The documentation is not available anymore as the PR was closed or merged.

Hi, @patrickvonplaten!

Could you possibly guide me on this issue and let me know if my approach towards solving #873 is correct or are we looking for something else altogether?

Thank you for your time!

Thanks for the PR - could you also add an example here: https://github.com/huggingface/diffusers/blob/main/examples/community/README.md :-)

Sure thing, will do in a while!

It seems that DiffusionPipeline.from_pretrained() only accepts one string for the pretrained-model path, but here we need 4. What should I do in this case, in your opinion?

One possible solution is to accept only the v1.4 model string while using the v1.1-3 models internally. Should I do that instead?

Would this be an apt example as shown below, in your opinion?

It seems that due to the excessive requirement of memory needed to lead 4 Stable Diffusion models into memory, even Google Colab is unable to handle the requirements and goes on to crash (even with torch.no_grad() on all graphed functions)

from diffusers import DiffusionPipeline
import matplotlib.pyplot as plt

pipe = DiffusionPipeline.from_pretrained('CompVis/stable-diffusion-v1-4', custom_pipeline='suvadityamuk/StableDiffusionComparison')
pipe.enable_attention_slicing()
pipe = pipe.to('cuda')
prompt = "an astronaut riding a horse on mars"
output = pipe(prompt)

plt.subplots(2,2,1)
plt.imshow(output.images[0])
plt.title('Stable Diffusion v1.1')
plt.axis('off')
plt.subplots(2,2,2)
plt.imshow(output.images[1])
plt.title('Stable Diffusion v1.2')
plt.axis('off')
plt.subplots(2,2,3)
plt.imshow(output.images[2])
plt.title('Stable Diffusion v1.3')
plt.axis('off')
plt.subplots(2,2,4)
plt.imshow(output.images[3])
plt.title('Stable Diffusion v1.4')
plt.axis('off')

plt.show()

Have added the code example and the updated file for the Pipeline where it accepts the v1.4 checkpoint and installs the v1.1-3 checkpoints internally. Is that fine?

Thanks a lot for adding this!