⚠️If you encounter a bug before commenting, try to redownload the workflow first as I am actively fixing bugs as they are reported. If it doesnt work let me know and I'll try to help.⚠️
⚙️ MiniMax H3 Prompt Enhancer — V7 Turn simple ideas into detailed, structured prompts for MiniMax video generation.
This ComfyUI workflow is a prompt enhancer built around the official MiniMax H3 prompting guide. It helps transform short, rough ideas into clearer and more effective video prompts — without any complex LLM installation.
Supports T2VA, I2VA, FL2VA, L2VA, and REF2V workflows.
List of models: https://huggingface.co/DavidAU/Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic-NEO-IMATRIX-MAX-MTP-GGUF
Choose the best for your hardware. *MTP models incompatible with the Custom LLM node*
Custom nodes needed (also in comfyui manager):
LLM Text Processor: ⚠️*Must be insalled manually from here, there is a bug for that node in the node manager*⚠️ https://github.com/KingManiya/ComfyUI-LLM-text-processor
Comfyui-RMBG: https://github.com/1038lab/ComfyUI-RMBG
rgthree-comfy: https://github.com/rgthree/rgthree-comfy
✨ What's New in V7
Added a video input node and a second QwenVL node (automatic download on first run for this model) with instructions on how to use.
Backed prompt refining.
Model Change. Now using Qwen3.5-9B-The-Defiant-Fable-Uncnr-Heretic-NEO-MAX-Q8_0. (Download directly in workflow. Extremely accurate, runs fast and easy on 16 VRAM)
🎯 The Goal
This workflow isn’t meant to blindly generate the “perfect” prompt for you. Instead, it gives you a well-structured starting point that you can quickly review, tweak, and customize before sending it to MiniMax.
Simple idea → Enhanced prompt → Your edits → Video generation
💡 Tip: The generated prompt is a starting point. Read through it and adjust the details, actions, camera movement, timing, and style to match your exact vision.
Feel free to use, modify, and integrate this workflow into your own projects. If you create something with it or improve the prompting method, share your results and prompting tips in the comments so the community can keep improving it!
If you use this workflow in your own workflow pack, a quick mention would be greatly appreciated. ❤️
Description
Added a short concept / prompt enhancer.
Added link for bigger model (better for REF2V but slower)
Added "Load Video UI" Node for cutting long video into smaller clips.
FAQ
Comments (45)
Amazing! Works great. You can improve it further by adding this node - https://github.com/phazei/ComfyUI-Prompt-Stash
It acts as a previewer, editor and prompt saver. And you can save prompts in different categories. As clip model I use Huihui-Qwen3-VL-4B-Instruct-abliterated-fp8_scaled with absolutely amazing results.
I'll look into that. Thanks for your comment and advice! ❤️
How should a ref model be write?
in a natural language, explain briefley what is what and describe the scene,
ex:
1-girl from picture 1 dancing, music from audio 1, movement from video 1,
A stunning anime girl performs a, fluid choreography in the center of the frame, her movements sharp and rhythmic. 10 second clip
2- Man from picture 1 in location from picture 2, voice tone from audio 1.
He pulls a pipe out of his pocket and says "that's what I needed"
How to add more reference images?
working on it, stay tuned for v3.1
In the meantime; just describe in more details what the other images are and edit the output prompt if needed.
ex: girl from picture 1 in the room from image 2, a dark candlelit room with a big window, moon visible through the window.
V4.0 out. Enjoy!
A-W-E-S-O-M-E
Glad you like it! 😊❤️
System Prompt starts with: Your task is to generate a perfect prompt optimized for the Minimax H3 model image to video. Use this guide to optimize a prompt inputed by the user, you must adhere strictly to this guide to output the final prompt. How few years old models like QWEN3 can know what MInIMAX 3 is? :D
well the guide is described aftewards so thats to make sure it doesn't try to output an other format or model prompt for example.
@KiraNugget fair enough, I did ask offline models like Gemma3.5 and qwen3 about minimax, they had vague knowledge about the name but not specifics.
@blhll if you have suggestion of models that run in comfyui load clip node for 16vram, that would be nice to know. I haven't realy played a lot with text gen before so my knowledge is still pretty basic; I use what I found was simple for this workflow but I know it can be improved.
@KiraNugget I just tried Qwen3.5-9B that was included in WF link, and it did not work At all, Generate Text node threw error, then I tried Qwen3VL-4B-abliterated and it actually followed System prompt structure pretty well, with proper dialogue formatting (putting [English] in brackets, etc. need do more testing with different ones that can be used in this manner.
@blhll 9b worked for me but is very slow, I find it more useful in the ref2v section becaus 4b sometimes mess the prompt structure with more complex prompts and multiple references
i need some help. whenever I enter my prompt, it finishes with a whole bunch of dots.
only dots or the prompt then dots? what model are you using?
https://huggingface.co/ahmed22xa/Huihui-Qwen3-VL-4B-Instruct-abliterated-comfy/blob/main/Huihui-Qwen3-VL-4B-Instruct-abliterated-fp8_scaled.safetensors this model was suggested in the comments maybe try that. I don't know what could be causing hese kind of issues. :(
@KiraNugget prompt dots
@Alkorr64 I realy don't know what sould be causing this... sorry ::(
@KiraNugget nvm, i figured it out
@Alkorr64 Same thing over here, how did you fix?
where i can download Qwen3.5-9B-abliterated.safetensors?
link is in the workflow, just click on it. :)
@KiraNugget forgive my ignorance, but do we need to also download all of the tokenizer, config files etc... in the repository listed on your workflow for the 9b model to work? When you click the link listed in the workflow, it just downloads a 'model.safetensors' file but that file just seems to throw errors when I attempt to use it in the workflow. The 4b model seems to work fine though.
@mjh02111964582 you only need the model file, rename it so you can find it easily and place it in models/text encoders. i'm working on V5 with a better model that should realease today if you can't make it work. stay tuned
I'm getting an error that it's missing the model minimax_h3_fl2v_lightx2v_turbo_4step_v0.1_comfy.safetensors , but that model is downloaded and in the loras folder. I assume it goes in the loras folder because the error message focuses on a turbo lora node that is lit up red. That node says "always on--don't touch", and there's no way to point it at that "missing" model. Fresh install with comfyui 0.31.0 .
This workflow only uses qwen3.5 models.... I have no idea where the error is coming from. Are you on the right workflow page?
@KiraNugget lol--i am not. i should take a nap. thanks for the fast response, though!
@matai515 I figured lol, enjoy your nap!
When i try it with Qwen3.5-9B-abliterated from you site it will only run when it no picture input only text can be use :/ how i can fix it ?
Likely an OOM issue. try the smaller model 4b. I Barely run it on 16 vram
@KiraNugget I have 32 Vram load 1-3gb in vram end then end :/ and it will not working with image
@tomookazaki87856 im working on v5 with a new smarter nad faster model with qwen3.6, should release today, stay tuned.
@KiraNugget i have try it all instal newest comfyui and nothing working :( the LLM dont work and the other what i write too not with images still the same
@tomookazaki87856 do you have comfyui installed with ComfyUi-Easy-Install program? I really don't know what could ba happenning I usually update everything with one click.. sorry :(
@KiraNugget Yes i have it with it Easy Install all new and it will not worked i have llm install all what i need nothing will working
@KiraNugget It will work now :) but no video anymore come it back in newer version ?
I tried using a version 4.0 workflow with comfyUI 3.32.0, but LoadVideoUI is unavailable. What should I do?
search for whatdreamscost in custom nodes manager and intall that pack, this is the github for manual install: https://github.com/WhatDreamsCost/WhatDreamsCost-ComfyUI
if it does not work you can tryy ani other load video node but you will have to cut/reseize your clips manually in any other way.
The node is now displayed correctly, thank you.
Question, works great when I use the --highvram augment but if not it fails on the generate text node, any way around this or nah?
try V5! 😉
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.