transformer_latent_diffusion icon indicating copy to clipboard operation
transformer_latent_diffusion copied to clipboard

Weird image when running the sample generator

Open metatl opened this issue 1 year ago • 3 comments

Just installed and give it a test of the trained model.

from tld.configs import LTDConfig, DenoiserConfig, TrainConfig from tld.diffusion import DiffusionTransformer

denoiser_cfg = DenoiserConfig(n_channels=4) #configure your model here. cfg = LTDConfig(denoiser_cfg=denoiser_cfg)

diffusion_transformer = DiffusionTransformer(cfg)

out = diffusion_transformer.generate_image_from_text(prompt="a cute cat") I got something like this: image

Not sure why.

metatl avatar May 03 '24 12:05 metatl

Hey @metatl, try using the legacy_dh_order branch - the model was trained with a small but annoying difference in ordering of the hidden dimensions and head dimensions and unfortunately I haven't found a simple way to fix that. Let me know if that doesn't work.

apapiu avatar May 09 '24 19:05 apapiu