Sign In

Unleashed Diffusion

173
1.1k
35
Verified:
SafeTensor
Type
Checkpoint Trained
Stats
913
Reviews
Published
Dec 19, 2023
Base Model
Other
Usage Tips
Clip Skip: 2
Hash
AutoV2
4A33DCDD26
License:

Unleashed Diffusion

A model inspired by my LoRA Unleashed.

Info

  • Model flexibility: Moderate

    • Due to the nature of the training and the recommended low CFG scale usage, the prompt may not always be entirely respected (e.g. hair color, eye color, etc...). For it to work, the prompts usually need to be simple...

    • The model is mostly trained with woman portraits ¯\_(ツ)_/¯

  • Artistic freedom: Moderate

    • Trying to apply other art styles on top of this model most likely will not work (realism, cartoon, etc...)

    • Trying to apply other art style concepts may work (painting style, sketch, etc...)

  • LoRA compatibility: Moderate

    • From my testing, many LoRAs wouldn't work on my first attempt. A bit of manual tweaking was often required

  • NSFW: Moderate

    • The model is not meant to generate nsfw, however you may still be able to.

Recommendations for best results

  • Use a low CFG scale value: between 3 and 5

  • Samplers: I found the best results with DPM++ 3M SDE Karras and DPM++ 2M SDE and Euler a

  • Using upscalers (recommended):

    • scaling: 1.5x or more with at least 12 Hires steps

    • upscaler: R-ESRGAN 4x+ Anime6B or R-ESRGAN 4x+ or 4xUltrasharp_4xUltrasharpV10

  • Sampling steps: 30 or more

  • No VAE required!

Troubleshooting

The style isn't being applied to my prompt !!!!

  • This usually means that the model does not know how to apply the style to what you're asking. Try adding more keywords that visually represent the art style such as clouds, flames, colorful, etc...

  • If it still doesn't work, you can always apply my Unleashed LoRA on top!

What's the secret formula?

  • It's a secret 👀!!!

Examples