Attempt to build sage attention 2.2+ it speeds up MiniMax significantly.
Tested on rtx5070ti.
Diffusion model:
Text encoder:
VAEs:
Turbo LoRA:
Ollama LLM (gemma 4 12B it) for prompt helper:
You can add as many references as needed..
DOWN BELOW IS OLD VERSION CONTENT.
DISABLE comfyui_smznodes if you have this extension!!!
(It breaks any LTXV workflow.)
Diffuser model (pick one):
Text encoders (both):
VAEs:
Latent upscaler:
LoRA (distilled LoRA is needed, do not disable):
Enjoy!
Description
Ditched LTXV2.3 upscaler, now SEEDVR2 is my best friend.
FAQ
Comments (4)
My mistake, brother. I've only been learning comfyui for a short time. I just downloaded the wrong model. I'm very sorry. Your workflow is very powerful and practical.
Now if only there were stable Blackwell support for Sage Attention...
Try to build your own using qwen chat or https://www.youtube.com/watch?v=YaOMeo5467U
I wouldn't bother with Sage Attention 3 (yet) because the quality is not good.
But you can use v2.2.0 on Blackwell and it works fine on my 5080 RTX https://huggingface.co/ussoewwin/Sage-Attention-for-Windows/tree/main
Then in ComfyUI you just have to use a custom node like ComfyUI-KJNodes. and put this node in your workflow between "Load Diffusion Model (Model ->) Patch Sage Attention KJ node (<- Model) Basic Guider or Basic Scheduler.