recognize the potential for misuse and are proactively taking steps to ensure that Rapid Voice Cloning is used responsibly and transparently.

Consent and Authorization

One of the cornerstones of ethical voice cloning is consent. We require that all users obtain explicit permission from the individuals whose voices they intend to clone. This consent must be freely given and informed, with the voice owner fully understanding how their voice will be used.

To facilitate this, we’ve built consent and authorization mechanisms directly into our platform. When creating a Rapid Voice Clone, users must confirm that they have obtained the necessary permissions and agree to use the voice clone in accordance with our terms of service.

Join Us in Responsible Innovation

To our users and the wider community, we invite you to join us in this commitment to ethical voice cloning. By using Rapid Voice Cloning responsibly, transparently, and with respect for voice owners’ rights, you can help shape a future where this transformative technology is used to create positive change.

Together, we can unlock the incredible potential of voice cloning while ensuring that it remains a tool for good. Let’s build a future where every voice matters and every innovation is grounded in ethics.

Original source

Resemble Enhance

While many AI companies race to find ways to use the technology to enhance or even create video, Resemble AI is focused on audio fidelity. The startup—which also offers an AI voice generator for businesses to use to create realistic human–like voiceovers—has launched 'Resemble Enhance,' an open-source tool designed to significantly upgrade the quality of historical audio.

The new service can take a distorted, fuzzy recording of a long-lost historical speech and then apply AI to make it sound like it was recorded or broadcast yesterday.

The Canadian company says Resemble Enhance is distinguished by its dual-module approach, combining a sophisticated denoiser—which removes static background hums and hisses—and an AI-powered speech enhancer. This combination not only removes unwanted noise but also enriches the overall quality of the audio.

Even though there are other audio restoration products on the market, Resemble’s combination of techniques could be a meaningful differentiator.

How does it work?

• The Resemble AI denoiser uses UNet, an AI model that helps to separate the different types of sounds that appear on a recording. It excels at filtering out unwanted noise from audio tracks, leaving just the speech as the focus.
• Once UNet does its job, the enhancer module kicks in, extending audio bandwidth and correcting distortions. This dual functionality, the company says, ensures that the final output is not just noise-free but also possesses the richness of contemporary recordings.

As an open-source tool, Resemble Enhance is accessible at no cost, a compelling option in the traditionally expensive market of media restoration services. The primary beneficiaries of Resemble Enhance are industries reliant on clear audio quality, such as podcasting, entertainment, and education. Additionally, this tool offers a new lease on life to historical recordings, potentially providing clearer insights into the past.

The tool's release comes when the demand for high-quality digital content is at an all-time high. Meanwhile, the open-source nature of Resemble Enhance positions it as a potentially disruptive force in a market currently dominated by high-cost proprietary solutions.

The convergence of AI in audio and video enhancement will likely pave the way for more comprehensive media restoration solutions.

By combining this tool with other video enhancers that use generative AI or other models to upscale and enhance images and faces—like GPEN or the well-known GFPGan—users can now achieve professional results with their own computers for a minimum investment.

To experiment with Resemble Enhance, users can visit the official Resemble AI website or download their models from the project’s official Github page.

Edited by Ryan Ozawa.

Original source

Generative AI in Film & TV (Special Report)

Over the last year, the entertainment industry narrative around generative AI has been intensely fraught. Now that the writers and actors strikes are over, Hollywood needs to consider exactly how studios and creatives will — and perhaps should — use gen AI models and software tools in the many varied creative processes involved in producing film or TV.

As the industry grapples with the game-changing technology in the coming months, Variety Intelligence Platform’s special report “Generative AI in Film & TV” examines the capabilities and limitations of gen AI models and emerging software through the lens of their present, near-term and future uses in film and television creation.

Generative AI offers an expansive range of possibilities at stages throughout the production value chain. In particular, we analyze the value and usability of this expanding, diverse and powerful set of AI models and tools for tasks in screenwriting, VFX, previsualization, content localization and sound editing.

And it’s already beginning to disrupt traditional methods, with generative AI tools currently used to automate some creative tasks. Still, its impact stands to be positive, as it eliminates rote work, speeds project timelines and allows productions to pursue previously impossible creative paths prohibited by constraints on cost, time and even physical reality. At the same time, its use promises to reduce the need for certain processes and as many workers to achieve the same level of output.

Since the first major public release of generative AI user applications last year, the exponential pace of research and development in the AI community has further improved the capabilities of these systems.

Alongside a rising tide of major tech companies variously developing, open-sourcing and productizing generative AI, several startups have emerged, offering best-in-class software and tools catering to entertainment production and post-production uses.

Research for this report partly draws from over 20 background conversations primarily conducted in August through November 2023 with media & entertainment (M&E) advisers; leaders and founders at generative AI startups; and independent filmmakers involving AI tools and techniques in their processes.

Company participants included Runway, Synthesia, Metaphysic, Wonder Dynamics, Digital Domain, Monsters Aliens Robots Zombies (MARZ), ElevenLabs, Deepdub, Resemble AI, DeepBrain AI, Luma AI and Fable Studio.

For the purposes of this report, our focus is the impact of generative AI in film and TV production. For additional research, we point back to our October edition, “Generative AI & Entertainment: Part 2,” examining the legal and labor risks of gen AI for the film and TV industry as well as potential mitigations. Our April 2023 report, “Generative AI & Entertainment,” presented the full scope of use cases across entertainment domains, including film, TV, music and gaming.

Read on to learn about:

• A full-page chart on generative AI capabilities, software and production uses

• How the tech will be used in screenwriting, VFX, content localization, more

• Ways advances in AI software development will impact content creation

Original source

With the launch of its new speech-to-speech feature, Resemble AI (https://www.resemble.ai/) is bringing natural-sounding AI voices to millions of developers and creators and continuing to expand the creative possibilities for human voices and beloved characters.

TORONTO and SAN FRANCISCO, Sept. 8, 2022 /PRNewswire-PRWeb/ -- Resemble AI today announced the launch of its speech-to-speech feature, to capture the unique style of human voices at scale and bring natural-sounding AI voices to millions of developers and creators across gaming, entertainment, e-learning, and more. Resemble AI's generative audio continues to expand the creative possibilities for human voices and beloved characters.

Now AI voices can perform a wide range of emotions, speaking styles or even singing using non-speech vocalizations. If the input audio is spoken in a different language, the resulting target voice will be able to speak in that language. To see speech-to-speech in action, watch this video.

“Resemble AIʼs mission is to make interactions with digital products as human and natural as possible,” says Resemble AI founder & CEO Zohaib Ahmed. “Now, developers and storytellers can create content in any language and any voice, without the expense or travel associated with recording studios.”

"We're really happy with the quality of voices we're able to develop for 'Animals Anonymous' using Resemble AI," says Fika Agency co-founder Adam Altman. "Now our entire team can record new episodes that feel consistent with the same voices our listeners are used to hearing."

The team at Resemble AI refined speech-to-speech when it used 3 minutes and 12 seconds of Andy Warhol's original voice recordings from the 1970s and 80s to produce synthetic voice narration for the Emmy-nominated and Dorian Award-winning Netflix docu-series, The Andy Warhol Diaries. The team made adjustments for emotion and pitch to the AI output of Andy Warhol's voice, and added human-like imperfections using reference audio clips of another speaker, as seen in this video about how Resemble AI's Style Transfer works.

"Resemble AIʼs mission is to make interactions with digital products as human and natural as possible," says Resemble AI founder and CEO Zohaib Ahmed. "We've dramatically accelerated and simplified the process of creating human-like AI voices by building hyper realistic synthetic voices that can match or expand the reach of voice actors. This means that developers, creators and storytellers can create content in any language, and with any voice, without the need for expense or travel associated with recording studios."

Starting today, Resemble AI customers on the Basic, Pro (new) and Enterprise plans will see a new option for speech-to-speech when creating a sentence in a clip within a project. Instead of using text as input, they can provide a spoken sentence by uploading a pre-recorded audio file or record directly through the interface. This enables high quality AI voices and allows the target voice to speak in a different language than the original voice–which must be the same voice that records a consent line at the start of any project.

Consent is a requirement for all Resemble AI projects and transparency is maintained throughout the process, as seen in this video about how Resemble AI voice cloning works. To learn more about how Resemble AI maintains ethical standards across the industry, visit https://www.resemble.ai/ethics/.

About Resemble AI

Resemble AI's technology is being used by some of the largest media companies in the world to create content that was previously impossible. Whether it's transferring a voice into dozens of other languages, creating thousands of dynamic personalized messages from celebrities, or creating unique real-time conversational agents, Resemble AI is changing how content is created.

With Resemble AI, creating engaging and high-quality voice content is now easier than ever, enabling content creators to add a whole new level of authenticity to their work, and will add a new level of immersion for the audience. Learn more at https://www.resemble.ai/our-solution/.

Media Contact

Amy Jackson, TaleSplash for Resemble AI, 1 4156092435, [email protected]

SOURCE TaleSplash for Resemble AI

Original source

More than 65,000 Users have Cloned 42,000 Voices

Toronto, Canada, Oct. 15, 2020 /PRNewswire-PRWeb/ -- Resemble.ai, a leader in generative deep learning voice technology today announced that it has created Localize -- voice AI technology to localize speech for the first time. Until now, entertainment companies, ad agencies, call centers and companies that needed to translate voices would have to use a different voice in each language. With Resemble.ai, voices can be cloned into any language - so George Clooney sounds like George Clooney, even when a movie has been dubbed into another language.

Resemble.ai clones voices at scale in seconds, as opposed to weeks. It has democratized this previously laborious, expensive process and cloned 42,000 voices for 65,000 users, including two of the largest global telecoms, two of the largest consulting companies, a top global broadcasting co, two of the largest entertainment conglomerates, one of the largest toy makers, and the leader in airport communications systems.

Localize fundamentally changes the way we think about speech. Much like text, video, and other mediums that have gone across borders, speech remains stuck to a single language. With deep learning and custom synthetic AI voices, we're breaking that barrier down.

Localize is able to keep any character's voice consistent across video games, movies, call centers, company videos, and more as they are translated to and from languages including French, German, Dutch, Italian and Spanish. A majority of the world speaks one of these six languages and the company also has near-term plans to introduce Localize for Korean, Japanese and Mandarin.

Normally, voice talent translation takes an average of two months and can tally hundreds of thousands of dollars. With Resemble.ai, it's accomplished in a week with maximum creative flexibility and efficiency. For entertainment companies, dubbing a script is logistically challenging and the fidelity of the production is oftentimes lost in translation. Resemble.ai offers a more attractive solution in a fraction of the time using the same talent whom production houses have already paid for.

Resemble Co-founder and CEO Zohaib Ahmed said, "Localize fundamentally changes the way we think about speech. Much like text, video, and other mediums that have gone across borders, speech remains stuck to a single language. With deep learning and custom synthetic AI voices, we're breaking that barrier down."

"It's hard to overstate how important audio has become in recent years -- or just how much bigger it's going to get in an AirPods-first world," said Peter Rojas, Partner at Betaworks Ventures. "Synthetic voice is going to be key to all this by transforming how audio is created. Demand for localized and translated spoken word content, whether it's in the form of podcasts or audiobooks, is exploding, and AI-based tools like Localize are the way to satisfy that demand."

How it works:

An audio recording is translated so that it accurately reflects not only the words and their specific meanings, but also colloquialisms and grammatical structures that are particular to that region and language. The service allows anyone to translate and hear it simultaneously - so you can find out immediately "how does Will Smith sound when he says this sentence?" Other examples include:

• Gaming - Resemble.ai makes it possible for a character's voice to sound the same in different languages. It retains the characters' voices for a different language.
• Influencers - YouTubers and TikTok stars can reach non-English speaking audiences and vice versa
• Call Centers, which serve scores of markets, can create one universal conversational chatbot with similar voices in numerous languages. Previously they had to hire talent internationally.
• Movies can expand internationally faster with the voices they originally intended because dubbing isn't needed.

About Resemble AI

Resemble AI, an artificial intelligence company that creates synthetic voices, is creating synthetic speech that is focused on human emotion. With state of the art Artificial Intelligence, its products Resemble Clone and Resemble Localize are designed to help entertainment, gaming, call center and other industry creatives synthetically produce, clone and dub high-quality voices. Founded in 2019, Resemble AI is headquartered in Toronto, Ontario and has secured $2M in funding from Craft Ventures, firstminute Capital, AET Fund, and Betaworks. To learn more, please visit https://www.resemble.ai.

SOURCE Resemble AI

Original source