MiniMax H3 · Character Motion Transfer

ABOUT THE WORKFLOW

Transfer Motion Onto a Character
Upload a character image and a driving video. Write a prompt describing the scene. The model maps the motion from the driving video onto the character, preserving the character's face, outfit, and proportions while following the driving clip's movement beat for beat. You get a video with synchronized stereo audio, plus a side-by-side comparison of the driving video and the output.

Model

• MiniMax H3 by MiniMax (also known as Hailuo 3.0). An open-weight omni-modal video model that generates video with synchronized stereo audio from images, text, or reference video. Strong at preserving character identity across motion transfer.

• Turbo LoRA (optional). Cuts generation from 20 steps to 4 for faster output at slightly lower quality.

HOW IT WORKS

Step 1. Upload a character image
The person or character whose appearance you want in the output video. A clear, well-lit photo with visible face and full outfit works best.
Works great with: portraits · full-body shots · illustrated characters · cosplay photos

Step 2. Upload a driving video
The video whose motion the character will follow. Clip length sets the output length, up to 15 seconds at 24 fps (360 frames max).
Works great with: dance clips · walk cycles · gestures · action sequences

Step 3. Write a prompt
Describe the scene, the motion, and how the character should appear. Be specific about actions and camera angles. "Transfer the dance onto the character in a softly lit studio, steady camera" is better than "make them dance."

Step 4. Hit run and download
The model generates a video with stereo audio, plus a side-by-side comparison so you can check the motion transfer against the original driving clip.
Ready for: Premiere Pro · DaVinci Resolve · After Effects · any video editor

First time? Leave every setting as-is. The defaults (1280×720 · 20 steps · Turbo off) are the right starting point for almost everyone.

RECOMMENDED SETTINGS

Quick-start guide. Find the goal that matches yours and copy the settings.

• Standard motion transfer (most people) — 1280×720 · 20 steps · Turbo off · random seed. The right starting point for almost everyone.

• Quick preview or test — Turn Turbo LoRA on. Cuts steps from 20 to 4. Faster output, slightly softer quality. Good for checking whether the motion lands before committing to a full run.

• Higher resolution output — Raise width and height (must be multiples of 32). More pixels, longer generation time. Match the aspect ratio to your driving video for clean results.

• Portrait / vertical video — Set width to 720, height to 1280 (or similar vertical ratio). Match the orientation of your driving clip.

• Reproduce a result you liked — Lock the seed to the value from that run. Same inputs and seed give the same output.

• Motion is not matching the driving clip — Keep the prompt focused on describing what moves and how. Mention the driving video's camera framing. Avoid adding unrelated scene details that pull the model away from the reference motion.

Prompt: Describe the motion, the scene, and camera framing. Reference the character and driving video by their placeholder names (Subject 1, Picture 1, Video 1) if you want precise control. "Transfer the walking motion onto the character in a clean studio, steady camera following the framing of the driving clip" is better than "person walking."

LEARN

📹 Videos

• Intro to Floyo

• ComfyUI 101 Free Course ft. Sebastian Kamph

• Floyo 101 for Team Collaboration

✨ Quick links

• Help docs

• LoRA Training guide

• Discover more workflows

USE CASES

🎬 Content Creators
Put yourself or a character into a dance, a sketch, or a trending clip without reshooting. Upload a photo, drop in the reference video, and get a motion-matched result with audio.

🎮 Game & VFX Artists
Generate character animation tests from a single reference image. Block out motion before committing to a full mocap or keyframe pass.

🎨 Concept Artists & Previsualization
Animate a character design to pitch movement and staging. Turn a static concept into a moving scene to show a director or client.

🥽 Virtual Production & Digital Doubles
Transfer a real performance onto an illustrated or CG character for previsualization, social content, or experimental short films.

WHAT WORKS BEST / WHAT TO AVOID

✅ Works great

• Clear, well-lit character photos with visible face and outfit

• Driving videos with distinct, readable motion (dance, walk, gesture)

• Matching aspect ratio between driving video and output resolution

• Prompts that describe motion, scene, and camera framing specifically

⚠️ May produce softer results

• Low-resolution or blurry character images

• Driving videos with fast cuts or camera shake

• Mismatched aspect ratios between driving video and output

• Overly long or vague prompts that conflict with the driving motion

FAQ

What is MiniMax H3 and what can it do?
MiniMax H3 (also known as Hailuo 3.0) is an omni-modal video generation model by MiniMax. It generates video with synchronized stereo audio from text, images, or reference video. This workflow uses its reference-to-video pipeline to transfer motion from a driving clip onto a character image.

How does character motion transfer work with MiniMax H3?
You upload a character image and a driving video. The model reads the motion, timing, and camera framing from the driving video and applies it to the character from your image, preserving the character's face, hair, outfit, and proportions. The output is a new video where the character performs the driving clip's movements with AI-generated stereo audio.

Does MiniMax H3 generate audio with the video?
Yes. MiniMax H3 generates native stereo audio synchronized to the video. The audio is produced by the model, not extracted from the driving video. Expect ambient room tone and movement sounds rather than speech or music.

What resolution and length does MiniMax H3 support?
Up to 2K resolution at 24 fps, with video duration between 4 and 15 seconds. The output length is set by the driving video's frame count, capped at 360 frames. Width and height must be multiples of 32.

What does the Turbo LoRA do and when should I use it?
The Turbo LoRA is a distilled version of the model that cuts generation from 20 steps to 4. It produces results faster at slightly lower quality. Use it for previews and tests. Turn it off for final output.

Is MiniMax H3 open source and can I use the output commercially?
MiniMax H3 is released under the MiniMax H3 Community License with open weights. Review the license terms for your specific use case before shipping commercially. All models in this workflow run locally through native ComfyUI nodes with no API keys or per-run fees.

How to run MiniMax H3 motion transfer online?
You can run MiniMax H3 motion transfer online through Floyo. No installation, no setup, no API key to wire up. Open the workflow in your browser, upload your inputs, and hit run. Free to try.

WHY FLOYO?

Floyo is the only platform with team collaboration for ComfyUI in the browser. You run workflows with no install. You share run history, assets, and models across your team. You pay only when you generate. Floyo supports open-source and closed-source models.

A designer runs an edit and likes the result. A teammate opens that exact run from shared history and keeps going. No file handoffs. No version confusion.

For studios and enterprise teams, Floyo adds private workspaces, pooled resources, and a team usage dashboard. Other ComfyUI cloud tools run for one person at a time. Floyo runs for the whole team, with transparent per-generation costs.

Ready to try it?
Upload a character image, drop in a driving video, and run it. The prompt and settings are already set.

→ Launch Workflow, Free

Questions? Watch the free course or check the FAQ above.