cog-stable-diffusion

cog-stable-diffusion copied to clipboard

Default output is PNG. It would be nice to have an input to select the output image type, so you'd get a compressed and more readily reusable file.

zeke

docker: invalid character 'c' looking for beginning of value.

9

Setup: Windows 10 Docker (last) full log (not sure how to make a spoiler): λ docker run -d -p 5000:5000 --gpus=all r8.im/stability-ai/stable-diffusion@sha256:a9758cbfbd5f3c2094457d996681af52552901775aa2d6dd0b17fd15df959bef Unable to find image 'r8.im/stability-ai/stable-diffusion@sha256:a9758cbfbd5f3c2094457d996681af52552901775aa2d6dd0b17fd15df959bef' locally r8.im/stability-ai/stable-diffusion@sha256:a9758cbfbd5f3c2094457d996681af52552901775aa2d6dd0b17fd15df959bef: Pulling...

Musickiller

CI pipeline

1

andreasjansson

Using a scheduler that works for pretrained dreambooth weights and default weights

2

We are in the process of consolidate the prediction from replicate/dreambooth-template into this repo, so the DreamBooth API will use the same predictor code that https://replicate.com/stability-ai/stable-diffusion uses, so all the...

zeke

CUDA out of memory - SD2.1

1

When asking for 4 outputs - with everything else the default, I sometimes get: ``` Output CUDA out of memory. Tried to allocate 12.66 GiB (GPU 0; 39.59 GiB total...

anotherjesse

Reducing inference timings for Sd2.1 Base model

1

I managed to shave off inference timings for SD2.1 by a few seconds for `512x512` (50 steps) and `768x768` (50 Steps). Using just few additions: ```py torch.backends.cudnn.benchmark = True torch.backends.cuda.matmul.allow_tf32...

pratos

Support for Dreambooth Weights

6

I've been experimenting with supporting weights from dreambooth in this model: diff --git a/predict.py b/predict.py index 5630646..2d23a87 100644 --- a/predict.py +++ b/predict.py @@ -10,6 +10,7 @@ from diffusers import (...

anotherjesse

Hey guys, did you take a look at: https://github.com/VoltaML/voltaML-fast-stable-diffusion Seems to be a valid way for speeding up generation by a lot (3x faster)... it optimizes the model using TensorRT...

sauvage7

As there are major differences between stable diffusion 1.x and the 2.x series, there will be folks who will stick to 1.5 vs 2.0 or 2.1 Having tags link git...

anotherjesse

Support for negative prompts & attention slicing

8

New Stable Diffusion diffusers have support for negative prompts & attention slicing. Can you switch to 0.4.1 to enable those options?

yekta