Home Artists Prompts GPT


Image to image XL turbo


SDXL Turbo achieves state-of-the-art performance with a new distillation technology, enabling single-step image generation with unprecedented quality, reducing the required step count from 50 to just one.






Advertisement

Transform a photo or a sketch with SDXL Turbo in near real time: upload an image, describe the result, and the model redraws it in a single step. Free, no account. SDXL Turbo is a distilled version of SDXL that trades some detail for speed, which makes it ideal for quick iterations.

How to use SDXL Turbo image-to-image

  1. Upload a source image or draw a rough sketch in the frame.
  2. Describe what the result should look like in a short prompt.
  3. Move the strength slider: low keeps your composition, high lets the model reinvent it.
  4. Iterate quickly, then reproduce the best result at full quality with the SDXL demo.

Limits and questions

Why is the quality lower than SDXL?

Turbo models generate in 1 to 4 steps instead of 25 to 30: faster, but with less detail and weaker prompt following. Use them to explore, then finalise with SDXL.

Does the negative prompt work?

Barely. Turbo models run at CFG 1 to 2 where the negative prompt has almost no effect. Put everything in the positive prompt.

Why is it not instant here?

This demo runs on shared CPU hardware, so each step takes seconds rather than milliseconds. This demo is hosted by a third party (Hugging Face Space) and embedded here; it can be slow, queued or temporarily unavailable when the host is busy. Nothing to install, no account needed.

Read the guide: When the negative prompt does nothing