stabilityai

/

stable-diffusion-2-1-base

like 627

Follow

Stability AI 5,897

I'm using Dream Studio's stable diffusion and set it up locally. I'm keeping the parameters the same but still, there is a difference in output. What could be the possible reason for this?
Parameters:

prompt: A digital illustration of an eagle, piano, red, mountains, landscape, 1987, symmetric, detailed, extremely detailed, trending in artstation, high resolution, 8K, fantasy vivid colors

model: stable diffusion v2.1 base

height: 512

width: 512

seed: 2851568514

Scheduler: PLMS and Euler (tried both)

guidance scale: 14

steps: 94

Dream studio is generating better eyes and crisp image. I would like to understand the reason for this difference.

Hi, I met the same problem. Have you solved it?

Try to change CLIP skip count

The reason is that actually Dream Studio is using a different commercial model, which is not open source. This has been confirmed by Emad, CEO of stabilityai.

Hello @adanish ! So that has been confirmed for v2.1, but what about v2.0, v1.5 and v1.4 ? They are also using different commercial models in Dream Studio compared to the open-sourced ones ?

If you could link the place where Emad was talking about this, that would be cool.

Thanks :)

I am also finding this discrepancy. And I am seeing it with all the available models. Has anyone learned more about this