SDXL generalist · open community model
DreamShaper XL — one checkpoint for many styles
DreamShaper XL is a Stable Diffusion XL fine-tune by Lykon, aimed at “doing a bit of everything well”: photos, art, anime, and comics. Explore the version line from alpha to Lightning, learn official few-step tips, and try a browser demo.
- Base
- SDXL 1.0
- Current focus
- Lightning
- Try in browser
- HF Space
Why DreamShaper XL
A general-purpose SDXL finetune, not a one-trick model
Creators reach for DreamShaper XL when they want one checkpoint that covers many looks, with newer Turbo and Lightning builds for fast sampling.
Generalist style coverage
Public model pages position DreamShaper XL as a multi-style tool for photos, art, anime, and comics — closer to a general image model than a single-niche realism fork.
Author-claimed gains over base SDXL
Lykon’s public notes say the finetune often improves people, edges, and NSFW quality versus stock SDXL 1.0, and that a separate refiner is usually unnecessary if you use highres fix or upscale + img2img.
Fast Turbo and Lightning variants
Beyond full-quality alphas, the line includes Turbo and Lightning DPM++ SDE builds for few steps and low CFG when you want quicker tests.
Open downloads on Civitai and HF
Primary distribution is Civitai model 112902, with Hugging Face mirrors under Lykon (1.0, Turbo, Lightning repos). License is CreativeML Open RAIL++-M / openrail++.
Lineage
From XL 0.9 alpha to Lightning
DreamShaper XL grew on Stability AI’s SDXL stack. Public pages show early alphas, then Turbo, then Lightning — with the author now steering people toward few-step builds.
- 1
Alpha1 on SDXL 0.9
Early public test on SDXL 0.9. Community workflows often refined outputs with older DreamShaper or AbsoluteReality highres passes.
- 2
Alpha2 on SDXL 1.0
Trained checkpoint (not a multi-merge soup): UNet and text encoders from SDXL 1.0, finetuned on a slice of DreamShaper 7 data, with the XL 0.9 VAE preferred by the author.
- 3
Turbo few-step build
Turbo brought low-CFG, few-step sampling (often DPM++ SDE Karras). The author notes extra commercial limits for Turbo unless Stability AI permits otherwise.
- 4
Lightning DPM++ SDE (current focus)
Released around 2024-02-21. Author guidance pushes Turbo/Lightning over older alphas for everyday use, with ~3–6 steps and CFG around 2.
How to use it
A simple starting recipe
Run weights in ComfyUI, Forge, or Automatic1111, or test ideas in our embedded Hugging Face Space. Match sampler, steps, and CFG to the variant you loaded.
1. Prefer Lightning or Turbo for speed
Lykon publicly recommends Turbo or Lightning instead of older Alpha2 for most new work.
2. Lightning defaults
Start around CFG 2, about 3–6 steps, sampler DPM++ SDE Karras (or Normal). Keep the first pass near 1024 on the long side.
3. Turbo defaults
CFG about 2, about 4–8 steps, DPM++ SDE Karras (not 2M). Turbo can also run “non-Turbo style” at higher steps with different samplers; Lightning should not be forced that way.
4. Test in the playground
Use the dual-tab playground to open the coralLight DreamShaper XL Space without installing weights first.
Playground
Try DreamShaper XL in the browser
Open the playground and switch tabs between playground 1 and playground 2. No local GPU install required to start.
The Space is third-party. Availability, loaded weights, and queue times can change upstream.
Launch playground
Resources
Where to go next
Features, history & future
A junior-high-friendly guide to what DreamShaper XL is, how versions grew, and what public sources say about the future.
Read articleLightning & latest features
What the current Lightning-focused line includes: settings, files, license notes, and where to download.
Read Lightning guideCivitai model page
Official gallery, version files, and community samples for DreamShaper XL.
Open Civitai