photo
RealVisXL V5.0 Lightning
Photoreal people and materials, distilled to five steps.
An SDXL fine-tune aimed squarely at photorealism, distilled so that five steps is enough. It runs with real classifier-free guidance and a negative prompt, which costs double the compute per step and is worth it: at guidance 1.0 it ignores much of the prompt and mangles hands, at 2.0 it follows and the anatomy holds.
- per image
- 7.4 s
- compute
- $0.0011
- steps
- 5 · CFG 2.0
- class
- 1024 px
- runs on
- GPU
What it is good at
- Skin, fabric and light on people
- Cheapest photoreal image in the catalog
- No approval and no schedule — open to every account
Where it struggles
- Illustration and flat styles — that is not what it was tuned for
- Four steps saves ten seconds and costs you the hands
Writing prompts for this model
- Five steps is what it was distilled for
- The default negative prompt is doing real work; clearing it costs anatomy
119 looks · 357 samples · every prompt published
Its own catalog
These looks were written for this model, not shared with the others — the same prompt on a different model tells you which prompt suits which model, not which model is better. Click a look to view it full size or copy the prompt; open it in Studio from there. Every image below is the first of 3 seeds.
Licence
CreativeML Open RAIL++-M — commercial use permitted.
Other models
- Anima 10 steps · 1024 px class · GPU
- Animagine XL 28 steps · 1024 px class · GPU
- Z-Image-Turbo 8 steps · up to 1.5 MP · GPU
- SDXL-Turbo 4 steps · 512 px class · CPU