Text-to-Image
Diffusers
stable-diffusion

Get black image when running on MacBookPro m3 16GB

#7
by foxonline - opened

In these days, I tried the SDXL, SDXL with refiner, SDXL Lighting, they all produce the black image. Also I tried the vae-fp16-fix, but no effect. Anyone has the solution?

That's usually because you are using the wrong VAE model, which can produce a black image or a very incorrectly colored image. Some need a specific VAE, some use the VAE from the model itself. There are SD and SDXL VAE models and they are not compatible with each other.

SD 1.5 model won't work with a SDXL VAE, and vice versa.

Also, the SDXL Inpainting model produces black images.

Sounds like you need to download the SDXL VAE (regardless of it you use it) or try using the model as the VAE. It's seems your using a VAE created for SD 1.4 or SD 1.5 or SD 2.0 or SD 2.1 which will not work with SDXL.

Try
https://huggingface.co/stabilityai/sdxl-vae/blob/main/sdxl_vae.safetensors

Or this FP16 version:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/blob/main/sdxl.vae.safetensors

This comment has been hidden

Actually, you need to UPDATE COMFYUI today.
They released the "x0" sampling method which is needed.
Also, you need to add a VAE Decode.
I attached a PNG image I made that has the workflow I used embedded for Lightning 1 step model.
ComfyUI_00069_.png

This is what the setup i modified looks like now: (this image of the layout doesn't have the workflow, use the picture of the lightning above for the workflow, or just modify yours)
SDXL-Lightning-1-step-layout.png

Sign up or log in to comment