AI model artifacts and deepfake detector directory
Technical breakdowns of the newest generative face-swapping, voice-cloning, and video-diffusion engines, with frame-by-frame artifact recognition.
Video
14 modelsMochi 1
Mochi 1 is an open-source video generation model from Genmo with 10 billion parameters. It produces high-motion video from text prompts, but exhibits distinctive fluid dynamics inconsistencies and unnatural physical inertia. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Seedance 2
ByteDance Seedance 2 (released in 2025–2026) is a breakthrough kinetic video generation engine specializing in dynamic character movement, acrobatic choreography, and viral dance performances. Due to its photorealistic rendering, it has become the #1 video generator exploited by romance scammers, catfishes, and spam farms on TikTok, Instagram Reels, and Facebook to generate fake "proof of life" videos and fabricated influencer clips. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Sora 2
OpenAI Sora 2 represents the frontier of foundation video models, using spatiotemporal transformer diffusion to generate high-definition video clips from prompt text. While cinematic and visually stunning, Sora 2 frequently fails complex physical interactions—liquids merging, hands deforming during object manipulation, and subtle background drift across camera pans. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Kling 1.5
Kwai Kling 1.5 is one of the most widely used generative video models in viral social media hoaxes, capable of producing 1080p video at 30 FPS. Scammers and content farms use Kling 1.5 to fabricate celebrity footage, fake disasters, and romance scam verification videos. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Google Veo
Google DeepMind Veo (Veo 2 and Veo 3) generates photorealistic 4K cinematic video with native synchronized audio and sophisticated camera control. Despite its cinematic realism, Veo exhibits telltale synthetic optical distortions, simulated bokeh anomalies, and audio-visual phase discrepancies. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Wan 2.1
Alibaba Wan 2.1 is an open-source 14B parameter foundation video model widely adopted by independent creators, open-source communities, and bad actors. Because the model weights are public, it can be run locally with zero content filters or watermarks. Sealed Rose exposes the unique 3D causal VAE latent artifacts and flow-matching trajectory anomalies inherent to Wan 2.1. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
MiniMax Hailuo
MiniMax Hailuo AI is known for producing expressive human characters with cinematic lighting, dynamic camera moves, and intricate facial gestures. It is heavily utilized in viral social media hoaxes, fake influencer channels, and romance scams. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Pika 2.0
Pika 2.0 introduced viral "Pikaffects" (melting, crushing, exploding, inflating objects) that took TikTok and Instagram by storm. Beyond surreal memes, Pika 2.0 is used to generate realistic short product clips, landscape footage, and fake breaking news scenes. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Luma Ray 2
Luma Ray 2 (Dream Machine) models video as unified 3D worlds, allowing realistic camera tracking, complex lighting, and dynamic action. It is commonly used for viral architectural teasers, fake historical footage, and product concepts. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Runway Gen-3
Runway Gen-3 Alpha and Gen-3 Turbo set a new benchmark for generative video with high fidelity and controllable camera trajectories. Because of its realism, Gen-3 is commonly used to produce deceptive social media clips, synthetic news footage, and impersonation videos. Sealed Rose extracts temporal optical flow across sampled frames to detect subtle motion warping, fluid dynamics violations, and synthetic camera acceleration. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Hunyuan Video
Tencent Hunyuan Video is a state-of-the-art open-weights video foundation model featuring an advanced 3D VAE architecture and 13B parameters. Because it is open-source, cybercriminals and scammers deploy Hunyuan Video locally to generate high-resolution deepfake clips and romance bait videos without cloud safety filters. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Higgsfield AI
Higgsfield AI specializes in hyper-dynamic cinematic camera movements, 360-degree pans, and character animation. Its cinematic controls are utilized by content creators and bad actors to produce convincing viral TikToks, fake action footage, and synthetic influencer travel clips. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Hedra Character
Hedra Character-1 allows anyone to generate expressive, talking human avatars from a single portrait and audio track. Because it models complex facial expressions, head tilts, and emotional micro-gestures, Hedra is widely adopted by scammers creating personalized video messages, romance catfish verification clips, and impersonation scams. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
CogVideoX
THUDM CogVideoX (developed by Tsinghua University and Zhipu AI) is an open-source text-to-video foundation model with 2B and 5B parameter variants. Featuring 3D VAE compression and expert transformer blocks, CogVideoX is widely used in local computing environments to generate uncensored video deepfakes and automated video feeds. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Image
10 modelsFlux.1
Flux.1 (developed by Black Forest Labs, founded by the original creators of Stable Diffusion) is an open-weights text-to-image architecture with 12 billion parameters and rectified flow transformers. Because Flux.1 renders realistic anatomy and skin textures, bad actors utilize it to fabricate catfish personas and synthetic photos. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Recraft v3
Recraft v3 is a state-of-the-art AI model tailored for professional graphic design, digital illustration, vector art, and logo generation. It is frequently used to fabricate fake corporate logos, counterfeit product packaging, and fake certification badges. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Midjourney
Midjourney is one of the most widely used text-to-image generators in the world. With version 6 and 7, Midjourney achieved cinematic realism, rendering natural skin textures, reflections, and intricate details that frequently deceive human observers. It is heavily utilized across social media, dating apps, and marketplace fraud to manufacture synthetic identities. Using our detection partners' models alongside our own checks, Sealed Rose looks for Midjourney’s distinctive spatial frequency profiles, lighting coherence inconsistencies, and latent diffusion noise residuals. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Stable Diffusion
Stable Diffusion (including SDXL and SD 3.5 Large) is the foundation of the open-source generative AI ecosystem. Open weights allow fraudsters to fine-tune custom checkpoints and LoRAs locally on consumer GPUs to generate targeted catfish personas, fake KYC credentials, and non-consensual imagery without guardrails. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Ideogram 2
Ideogram 2.0 is renowned for its ability to render accurate, coherent typography inside generative images, making it popular for fabricating fake posters, fraudulent invoices, receipts, and social media branding. Despite crisp lettering, Ideogram exhibits telltale diffusion artifacts in non-text regions and distinct edge anti-aliasing signatures. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Hunyuan-DiT
Tencent Hunyuan-DiT is an open-source diffusion transformer model tailored for bilingual text-to-image synthesis. Scammers deploy Hunyuan-DiT to generate convincing identity documents, synthetic marketing collateral, and photorealistic social media personas. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Qwen-Image
Alibaba’s Qwen-Image (developed within the open Qwen foundation model ecosystem) excels at photorealistic human portraiture and complex multilingual visual scenes. Fraud rings deploy Qwen-Image models to manufacture synthetic influencer profiles, romance scam personas, and e-commerce model photos. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Grok & Aurora
xAI Grok-2 incorporates image generation (powered by Black Forest Labs technology and xAI’s Aurora pipeline) available directly on X (Twitter). Due to minimal guardrails, Grok-2 is frequently weaponized to fabricate viral political deepfakes, celebrity hoaxes, and fake emergency news photos that spread rapidly across social networks. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
DALL-E 3
OpenAI DALL-E 3 and GPT-4o image generation are built directly into ChatGPT, allowing hundreds of millions of users to synthesize images from natural language. Scammers utilize ChatGPT to generate realistic profile avatars, fraudulent invoice receipts, and misleading product photos. While original OpenAI files include embedded C2PA provenance metadata, bad actors routinely screenshot or re-encode images to strip this metadata. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
SeaDream
ByteDance SeaDream is a high-performance image generation engine powering visual creation across TikTok and ByteDance creative apps. It generates realistic consumer photos, viral product showcases, and digital human likenesses. Scammers leverage SeaDream to manufacture deceptive TikTok profile pictures and synthetic viral marketing imagery. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Audio
4 modelsF5-TTS
F5-TTS is a state-of-the-art open-source text-to-speech model based on flow matching. With just 3 to 5 seconds of audio from a podcast, voicemail, or Instagram story, F5-TTS clones an individual’s voice with unprecedented fidelity, fueling grandparents-in-jail phone scams and CEO fraud. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
ElevenLabs
ElevenLabs v3 is a leading platform for expressive voice cloning, capable of synthesizing speech from short audio references with emotional inflection and conversational cadences. Synthetic voice clones are frequently weaponized in CEO fraud, grandparent emergency scams, and political robocalls. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Suno AI
Suno AI is a generative music diffusion engine capable of producing complete multi-instrumental songs with synthetic vocals from text prompts. Prolific creators upload AI-generated tracks to streaming platforms under virtual artist personas. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Udio
Udio emerged in 2024 as a high-fidelity generative music platform, praised for realistic instrument separation, expressive vocal synthesis, and complex genre styling. However, Udio tracks are frequently utilized to flood digital streaming services with synthetic music and deceptive AI covers. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Face-swap
4 modelsDeep-Live-Cam
Deep Live Cam went viral in 2024–2025 as an open-source tool enabling anyone to swap their face in real-time onto a live webcam stream using a single reference photo. It is heavily utilized by romance scammers, catfishes, and remote interview impostors on Zoom, Teams, and Omegle. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
LivePortrait
LivePortrait (developed by Kuaishou) generates lifelike animated talking videos from a single still portrait by transferring head movements, blinks, and emotional expressions from a driver video. It is widely used by catfish syndicates to turn stolen photos into "video proof" of their identity. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
Roop Unleashed
Roop Unleashed is a widely used desktop software application that automates face-swapping across full video clips without requiring dataset training. Attackers use it to generate non-consensual deepfakes and fabricate fake celebrity adult content. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.
FaceFusion
FaceFusion is the premier open-source face-swapping and face-enhancement framework. It combines high-resolution face swapping (Ghost, Inswapper) with neural restoration (CodeFormer, GFPGAN) and frame color matching. Sealed Rose scores the file with AI-detection models and returns a confidence score and the likely generator.