Text to Image
model output
Note: we now use compel syntax for prompt weights (like InvokeAI). Simple A1111-style weights will be converted automatically.
stabilityai/stable-diffusion-2-1-base
Stable Diffusion v2.1
This is the default model, and great to start with. However, you may get much better results with the specialized models, particularly ICBINP-Final, DreamShaper_7 and the various anime models. Some models take longer and use more credits with their default settings. It's worth looking at what people have shared on the front page, and checking which models and settings were used to produce them.
Addons
Browse CivitAI and filter for
LORA
models or
Embeddings
(aka Textual Inversions).
    This can be temperamental. See usage notes.
    1. We're still working on this. Expect things to break! Your feedback is greatly appreciated and you can give it in the forums.
    2. LoRAs: 1) Currently, only one LoRA can be used at a time (tracked upstream at diffusers#2613). 2) LoRAs work best on the same model they were trained on; results can appear very garbled otherwise. We hope that moving forwards, the community will move away from fine-tuned models and we'll simply have many LoRAs for the original SDXL base. Let's see.
    3. Textual Inversions: currently, once a token has been added to a model by any user, it cannot be removed until a new model is loaded. This means it is currently not possible to easily update a token with a newer version of the embedding. For most users, this is unlikely to be an issue.

    Classifier-Free Guidance (Scale)

    Number of Inference Steps

    👣
    New DPM sampler needs just 20 steps! 🎉

    Width

    ⭤

    Height

    Copyright © 2022 by Gadi Cohen (gadicc)
    and Wastelands Networking Ltd.
    SD-MUI on GitHub | Terms of Service | Privacy Policy