DreamActor-M1 From ByteDance
ByteDance · Video
DreamActor-M1 is ByteDance's AI human animation model. It turns one reference photo into a moving video by copying the motion of a driving clip.
About DreamActor-M1 From ByteDance
What Is DreamActor-M1
DreamActor-M1 is a human image animation model from ByteDance Intelligent Creation, the research group behind the company's generative media work. It takes a single reference image of a person and a driving video, then rebuilds that person moving the way the video does. The output keeps the face, the body and the overall identity intact while changing the pose, expression and camera angle.
The model targets a gap that earlier animation tools left open. Most systems handle either a close-up portrait or a full body, and they drift off-model once a limb leaves the frame or the clip runs long. DreamActor-M1 claims to cover the whole range, from a tight head-and-shoulders shot to a full-body view, and to hold consistency across longer videos. It's a research release, published as a paper and a project page rather than a consumer app.
That distinction matters if you're shopping for a tool. There's no signup, no pricing page and no download link, so you can't run DreamActor-M1 yourself today. It's a look at where ByteDance's animation work is heading, and a set of results you can compare against the animation features inside products like Dreamina and CapCut.
Getting Started
- Open the official project page and watch the side-by-side demos, which show the driving clip next to the generated result
- Read the paper (arXiv:2504.01724) for the method, the training stages and the comparison tables against other models
- Note that no hosted demo, API or repository is published, so there's no public way to upload your own image and clip
- If you need human animation today, look at shipped ByteDance tools such as Dreamina and CapCut, or rival models with public access
Product Information
A quick look at DreamActor-M1 From ByteDance's pricing, supported platforms, and performance.
Best for
The users, tasks, and scenarios where this tool fits best.
Users
- AI researchers and students
- Video creators and animators
- Developers evaluating animation stacks
- Product teams at media and e-commerce firms
Tasks
- Turning a still portrait into a talking or moving clip
- Pose transfer
- Partial motion control
- Audio-driven facial animation
- Comparing animation quality
Scenarios
- Sourcing assets for a short film or ad
- Reviving archived photos
- Prototyping an avatar pipeline
- Teaching a class on generative video
- Evaluating a vendor
Key features
Hybrid Motion Guidance
The control system blends three signals: implicit facial representations, 3D head spheres, and 3D body skeletons. The head spheres track where the head sits and which way it faces, the skeletons handle body movement, and the facial encoding carries expression. According to ByteDance, this mix drives faces and bodies together instead of trading one for the other, which is the usual weak spot in skeleton-only animation.
Multi-Scale Adaptability
DreamActor-M1 is built to work at any framing, from a tight portrait to a full-body shot, without losing quality as the subject gets smaller in frame. ByteDance trained it in progressive stages on data at different resolutions and scales. That's the part that decides whether a single model can cover both a headshot and a dance clip, or whether you need to pick one.
Long-Term Temporal Coherence
For longer clips, the model pulls motion patterns from sequential frames and adds extra visual references, which does most of the work for body parts the reference image never shows. The project demos lean on this to keep a hand or a back looking stable as it swings in and out of view, the point where a lot of animation tools start to flicker or warp.
Partial Motion Transfer
You can move only part of the motion. The project shows results where only the face and head transfer while the body stays in its original pose, and other cases that adjust bone length so a taller or shorter subject animates without breaking shape. That's useful when you want a subtle expressiveness pass rather than a full-body reenactment.
Audio-Driven Facial Animation
Beyond video-driven motion, the model extends to speech. It generates lip-sync from audio in several languages, which points to dubbing and virtual-presenter work. The demo shows the mouth matching the audio while the rest of the performance stays consistent.
Diffusion Transformer Backbone
The engine is a diffusion transformer (DiT), the same class of architecture behind many current image and video models. A 3D VAE encodes the clip, a pose encoder handles the skeletons and head spheres, and a face motion encoder adds expression through cross-attention. The reference image runs through the same transformer weights as the noisy video, which is how identity stays put across the output.
Pros and cons
Pros
- Covers portrait, upper-body and full-body animation in one model, so you don't switch tools by shot size
- Holds identity and detail across longer clips, including parts the reference image doesn't show
- Supports partial transfer, so you can drive just the face and head when the full body isn't wanted
- Adds audio-driven lip-sync in multiple languages, which widens it beyond video-to-video use
- Backed by a published paper with method details and comparisons, so the claims are checkable
Cons
- No public demo, API or code, so you can't run it on your own image and clip today
- No pricing, signup or documentation for end users, which rules it out as a buy-now tool
- Judging quality means relying on the project's own demos and tables rather than your own tests
- Research output like this can stall before it reaches a product, so there's no timeline to plan against
Frequently asked questions
DreamActor-M1 is a human image animation model from ByteDance. Given one reference photo and a driving video, it produces a video of that person moving like the video, keeping their face and identity intact. It was presented at ICCV 2025 by the ByteDance Intelligent Creation team.
Related content
Explore related tools, skills, and articles for DreamActor-M1 From ByteDance.
DreamActor-M1 From ByteDance Alternatives
Griffin by Tavus
Tavus · Voice & Language · Video · ChatbotGriffin by Tavus is the first Human Interaction Model: a full-duplex, video-to-video system that watches, listens, and answers during a live video call in real time.
Syllaby AI Avatar 2.0
Syllaby · Video · MarketingSyllaby AI Avatar 2.0 turns a script into a presenter-led video with a reusable AI avatar, natural voice, B-roll and automatic subtitles, all in one browser workflow.
Vitra.ai
Vitra.ai · Voice & Language · Video · MarketingVitra.ai is an agentic content platform that creates, translates, reviews, and publishes video, images, documents, and app content in 120 languages from one shared brief.