search

liuliu / swift-diffusion

BSD 3-Clause "New" or "Revised" License
423 stars 33 forks source link

Any chance you could share all the .ckpt files used in the canny2img example? #39

Closed ghost closed 1 year ago

liuliu commented 1 year ago

These are the same you downloaded from the Draw Things app (and if you sniff the traffic, these are from https://static.libnnc.org/controlnet_scribble_2.x_f16.ckpt and sd v2.1 models (probably sd_v2.1_f16.ckpt))

ghost commented 1 year ago

sd_v2.1_f16.ckpt only has unet weights not text encoder and others.

ghost commented 1 year ago

actually they are in : open_clip_vit_h14_f16.ckpt

liuliu commented 1 year ago

Yeah, later models (all accessible through https://static.libnnc.org/xxx) have text encoder / autoencoder weights separated out.