search

google / prompt-to-prompt

Apache License 2.0
3.07k stars 285 forks source link

Failure case of DDIM inversion #27

Open zhihongp opened 1 year ago

zhihongp commented 1 year ago

I have used gnochi_mirror.jpeg and the associated prompt "A cat sitting next to a mirror" to try DDIM inversion using SD v1.4 (50 steps) but found the reconstruction quality is fine. What is the settings (and codes if available) to generate the failure case as shown in the Null-Text paper?

Thanks

FeiiYin commented 1 year ago

Same, my reconstruction code is as below:

images, x_t = text2image_ldm_stable(ldm_stable, prompts, controller, latent=x_t, num_inference_steps=NUM_DDIM_STEPS, guidance_scale=1.0, generator=None, uncond_embeddings=None)
pil_img = ptp_utils.view_images(images)
FeiiYin commented 1 year ago

DDIM inversion works well when the guidance scale is 1 But editing won‘t work

When the guidance scale is 7.5 DDIM inversion will fail