Output image quality is unstable

[PetitPrinceb612]

Hi, thank you for developing the comfyui-ELLA. I got it for the first time and compared it with the SD15 general process. I don’t know if it is a problem with my settings. I tested a lot of prompts and found that the Ella img rendering effect is very unstable and not good. And No matter how you adjust it, it is difficult to improve the img quality. Do you have better preset parameters?

workflow is here

[user-vm]

I'm getting similar issues, it seems to not like batch sizes different from 1. With batch size 4, I get one clear image that doesn't fully match the prompt, two low quality ones, and one broken and noisy image. The simple diffusers wrapper version works properly for batch size 4 - https://github.com/kijai/ComfyUI-ELLA-wrapper . Same settings I think: workflow_ella.json workflow_ella_diffusers.json

[JettHu]

I'm getting similar issues, it seems to not like batch sizes different from 1. With batch size 4, I get one clear image that doesn't fully match the prompt, two low quality ones, and one broken and noisy image. The simple diffusers wrapper version works properly for batch size 4 - https://github.com/kijai/ComfyUI-ELLA-wrapper . Same settings I think: workflow_ella.json workflow_ella_diffusers.json

Thanks for raising the possible issue. I will look at the multi-batch situation.

[quixot1c]

FYI, batch sizes of 3, 5, 6, 7, 9, 10, 11, 12, 13, 14 and 15 actually give errors, while batch size 1, 2, 4, 8 and 16 work (obv. powers of two). I did not test higher batch sizes.

Batch size 8 gives some interesting results:

[JettHu]

Fixed! it will be released as soon as possible after testing