Skip to content

Conversation

gpetters94
Copy link
Contributor

The preprocessors (except canny) need some more work, so they'll go in a separate PR.

Copy link
Contributor

@monorimet monorimet left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice, left a comment -- will also need flag for precision (f16 is only viable path for current GPU optimizations)

if hf_model_name == "stabilityai/stable-diffusion-2-1-base":
encoder_hidden_states_sizes = (2, 77, 1024)

sample = (batch_size, unet_model.unet.config.in_channels, height // 8, width // 8)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do we not use the scaling as such for controlled unet? Can this be made conditional?

@dan-garvey
Copy link
Member

do we not need this anymore?

@monorimet
Copy link
Contributor

do we not need this anymore?

we definitely need it, though controlnet is lower priority until we have base sd/sdxl inference (+ img2img) locked down

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants