A workflow uses the latest Open Source MiniMax H3 to generate videos from multiple references.
Other model details are within the workflow.
Try the workflow online(Faster):
Eros 8Step LoRA:
https://www.runninghub.ai/post/2097621331332780033
Dual Sampler:
https://www.runninghub.ai/post/2086790597210746881
Latent Upscaler:
https://www.runninghub.ai/post/2089678677750923266
Basic:
https://www.runninghub.ai/post/2084116916972908545
Please check the video tutorial before you use this workflow:
Eros 8step lora + authenticity skill: https://rumble.com/v7fb9yi-authenticity-gpt-image2.5-minimax-h3-eros-lora-8steps20steps.html
Basic: https://rumble.com/v7dotpw-seedance-is-dead-minimax-h3-rebornfree-open-source.html
Dual Sampler: https://rumble.com/v7e3llc-free-open-source-gods-i-make-the-minimax-h3-recreate-a-classic-60-second-sh.html
Latent Upscaler+Local Prompt: https://rumble.com/v7een90-minimax-h3-latent-upscaler-local-prompt-skill-nodeqwen3.8-gguf.html
Custom Nodes Used In Workflow:
Custom Nodes:
https://github.com/T8mars/comfyui-minimax-h3-audio-T8
https://github.com/ptmaster/ComfyUI-PT_H3ConcatAVLatent
https://github.com/WhatDreamsCost/WhatDreamsCost-ComfyUI
https://github.com/smthemex/ComfyUI_UniBlockSwap
https://github.com/chflame163/ComfyUI_Qwen_H3_Prompt
https://github.com/LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler
Skill:
https://github.com/landon2022/minimax-h3-video-prompt
Description
FAQ
Comments (6)
I swear when I drag the json file into comfyui it just populates like a text. It doesn't deploy the workflow like normal
I tried, it just works as any other workflow file.
Try dragging the file directly over the center grid canvas of ComfyUI.
Alternatively, bypass drag-and-drop entirely: click the Load button on the ComfyUI menu panel and select your .json file manually through your file explorer.
sorry for stupid question, are u unziping the file u download?
@SNEHCORE I just got it and noticed it was a ZIP, cool and all, but, I saw uncompressed it's barely 45KB. It would be either useful to upload it as a *.json anyways, in case more users face the same issues. XD
I got it working just fine, btw. I was just wondering since I'm somewhat new in H3, how many character references it allows and how does they should be, in your Wukong example they are (Front/Side/Back) but what if I wanted highly detailed hands, feet, horns, eyes, etc. And if more than one character perhaps...
Thanks for the WF tho, it worked flawlessly.
@ad_Marzzel I managed now 6 character references (face, full front clothed, full back clothed, naked breasts, naked ass, naked pussy, I think I need a side body view too) + 1 background + 1 other character full view, everything works as planned, especially at close-ups (well, genitals sometimes buggy, but it's a miracle that we can add them trough ref2va even if model doesn't do them well).
9 reference images max that "H3 Reference to Video" allows. Seems to be they all maintained if described properly.