MiniMax H3 · Reference to Multi-Shot Video
Upload a reference image and up to three guide images as timed keyframes, write a prompt describing the motion, and MiniMax H3 generates a multi-shot video with synchronized audio in one pass.
keyframe video
minimax h3
multi-shot
reference to video
310
ABOUT THE WORKFLOW
Generate a Multi-Shot Video from Reference Images
Upload a reference image to set the subject and opening frame. Add up to three guide images that lock keyframes at specific timestamps in the clip. Write a prompt describing the motion, camera work, cuts, and sound. The model generates video and stereo audio together in one pass.
Model
MiniMax H3 by MiniMax (also called Hailuo 3.0). An open-source video model strong at multi-shot sequences with consistent subject identity across cuts and native synchronized audio.
HOW IT WORKS
Step 1. Upload your reference image
This sets the subject, scene, and opening frame. The model carries the subject’s identity forward through the clip.
Works great with: portraits · fashion · product shots · characters
Step 2. Add guide images (optional)
Upload up to three more images. Each one locks a keyframe at a specific timestamp in the video. Set the timestamp for each guide image in the Frame Index fields.
Step 3. Write your prompt
Describe the motion, camera movement, lighting, shot cuts, and sound. Use structured prompts with shot tags for multi-shot sequences.
Step 4. Hit run and download
The model generates video and audio together, then combines them into one file. Output is 768p at 24 fps with stereo audio.
Ready for: Premiere · DaVinci Resolve · After Effects · any NLE
First time? Leave every setting as-is. The defaults (16:9, 5 seconds, Lightning LoRA 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 multi-shot video (most people) — 16:9 · 5 seconds · Lightning LoRA off · random seed. The right starting point for almost everyone.
Quick preview before a full render — Turn Lightning LoRA on. Cuts steps from 20 to 4 for a faster, rougher result you can evaluate before committing.
Longer sequence with more shots — Increase duration up to 15 seconds. Space your guide image timestamps across the clip to control each shot transition.
Consistent subject across cuts — Use guide images that share the same subject in different poses, outfits, or settings. The model carries identity from the reference image forward.
Video with specific sound design — Describe the audio in your prompt: music genre, tempo, Foley effects, ambient tone. The model generates synchronized stereo audio from the text.
Transitions between guide images look rough — Make sure guide images share similar lighting and color temperature. Reduce the number of guide images or space timestamps further apart.
Prompt: Structured prompts with shot tags work best. “The camera pushes in on the subject against a blue background, then cuts to a side angle with dramatic shadow” is clearer than “cool cinematic video.” Describe motion, camera, lighting, and sound explicitly for each shot.
LEARN
📹 Videos
ComfyUI 101 Free Course ft. Sebastian Kamph
Floyo 101 for Team Collaboration
✨ Quick links
USE CASES
🎬 Fashion and Editorial Film
Upload a model’s hero shot as the reference, add wardrobe or set changes as guide images, and get a multi-shot fashion film with synchronized music.
🎮 Game and VFX Pre-Vis
Block out a multi-shot sequence from concept art keyframes. Test timing, camera movement, and shot transitions before committing to a full CG pipeline.
🎵 Music Video Prototyping
Set up a sequence of visual moments as guide images, describe the rhythm and sound in the prompt, and generate a rough cut with synced audio to pitch or iterate on.
📖 Storyboard to Motion
Turn storyboard panels into a moving sequence. Each guide image becomes a shot, and the model fills in motion and transitions between them.
🎥 Short-Form Content
Generate 5 to 15 second clips with multiple cuts for social posts, ads, or teasers from a set of reference images.
WHAT WORKS BEST / WHAT TO AVOID
✅ Works great
Guide images that share a consistent subject
Structured prompts with shot-by-shot descriptions
Well-lit, clear reference images with a single subject
4 to 10 second clips with 2 to 4 shots
⚠️ May produce softer results
Guide images with wildly different subjects or styles
Vague prompts without shot structure or motion description
Clips longer than 15 seconds (needs a separate extend step)
Too many guide images crammed into a short duration
FAQ
What is MiniMax H3 and how is it different from other AI video models?
MiniMax H3, also called Hailuo 3.0, is an open-source video model by MiniMax. It generates video and stereo audio together in one pass, which most other video models do not do natively. It also supports multi-frame reference generation, so you can lock keyframes at specific timestamps and the model fills in motion between them while keeping subject identity consistent across cuts.
How do guide images and timestamps work?
Guide images act as visual anchors at specific points in the video. You upload up to three guide images and set a timestamp (in seconds) for each one. The model generates the opening frame from your reference image, hits each guide image at its timestamp, and fills in the motion between them. Default timestamps are 1.5s, 3s, and 5s.
Does MiniMax H3 generate audio with the video?
Yes. The model generates native 32 kHz stereo audio synchronized to the video in the same pass. Describe the sound you want in the prompt: music genre, tempo, Foley effects, ambient tone. The audio and video decode separately and combine into one file.
What does the Lightning LoRA do?
The Lightning LoRA cuts sampling steps from 20 to 4, which makes generation faster but produces rougher output. Use it for quick previews when you want to test a prompt or guide image arrangement before running a full-quality render.
What resolution and frame rate does MiniMax H3 output?
Output is 768p at 24 fps with stereo audio. Duration ranges from about 4 to 15 seconds. The aspect ratio defaults to 16:9 widescreen. Higher megapixel settings and longer durations increase generation time.
Is MiniMax H3 free to use commercially?
MiniMax H3 is released under the MiniMax Community License. Commercial use is allowed for organizations with annual revenue under $20 million. Organizations above that threshold need a separate commercial agreement with MiniMax.
How to run MiniMax H3 online?
You can run MiniMax H3 online through Floyo. No installation, no setup, no API key to wire up. Open the workflow in your browser, upload your reference and guide images, 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 director runs a multi-shot sequence and likes the result. A compositor 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 reference image, add your guide frames, describe the motion, and hit run.
Questions? Watch the free course or check the FAQ above.
Read more
