[FAQ] docker-diffusers-api

Frequently Asked Questions


  • My results look bad or don’t look as good as AUTOMATIC1111, why?

    1. Resolution: make sure you explicitly set the correct resolution for your model, e.g. 512x512 for original SD models and SDv2+ “base” models, and 768x768 for SDv2+ models, etc.

    2. Prompt length and weights: for long prompts (more than 77 tokens) and prompts with weights like (((big eyes))), (blue hair:1.2), etc, it will be ignored by the built-in StableDiffusion pipelines; instead use Lpw_stable_diffusion pipeline (longer prompts, prompt weights!).

    3. Use the same seed when comparing to AUTOMATIC1111 or others, and for reporting bugs / questions.

    4. Check the scheduler (“sampler”) and num_inference_steps. Certain schedulers may work much better or worse on certain models.


  • What models does it run?

    • All models supported by the huggingface/diffusers library
    • All checkpoints that can be converted to diffusers format
      (downloaded & converted at build time if you set CHECKPOINT_URL)

Dev / Deployment