Tutorial inYouTube：
Register through my online workflow to receive 1,000 free RunningHub credits:
https://www.runninghub.ai/post/2099399709893853185/?inviteCode=rh-v1497
ComfyUI-Yusu-MiniMaxH3-Unified:
https://github.com/yusu-02/ComfyUI-Yusu-MiniMaxH3-Unified
RunningHub Canvas Version
https://rhtv.runninghub.ai/camp/view/2100066397379215361/?inviteCode=rh-v1497
Main Features
MiniMax H3 video generation
LightX2V 8-step accelerated workflow
Text-to-video generation
Image-to-video generation
Reference-to-video generation
Unified multi-purpose workflow layout
Optional cinematic LoRA
Simplified prompt and reference-image inputs
Adjustable duration and resolution
Suitable for product videos, AI UGC ads, character videos, and cinematic experiments
Cinematic LoRA
The workflow includes a cinematic LoRA stage that can help reduce some of the unnatural facial lighting often seen in AI-generated videos while adding a more grounded cinematic finish.
I normally use a relatively low LoRA strength of approximately 0.3–0.5. Higher values may change the original image or character appearance too aggressively, so I recommend starting low.
Recommended Workflow
Select the required generation mode.
Enter the video prompt.
Upload the reference images in the order specified by the prompt.
Confirm which reference is Image 1, Image 2, and so on.
Set the duration and output resolution.
Run a lower-resolution test first.
Review motion, character consistency, hand movement, and prompt logic.
Increase the resolution only after the test result is stable.
MiniMax H3 follows prompts closely. This is useful when the instructions are clear, but logical problems in a prompt may also be reproduced very literally. Complex hand movements and detailed physical interactions should be tested carefully.
Notes
This workflow does not remove the need to review prompts and test generated results. Different models have different strengths, and some shots may still require prompt revisions, rerolls, or editing.
The attached video demonstrates the workflow, real product-video examples, common generation errors, and the practical differences between MiniMax H3 and Seedance.