lodestones

/

Chroma1-Radiance

like 114

What has changed? The version of ComfyUI is still the same v0.3.59

What has changed? The version of ComfyUI is still the same v0.3.59

https://github.com/comfyanonymous/ComfyUI/commit/c1297f4eb38a63e2f99c9fa76e32e3a36c933b85

thanks, I'll try to reinstall comfy, maybe I broke something.

Help.

I am only getting noise, ComfyUI version: 0.3.59 ComfyUI frontend version: 1.26.11, pulled from the latest master git
Any tips?

Did you change the weight_dtype in the model loader ? if i change it from default to fp8 it makes noise.

Did you change the weight_dtype in the model loader ? if i change it from default to fp8 it makes noise.

Thanks for the tip!. I was launching with "--fp8_e5m2-unet ". After removing that argument the noise is gone and the images came.

Radiance support has reached the ComfyUI release version today, v3.60.

https://blog.comfy.org/p/humo-and-chroma1-radiance-support

Any turbo lora support instead of 30 steps?

Unfortunately, I'm not sure there's a viable way to run this on anything less than 16GB VRAM...

I've tried running the FP8 and various GGUF versions, as well as versions of the new x0 model... Using the Comfy template (30 steps), I can sometimes power through the 98-99% VRAM usage on my 12GB 5070 to generate an image, but that image still looks undeveloped with significant undercooking of hands, faces etc. It's a shame, as I have been a big fan of Chroma since the start, but I just don't seem to be able to get any decent results from Radiance or Chroma1-HD and its many variant models unless I use acceleration LoRAs which then impart unwanted styles and compromises to the image and seem to negatively affect prompt adherence...