CompVis / latent-diffusion

High-Resolution Image Synthesis with Latent Diffusion Models
MIT License
11.83k stars 1.53k forks source link

The text2img pretrained weight seems not working #264

Open zsh2000 opened 1 year ago

zsh2000 commented 1 year ago

Thanks for sharing the great work! I downloaded the text-to-image weight https://ommer-lab.com/files/latent-diffusion/text2img.zip with the setting in "./models/ldm/text2img256/". However, it seems that the pretrained weight has some issue and it produces meaningless images when putting the text prompt "A living room with a large window and a large wooden table." image

Instead, the weight downloaded from https://ommer-lab.com/files/latent-diffusion/nitro/txt2img-f8-large/model.ckpt and use the setting in "./configs/latent-diffusion/txt2img-1p4B-eval.yaml" works well and produces decent results.

Has anyone also come across similar issues?

ndjaitly commented 1 year ago

Sorry to add to a very old thread -- but maybe it helps someone.. I faced a similar issue with text2img256 -- In fact all the images I got from this model seem to be meaningless..However, the above model (txt2img-f8-large) indeed worked just fine.

order-a-lemonade commented 11 months ago

same problem.

Kelvar00 commented 3 months ago

I met the same problem, is there any solutions? I wanted to figure out how to generate meaningful figures using this ckpt