Experimental Page of AingDiffusion XL
Refer to AingDiffusion XL main page for model info
Check "About Version" to learn more about the issue and how to mitigate the issue.
Often unstable and has a lot of issues but outperforms its AingDiffusion XL counterpart in poses and styles. Issues are mostly regarding artifacts/noise.
If you have any idea on how to fix the model issue, please leave a comment.
How to read version names
Release candidate of what ADXL version - Issue that the model have
Example:
v1.2 RC - Artifacts means that the model is an alternate version of AingDiffusion XL v1.2, and it has artifact issues in the images generated by the model.
===
The model is using Fair AI Public License 1.0-SD license.
Description
v1.2 RC (also identified as ADXL-tueng-M3)
This model has artifacts/noise issues on some seeds with a "perfect prompt combination". This issue happens on AUTO1111/Forge and ComfyUI with CLIPTextEncodeSDXL. The issue happens when you set a different target size from your output size. AUTO1111/Forge uses 4096x4096. While for Comfy, the target size is adjustable but uses 1024x1024 by default. The issue does not affect diffusers.
It outperforms AingDiffusion XL v1.2 by a ton because the first fine-tuning layer uses epoch 8, while ADXL v1.2 uses epoch 5.
To fix the issue as a user, you have to use Hires fix. I use YandereNeo XL with ~0.1-0.2 denoising strength and ~10-15 hires steps to fix the issue.
FAQ
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.