4 ms·

Show HN: I made a local wrapper for Automatic 1111

9 points by saketh105 3y ago

I made an open-source Python library for the Stable Diffusion Web UI. It's a direct alternative to Huggingface Diffusers except it has more features + runs the same scripts as A1111 so the results are replicable. Please give it a star on Github! https://github.com/saketh12/Auto1111SDK https://github.com/saketh12/Auto1111SDK

bArray 3y ago
Seems like an interesting project. Regarding the name, is there permission to use something so similar to AUTOMATIC1111 [1]? > Diffusers will Cuda out of memory/perform very slowly for huge generations, like 2048x2048 images, while Auto 1111 SDK won't. Do we have some numbers on this? I have seen AUTOMATIC1111 fall-over whilst using only half the available of GPU VRAM - there seems to be some weirdness where it tries to allocate before de-allocating the last batch or something. > You can use any of the 6 compatible RealEsrgran models/weights with our RealEsrgran pipeline for upscaling images. Here are the model ids: I've previously had trouble trying to use AUTOMATIC1111 upscalers, it seems like it needs more GPU VRAM than just generating the image already upscaled. [1] https://github.com/AUTOMATIC1111/stable-diffusion-webui https://github.com/AUTOMATIC1111/stable-diffusion-webui
- saketh105 3y ago
  - We will release some benchmarks on the results we found on the Github soon. - Let us know if you run into any bugs with GPU VRAM with Automatic 1111 with the upscalers/benchmarks. We are shipping a feature that will allow you to customize it even further and specify certain flags that can significantly reduce the VRAM (but may slow down the generation speed). These flags are already available in the Web UI; we'll have them easily available in the SDK.
cchance 3y ago
My question is why base a wrapper around a wrapper, a111 is by its definition a wrapper around a bunch of other things, and even then why not just write a wrapper around a more ... easily interfaced with system ... like comfy's api?
- saketh105 3y ago
  The wrapper is the Python library for Automatic 1111. It has a lot of features that don't exist in Huggingface Diffusers and Diffusers also has several limitations that we list here(https://flush-ai.gitbook.io/automatic-1111-sdk/auto-1111-sdk-vs-huggingface-diffusers https://flush-ai.gitbook.io/automatic-1111-sdk/auto-1111-sdk...). The goal is to be the one true Python library for all things Diffusion.

9 points by saketh105 3y ago

bArray 3y ago

Seems like an interesting project. Regarding the name, is there permission to use something so similar to AUTOMATIC1111 [1]? > Diffusers will Cuda out of memory/perform very slowly for huge generations, like 2048x2048 images, while Auto 1111 SDK won't. Do we have some numbers on this? I have seen AUTOMATIC1111 fall-over whilst using only half the available of GPU VRAM - there seems to be some weirdness where it tries to allocate before de-allocating the last batch or something. > You can use any of the 6 compatible RealEsrgran models/weights with our RealEsrgran pipeline for upscaling images. Here are the model ids: I've previously had trouble trying to use AUTOMATIC1111 upscalers, it seems like it needs more GPU VRAM than just generating the image already upscaled. [1] https://github.com/AUTOMATIC1111/stable-diffusion-webui https://github.com/AUTOMATIC1111/stable-diffusion-webui

saketh105 3y ago
- We will release some benchmarks on the results we found on the Github soon. - Let us know if you run into any bugs with GPU VRAM with Automatic 1111 with the upscalers/benchmarks. We are shipping a feature that will allow you to customize it even further and specify certain flags that can significantly reduce the VRAM (but may slow down the generation speed). These flags are already available in the Web UI; we'll have them easily available in the SDK.

cchance 3y ago

My question is why base a wrapper around a wrapper, a111 is by its definition a wrapper around a bunch of other things, and even then why not just write a wrapper around a more ... easily interfaced with system ... like comfy's api?

saketh105 3y ago
The wrapper is the Python library for Automatic 1111. It has a lot of features that don't exist in Huggingface Diffusers and Diffusers also has several limitations that we list here(https://flush-ai.gitbook.io/automatic-1111-sdk/auto-1111-sdk-vs-huggingface-diffusers https://flush-ai.gitbook.io/automatic-1111-sdk/auto-1111-sdk...). The goal is to be the one true Python library for all things Diffusion.