Minimax H3 ref2va + fl2va Workflow
Simple toggle between ref2va and fl2va
Uses stock Minimax H3 sampler settings with Spectrum for a ~30-40% speed boost
I have yet to find a single other combination that doesn't have some kind of worse downside - The turbo LoRAs suck.
Motion Context video extension using save and load latents
How to use in workflow
Can lead to gamma shift after extending - may need to manually adjust brightness/contrast
FaceRefine for fixing distant garbled faces
required custom_nodes that aren't on the ComfyUI Manager included in archive
Simple postproduction settings
Interpolation using https://github.com/ethanfel/ComfyUI-Tween
GIMM VFI
SPEED VFI (seems better than RIFE at about the same speed)
Upscale
Tensorrt frame by frame upscale
RTX VSR
RTX Denoise/Deblur
Postproduction looping tools for fl2va using the same first and last frame
clip still end or starting frames + audio
Smoothen out the loop with rife generated frames using Seamless Loop
Generation Preview
Batch or single video upscale using GIMM VFI+Tensorrt as toggle below the main workflow
throw your videos into any folder inside ComfyUI/input/... and toggle the batch upscale to get everything running
No workflow will help, if your prompt doesn't follow the guidelines. I recommend looking through the following prompting guides if you have no idea about Minimax H3 prompt and then use other's or my prompts as help to get started.
I added the full workflows, including prompt and resources I used in the example videos, to the workflow archive.
https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_base_en.md
https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_ref_en.md
Resources
https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/vae/minimax_h3_audio_vae_fp32.safetensors
for low VRAM/RAM users
Thanks to
Description
FAQ
Comments (10)
If you have any questions just ask.
This is mostly just a WF made for myself, uploaded it with some examples.
MiniMaxH3NativeAudioLock node is missing. Comfy 0.37.0. All packs from resource links is installed. =(
@PerfectLoop It's included in the workflow zipfile :). It's part of the facerefine and the author didn't include it in his github but it's hidden inside a folder of another git so I included it to make it easier. You can find it inside the custom_nodes folder here otherwise if you want to download it directly: https://github.com/Shrek3OnVH5/MiniMax-H3-NativeAudio-MusicVideo-Workflow/tree/master/custom_nodes
I do like me some pink headed catgirls. <3
Before i test this H3_Face_Refine_Auto_Select.json workflow i just want to know something: Will this work on my 5060 16gb GPU and 32 Gb system RAM?
16GB VRAM is enough but you may have to drop down to a w4a8 model and nvfp4 or any smaller clip to not run out of RAM. I personally am using a 5080 and 96GB RAM. RAM is just as if not more important than VRAM since ComfyUI can do offloading very well.
I also recommend increasing the size of your pagefile/swap so your system doesn't crash if it does overflow your RAM.
Judging by the sample video, the workflow is excellent, and I would really like to use it.
However, it is massive and complex, and there are several nodes that do not function correctly with ComfyUI version V0.36.0.
I have been unable to resolve this issue.
I personally used it on v0.36, I could try to help you if you tell me what the issue is.
Most important thing is to not miss the "custom_nodes" inside the zip file, you have to manually copy them into your ComfyUI/custom_nodes folder since they can't be automatically downloaded using ComfyUI Manager, they are part of the FaceRefine custom nodes but the author didn't include them and they are hidden inside a folder of another git...
Everything else should be resolved by just downloading the few necessary custom_nodes via ComfyUI Manager.
It's definitely not a beginner workflow though since I made it mostly for myself.
Added a "Resources" version which I'll use as my resource dump for my H3 videos - this is the easiest way to share it. I'll update the file every now and then with the resources of some of my videos on request or whenever I feel like it.
Too bad there's no way to attach archives to video posts directly.