comfyanonymous comments

Results 503 comments of


                                            comfyanonymous

Flux Text encoder loras don't work, results are either blurry or the clip is not loaded

any example loras?

Almost unusably-slow - nodes fail to cache after recent update. ComfyUI now redundantly reprocesses unchanged inputs at every queue (IPAdapter reloads its model, ControlNet reloads depth map image, etc.)

Can you post the actual workflow?

Almost unusably-slow - nodes fail to cache after recent update. ComfyUI now redundantly reprocesses unchanged inputs at every queue (IPAdapter reloads its model, ControlNet reloads depth map image, etc.)

if you replace that Zoe-DepthMapPreprocessor with the "Canny" node does it have the same behaviour?

Almost unusably-slow - nodes fail to cache after recent update. ComfyUI now redundantly reprocesses unchanged inputs at every queue (IPAdapter reloads its model, ControlNet reloads depth map image, etc.)

Can you reproduce it on a simple workflow with core only nodes?

Absurd memory allocation issues runs fine on RTX 3060 crashes and errors out on 4090 on lower settings/similar settings (FLUX)

If you try without that custom node does it work?

ComfyUI cascade stage b implementation introduces artifacts

The correct sampler for this if I remember correctly is "euler_ancestral" so if you use that one you should get good results.

Flux SimpleTuner Lora "lora key not loaded"

I already support the transformer.diffusers_key_name format But this is: transformer.base_model.model.diffusers_key_name format At this point it's up to the training tool to output loras in a consistent format.

Flux bnb_nf4 'ForgeParams4bit' object has no attribute 'quant_storage'

Yes the right place to report issues is the custom node repo: https://github.com/comfyanonymous/ComfyUI_bitsandbytes_NF4

FLUX Issue | MPS framework doesn't support float64

https://github.com/comfyanonymous/ComfyUI/commit/48eb1399c02bdae7e14b2208c448b69b382d0090 can you check if this fixes it.

OOM when torch complie added into wf

For torch compile only the latest pytorch version is supported.