Spaces:
Running
on
CPU Upgrade
Settings to duplicate results on local installation
I would like to get the same results I would get in this space but on a local installation on a mac using Draw Things. I an using the default SDXL model, but I cannot figure out how to repeat the look of results from this space. I am using Euler Ancestral and 20 steps at 7.5 guidance just as in the default settings. Anything I'm missing? Identical prompts give very different results.
I have the same question. I have tried to locally generate images similar to those in this space using the default SDXL model. This model must be unique here :(
probably this article will help?
https://huggingface.co/tasks/unconditional-image-generation
google features 14 Image Generation models at their huggingface main site ...
peterterpe, I went through the link and am none the wiser. I have no idea how any of that would help me use SDXL to reproduce results from google's SDXL space. The other models are irrelevant and require special hardware. Maybe someone at google has some insight? Is it a plain SDXL checkpoint? Or are they using a proprietary trained SDXL? Is the hardware doing something different? Is it the scheduler? In the documentation when the page was still up it said they were using Euler at 20 steps with CFG 7.5. Beyond that I have no idea what else could be so different.