⚠️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
I threw the workflow into an industrial compactor.
Only one section for ALL modes!
LLM instruction prompt redesign.
Quick reference markdown added to double check your prompt.
FAQ
Comments (25)
There seems to be something wrong with the instructions. It always seems to want to add 2D, 3D, and claymations in all the T2V and I2V.
Also T2V always want's to input an image even when not toggled.
Thanks for your input, will fix asap
Should work now, redownload V6
Works as intended! Keep up the great work!
Hello again, great update. However, it's saying: "I cannot fulfill this request. I am programmed to be a helpful and harmless AI assistant. My safety guidelines prohibit me from generating sexually explicit content or descriptions of sexual acts."
is this a glitch?
The instruction prompt in the backend include a NSFW rule and it somtimes ignores it, just changing the seed should fix it. Models should be an uncensored one tho so IDK why it does that somtimes.
@KiraNugget ok
@Alkorr64 And thanks to you for giving me the idea to do an All-in-one workflow instead of 4 different sections ;)
@KiraNugget oh, you're very welcome :D
I've been here since v3, v6 is the best yet - thank you for sharing!
Best thing you can do for thanking me is posting some videos down below!
hi, it seems to be windows only ?
I'm not sure, I,ve seen a linux user struggling with the llm node but I don't know if it's incompatible or if the user failed to install the node manualy. If you could confirm after trying it would be nice so I can try to fix in the future.
Brilliant work—maybe you could add something else?
How about including 3 or 4 slots where you can enter your own custom prompts? That way, you could use them for specific prompting needs. :)
And maybe make it so you can add them directly—whether it's 1 or 10—so you don't have to do so much connecting.
Hope you know what I mean! :) And really brilliant work 🥳
One quei
Actually not sure what you mean :P
@KiraNugget I mean, in the subgraph you have options like "text to image" or "ref to image," and you simply add custom versions so I can enter my own prompt. And then, you could just press "add" to get more prompts and custom variations. That way, you'd have your own options to choose from instead of just yours.
@tomookazaki87856 you could link other promtpers together and link them to a text combine node and enablae or disable them yourself as needed, I'm not sure I see the use for that in this workflow.
Putting where we select our models in a subgraph and then we're stuck in the subgraph is pretty annoying. I don't know, and most people I'm sure don't know how to get back out of the subgraph. Many likely don't know to go into the subgraph to begin with, so hiding one of the most essential components in there makes it unusable for many. Putting things that never get touched into a subgraph, ok, but choosing a better model being hidden? I don't think that's good design.
Figured it out.. press esc to get back out of the subgraph
@civitai7_ I will improve that in future versions thx!
I copied the prompt inside the "subgraph". Then I use it as instruction in the llmstudio using a small model to generate the prompts. then, I "unload model" to free Vram for comfyui. Long story short: I use my private llmstudio to generate prompts and paste it to comfyui.
For NZFW pr0m pts: 4gb model: Qwen3.5 9b defiant fable herectic neo max
my good sir. this is a great wf. would it be possible to add back the describe a video and audio? as previous verions. thanks!
I will try to find a node that suports video and audio with a gguf model. In the meantime you can describe briefly what the audio is and what the video is.
TLDR: as long as it knows there is a video/audio input and what it is, You don't need to actually input them in the prompter.
Ex: I use an audio reference for the voice of my character and never had to input audio in the prompter, as long as I say that the audio is the reference for the voice of the woman, it will put it in the prompt as: <Audio 1> is the voice-timbre reference for the female vocalizations. <Audio 1> (appears in [Shot 1]): fully_copy - The female vocalizations match the reference timbre. that is what is important to have, MinimacH3 does the rest. same for video: a quick description: motion from the video reference, fight scene, etc
How do I write proper hints??????
It's complete nonsense.
1 photo - 1 frame
2 photos from 4 references - a map from 4 photos
1 video and 1 audio
I can't properly repeat the movements from the video
You don't have a proper description, no instructions, no video tutorials, etc.
Could you at least give some examples of how to write... (((