V2.2 UPDATE - Hotfix
Small Hotfix after some update broke the way I had the nodes working. It's fixed for I2V now.
V2.0 UPDATE
New version naming. New nodes for better I2V generations (previous one had cases where it would create a bad first frame), Chunking node for VRAM optimization, new node in first pass for better micro-details. New nodes in workflow to install via ComfyUI Manager: https://github.com/yichengup/ComfyUI-YCNodes-MiniMax-H3
***- PLEASE leave some kind of feedback after trying -
Base Information
Now actually the fastest HQ audio + video generation workflow. See for yourself in the examples. I can currently create these 8s, 1 megapixel videos in 3.5 minutes which is INSANE if you ask me, while keeping good audio!
I'm still experimenting with different scenarios, tweaking for the ultimate best balance of quality vs speed. There might be more updates very soon.
WHY 2 MODELS AND REFINEMENT AT ALL??
Yes, valid question. My answer: creating high resolution videos with the Fast H3 model takes much longer than this workflow, and the quality of the short Taomate turbo workflow is outright bad with low steps. The Solution: We use the power of the audio quality and base movement of the Fast H3 model while then using the ultra-rapid speed of the Taomate 3 step turbo lora to refine and enhance the video to High Quality!
PLEASE let me know how this works for you!
This is my very first attempt at making a FAST workflow that runs on my 16gb + 32gb System RAM Setup and I'm looking forward to see if you guys have the same results like me.
Please try it out and let me know what you think!
INSTRUCTIONS:
Use the green nodes to configure your resolution, aspect ratio & duration
Keep the aspect ratio the same for the first pass and second pass resolution selector
Add your desired LoRAs in the purple node
Enable or Bypass your desired 1st Frame or Last Frame input Image Groups on the left side of the workflow
Adjust to your desired VAE's, text encoders that you might already have (I'm using a new KJ video vae that is smaller)
Don't forget to adjust the filename and path in the Save Video node to your prefered syntax and location (the one in the workflow creates this type of filename syntax with MP, date and time of generation: 1MP_Video_210926-150209_24fps_00001_.mp4)
MODELS
Base FLF2VA / T2V model: https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors
Text Encoder for 50xx Series Nvidia GPUs: https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors
Text Encoder for older GPUs: https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/text_encoders/qwen3vl_32b_minimax_h3_int8_convrot.safetensors
Audio VAE: https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/vae/minimax_h3_audio_vae_fp32.safetensors
Taomate 3 Step Turbo Lora: https://huggingface.co/drbaph/MiniMax-H3-Turbo-Lora-ComfyUI/blob/main/experimental/minimax_h3_taomate_fl2va_3step_ema_comfyui.safetensors
H3 Latent Upscaler: https://github.com/LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler
Custom Nodes Links
GENERATION SPEED:
On 5070Ti with 32gb System RAM:
8 seconds - 1 MP resolution: ~3:30 minutes
8 seconds - 0.5 MP resolution: ~1:45 minutes
10 seconds - 1 MP resolution: ~4:30 minutes
10 seconds - 0.7 MP resolution: ~3:10 minutes
Description
Completely reworked 2nd pass to actually good refinement and quality
FAQ
Comments (26)
Please share your experience/results with me, so I know if I should keep sharing these or not.
I'm 4060ti 16g + 32 g 2400 MT. I'm downloading and give it to try
@lilifirenotalone542 looking forward to hear your experience with it!
@zkyt sended
I would love to hear how this runs on systems from you guys with more or less RAM available, aswell!
Please let me know of any issues you encounter!
I'm really looking for some feedback here :D
Thanks this looks promising. I will try and report back.
i can report that.....no links for the required nodes and model.
@Crezjcm I updated the description with this info, thank you. Let me know if there is something missing still.
@zkyt still not exactly the sol-att node that is in your workflow, but thank you for the links. i 'll add another new comment about the experience.
@Crezjcm I'm so sorry! You are totally right, here is the one from the workflow. I will also add more links to the rest:
https://github.com/kijai/ComfyUI-SolAttn_triton
@Crezjcm I have updated the description with everything I could think of. Please let me know if there is still something missing. I'm still learning :D
@zkyt no problem, maybe it's even better to update the workflow with latest node versions and share it again. the best way to troubleshoot reports/problems/comments anybody will share, is to first have the exact same workflow ;)
@zkyt it's still not the exact SollAttnMiniMax node that is in the workflow, i tried to find it with no luck, anyway i'm trying to find the reason for the ripple effect right now.
@Crezjcm You just made me realize something.. The node may be completely redundant now because Kijai is now calling this node deprecated. I just started testing for a better solution. I will update the workflow once it is working. Thanks for the hint!
Good workflow, much better than the 0.5 one-pass. But I cannot find the SolAttnMiniMax node. and ComfyUI Minimax H3 Latent Upscaler is Comfyui_Minimax_h3_latent_Upscaler-Plus
@vrykol Thank you! Trythis for the Sol-Attn node:
i saw a laconic comment somewhere else about using TaoMate for the upscaling pass and this just provided the "how to do it", it'spretty simple and well layed out. It works fine and fast, good job. I just have a little issue, the first 2-3 frames of the video have a "ripple distorsion" effect. i did not take time to try differetn settings to find the culprit though. Thank you fo the workflow.
Thank you very much @Crezjcm ! I did no have this issue myself to be honest. I will see if I can find something about it. Otherwise I can only recommend to keep ComfyUI and your nodes updated, as this stuff advances really fast.
@zkyt I found the culprit, it was a "node" between the keyboard and the desk chair....as 99% of IT problems....
I immediately tried your workflow as I2V, adding an image as first frame, causes the ripple effect on the first 2-3 fames of the video.
For T2V, which is the purpose of this wf, there is absolutely no problem with the generated video.
for info, with everything up to date, fresh comfy start, rtx4090+64 ram, --fast-disk argument, original wf settings (with solattnminimax node bypassed) 8steps 0.2Mp + 3 steps 1.0Mp for a 8 seconds video output:
first run T2V 137 seconds. (2mn17)
second run I2V 121 seconds. (2mn01)
@Crezjcm I totally know what that feels like :D Thanks for sharing! Also, I just uploaded a new version after you brought my attention to the Sol-Attn node. That node seemed to be deprecated and didn't do anything for the generations, so I replaced it with the Comfy Kitchen Attention node.
Also thanks for sharing those numbers, is this noticeably faster than your previous workflows?
@zkyt it's pretty fast yes, i don't really compare to other workflows because i exclusively do I2V, i change too much settings at the same time (which is stupid), models/loras, resolution,1 pass, 2 passes, etc.... i'm hasty and don't settle, but i can say that your wf works as it is without spending xx minutes changing stupid colored node or...looking for nodes ;)
@Crezjcm that' good to hear :D I will now try to see if I can make it work for I2V aswell. If I can get it to work, I will update the workflow with a new version.
@Crezjcm took me a minute of research but I got it working. Check out the new version supporting First and Last frame inputs aswell now. Please let me know if you encounter any issues!
@zkyt will do, and i will certainly switch to "power lora loader v2 (qq)" from https://github.com/siraxe/ComfyUI-SA-Nodes-QQ and i think you will do too ;) taking advantage of the high and low model I/O and right click on the dot at the right of each lora to choose where do you want the lora to apply. it's my favorite lora loader "since" Wan 2.2
@Crezjcm thanks for the hint, I'm checking it out right now :)