How to run the SDXL Turbo model on AUTOMATIC1111?

The SDXL Turbo model is a fine-tuned Stable Diffusion model trained to generate sharp 512x512 images in one sampling step. To run it on AUTOMATIC1111 (even though there is no official support):

Step 1. Download the SDXL Turbo model file(sd_xl_turbo_1.0_fp16.safetensors) from huggingface and put it in the stable-diffusion-webui/models/Stable-Diffusion folder.

SDXL Turbo sd_xl_turbo_1.0_fp16.safetensors

Step 2. On the txt2img page, select the "sd_xl_turbo_1.0_fp16" model from the Stable Diffusion checkpoint dropdown menu.

Step 3. Set the sampling method to Euler a, sampling steps to 1, size to 512x512, and CFG scale to 1.

Step 4. Enter your prompt and generate the image.

The key settings are using the Euler a sampler, 1 sampling step, and CFG scale of 1. This lines up with the SDXL Turbo model's training to generate sharp images in one step. The noise schedule and number of steps also impact image quality based on the analysis.

Copyright © 2023 SDXL Turbo Online. All rights reserved.