inpaint pipelines

[heorhiikalaichev]

Hi!

Thank you for your great job!!

Do you provide a possibility to use inpainting pipelines (StableDiffusionInpaintPipeline)?

[deltheil]

Hi @heorhiikalaichev - yes, we do support inpainting. You can find an example with SD 1.5 part of the test suite:

refiners/tests/e2e/test_diffusion.py

Lines 764 to 784 in 7839c54

	sd15 = sd15_inpainting
	n_steps = 30
	prompt = "a large white cat, detailed high-quality professional image, sitting on a chair, in a kitchen"
	negative_prompt = "lowres, bad anatomy, bad hands, cropped, worst quality"
	clip_text_embedding = sd15.compute_clip_text_embedding(text=prompt, negative_text=negative_prompt)
	sd15.set_num_inference_steps(n_steps)
	sd15.set_inpainting_conditions(kitchen_dog, kitchen_dog_mask)
	manual_seed(2)
	x = torch.randn(1, 4, 64, 64, device=test_device)
	for step in sd15.steps:
	x = sd15(
	x,
	step=step,
	clip_text_embedding=clip_text_embedding,
	condition_scale=7.5,
	)
	predicted_image = sd15.lda.decode_latents(x)

(Also, the weights conversion script handles runwayml/stable-diffusion-inpainting by default:

refiners/scripts/prepare_test_weights.py

Line 631 in 7839c54

download_sd15("runwayml/stable-diffusion-inpainting")

)

Let us know if you have any other questions!