Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Sgm uniform scheduler for SDXL-Lightning models #15325

Merged
merged 1 commit into from
Mar 19, 2024
Merged

Conversation

KohakuBlueleaf
Copy link
Collaborator

Description

Based on the example code/workflow from https://huggingface.co/ByteDance/SDXL-Lightning
I implement the sgm_uniform scheduler for SDXL-lightning series models (also support the SDXL Lightning LoRA models)

I create a new file for this custom scheduler since I guess we will have more and more custom scheduler in the future.

I don't know if we should make some extension API on it. But I will keep it clean at first.

SDXL Lightning Examples (4step Euler):
xyz_grid-1843-2024-03-19_e0d996ee00_sdxl_lightning_4step_2465934700_3f0abdd7-3352x1494

SDXL Lightning LoRA + KXL Delta Examples (4step Euler):
xyz_grid-1844-2024-03-19_45c05e1d94_kohaku-xl-delta-rev1_1187998011_b3da2af2-4192x1494

Checklist:

@MoonRide303
Copy link

@KohakuBlueleaf I've tested it, seems to be working fine - results almost identical as from forge and ComfyUI.

photography of a nice girl
Steps: 8, Sampler: Euler, CFG scale: 1, Seed: 0, Size: 896x1152, Model hash: 43f0501ac4, Model: sdxl_lightning_8step, Clip skip: 2, RNG: CPU, Schedule type: sgm_uniform, SGM noise multiplier: True, Version: v1.8.0-229-ga6b5a513

ComfyUI:
girl-comfy

Forge:
girl-forge

A1111:
girl-a1111

@AUTOMATIC1111 AUTOMATIC1111 merged commit d44b8aa into dev Mar 19, 2024
6 checks passed
@AUTOMATIC1111 AUTOMATIC1111 deleted the sgm_uniform branch March 19, 2024 18:37
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