12 SDXL

cmdr2 edited this page 2023-08-31 21:34:12 +05:30

Table of Contents

• How to use SDXL:
• Usage tips:

SDXL models are trained to create larger images with better image quality. They can also make good images at 512x512 resolution, so they're often a good replacement for SD 1 or 2 models, in terms of image quality.

They do, however, consume more GPU and system RAM, so please experiment with your image sizes and VRAM usage level setting (in the Settings tab) while using SDXL models.

How to use SDXL:

1. Download whichever SDXL model you want:

• Base model (works for text-to-image, image-to-image, and inpainting): base model download link
• Refiner model (image-to-image only): refiner model download link

2. Save the model to the models/stable-diffusion folder.

Finally, in the Easy Diffusion UI, click the "refresh" icon next to the Models dropdown, and select the downloaded model and generate the image.

Usage tips:

1. If you get "Out of Memory" errors, try setting the VRAM Usage Level setting to "low" in the Settings tab, and press Save.
2. Use small values for Prompt Strength when using the "Refiner" model. E.g. 0.2
3. The "Refiner" model supports text prompt, along with an image. You can use text to describe how your image should be refined.

Basics

• Installation
• How to Use
• Writing Prompts
• Image Modifiers
• Inpainting
• Samplers
• Summary of every UI option
• Common error messages (and solutions)

Intermediate

• Custom Models
• Prompt Syntax
• UI Plugins
• Embeddings
• LoRA
• SDXL
• ControlNet
• Seamless Tiling
• xFormers
• The beta channel

Advanced topics

• Run on Multiple GPUs
• Model Merging
• Custom Modifiers
• Config settings