For more Workflows and updates, Join my Discord:
https://discord.gg/ZVWVhT43GW
You can deploy MiniMax on RunPod with my template: https://get.runpod.io/minimax-template
Workflow:
Minimax supports up to 18 inputs at once, wiring and bypassing nodes is a pain.
So I created this custom node that allows you to add/remove references for Minimax ReferenceToVideo, you only have to wire it once.
Some more neat features:
- Automatic prompt writing via OpenRouter/Local LLMs, returns structured Minimax prompts based on your description (opt in)
- Save prompt/reference packs and reuse them.
Custom node:
Nodes:
https://github.com/Hearmeman24/ComfyUI-MiniMaxRefPack
This workflow also come preinstalled on my Runpod template.
Description
FAQ
Comments (34)
Awesome will give it a shot...I was just about to vibecode a similar node today.... glad you beat me to it!!
I beat you to it, so you can beat your meat to it!
Cheers!
Hi just got the wf. How can I resolve
Unknown pack: MinimaxH3referencepack andand taeh3.safetensors shows that it needs to be put in tinyvae folder but there no such folder.You need custom nodes, linked in this post.
@HearmemanAI hey thanks, got it working but it asks openrouter key. how can i avoid that?
@Apprehensive6311 set openrouter to false
@HearmemanAI thank you so much. got it working but somewhow it still retains the face in video and just brings the dress from image. Any strength to be tweaked?
@Apprehensive6311 it looks like theres a bug in the prompt. But honestly after trying of like 30 mins it's not so simple lol. The OPs solution of specifying subject 2 as the clothing and movement itself seems like a decent solution, and maybe if I incorporated it mine would work. But I still tried; I can get the face to swap pretty easily, but'll often either forget the clothing or the body. I can't quite get it. This is my attempt:
"Use the first 00:05 seconds of <Video 1> as the source video for the target video edit. Replace the character <Subject 1> from the video <Video 1> with the character <Subject 2> from the photo <Photo 1>. Make sure the resulting character in the video is <Subject 2> but dressed in the same clothes and moving exactly as <Subject 1> moved in the original <Video 1>. Adjust the <Subject 1> clothing accordingly to fit on the character <Subject 2>. Use <Audio 1> as the original video reference sample, and use <Audio 2> is a reference sample for the voice of the newly dressed <Subject 2> in the edited video"
The easiest way would be to use F2K to take the first frame of the video and the picture and then make clothing swap there. Cause then you could cleanly add the person to the video.
Nice, but can openrouter models be replaced to other API, or local ollama models?
Added support for local LLMs in V2.
Why niot local LLM ?, not everyone want use paid services
Instead of complaining,
feel free to open a PR and I'll merge it.
Added support for local LLMs in V2.
OpenRouter is the worst node you could ever install.
I want to add my own prompts,Gemini sucks
Set openrouter to false and add them in the prompt box
@HearmemanAI but there is no prompt input only a generated prompt output
Added support for local LLMs in V2.
whats a better model for this WF?
Brilliant work! Is it possible to create an offline version? One with LLM and one completely without, so that users have a choice?Added support for local LLMs in V2.
Can you add openrouter/free?
Added support for local LLMs in V2.
pra mim aqui funcionou perfeitamente, selecionei o OpenRouter BETA fui no site deles e gerei uma Key, e pronto, rodou normal e gera muito rapido, so precisa de uns ajustes ali mesmo mas é normal, meu parabens!
The value for MiniMax References Manager's max_reference_edge couldn't be converted to INT.
set it to 2048
@HearmemanAI
did not work, same error.
I'm using the pruned ref2va int8 convrot model
@drfaker911219 I get the same error. I cant find any fix.
@drfaker911219 where did you set it?
@HearmemanAI ok corrected it. set 2048 at max reference edge.
Shitty openrouter needs money even though I'm using free model, nemotron
# ComfyUI Error Report ## Error Details - Node ID: 185 - Node Type: MiniMaxH3ReferencePack - Exception Type: ValueError - Exception Message: ValueError: prompt generation failed: OpenRouter returned 402: This request requires at least $1.00 in balance for video
Did anyone already mention something with local models? 😏
good stuff HearmemanAI 😉
Proprietary workflow. I advise you find another one that doesnt corral you into authors ecosystem.
Added support for local LLMs in V2.