Model
Trends
.ai
Model
Trends
.ai
de bästa AI-modellerna med öppen källkod

SDXL-Lightning: four-step generation on any SDXL checkpoint

Ämne: Modeller
Av Captain
Publicerad 2026-09-30
ByteDance's SDXL-Lightning LoRA turns a 30-step SDXL render into a 4-step one. Learn the step counts, the CFG you must use, the sampler, and how to combine it with your favorite checkpoints and style LoRAs.
Den här sidan har inte översatts ännu, så den visas på engelska.

Översikt

SDXL-Lightning is a progressive adversarial distillation of
SDXL
released by ByteDance. The
LoRA
versions are applied on top of any SDXL
checkpoint
and let it produce a finished image in 2, 4 or 8 steps. The catalog entry is the LoRA, which is the flexible way to use it because it keeps your checkpoint's style.
Because the distillation already decides how the image converges, classic guidance scales break it:
Lightning
wants a
CFG
near 1 and a plain Euler sampler with the sgm_uniform scheduler.

Referens

Namn
Typ
Vad det är
LoRA weight
1.0
Use full strength; lowering it reintroduces noise.
Steps
4 or 8
Match the LoRA variant. 4 steps is the sweet spot of speed and quality.
CFG
1-2
1 for the 4-step LoRA; up to 2 for the 8-step variant.
Sampler
Euler
Euler with sgm_uniform. Ancestral or Karras schedulers give blotchy results.
Resolution
1024
Standard SDXL buckets; Lightning does not change the native size.
AD
Inget att installera - kör AI-modeller i molnet
BitVector Prism är det enklaste sättet att komma igång: välj en modell, skriv en prompt och generera i en ren webbapp, utan någon installation alls. BitVector finns också på Discord och på webben (SpyGlass).

Steg för steg

  1. Load an SDXL checkpoint you like (realistic, anime or painterly) and attach the Lightning LoRA at 1.0.
  2. Set steps to 4, CFG to 1, sampler Euler, scheduler sgm_uniform.
  3. Prompt as you normally would for that checkpoint;
    negative prompts
    have little effect at CFG 1, so lean on the positive.
  4. Generate a batch; Lightning is fast enough to explore eight seeds at a time.
  5. For a final image, switch to the 8-step variant at CFG 1.5 for a little more detail.

Exempel

Fast portrait with a realistic checkpoint

photo of a fisherman mending nets at a harbor, morning fog, warm side light, 35mm, natural skin texture
Steps 4, CFG 1, Euler, sgm_uniform, Lightning LoRA 1.0

Anime checkpoint plus style LoRA

1girl, reading on a train, window light, autumn fields outside, soft colors, detailed background
Lightning LoRA 1.0 plus a style LoRA at 0.7, 8 steps, CFG 1.5

Creative idea: rapid storyboard

storyboard frame, wide shot, a courier cyclist weaving through monsoon traffic, neon reflections on wet asphalt, cinematic anamorphic look
Generate 12 variations and pick the sequence

Tips

  • Lightning is the fastest way to audition style LoRAs: stack one at 0.7 and iterate in seconds.
  • Hands and small faces suffer most at 4 steps; use 8 steps or an inpaint pass for those.
  • Hires fix still works: 1.5x at 0.4 denoise with the same 4-step settings.
  • If colors look burnt, your CFG is too high; it is the single most common mistake.

Felsökning

Over-saturated, crunchy images

Varför det händer
CFG above 2.

Så löser du det
Set CFG to 1 and keep Euler with sgm_uniform.

Noisy or unfinished output

Varför det händer
Step count lower than the LoRA variant, or LoRA weight below 1.

Så löser du det
Match steps to the variant (4 or 8) and set the weight to 1.0.

AD
PirateDiffusion
Inget att installera - kör AI-modeller i molnet
PirateDiffusion finns bara på Telegram och är byggt för proffs: tusentals modeller, LoRA och arbetsflöden som styrs med chattkommandon, med obegränsad generering till ett fast pris.

Frågor

Does Lightning work with Pony and Illustrious checkpoints?

Yes, they are SDXL-based. Expect slight style shifts; the 8-step variant is gentler.

Can I use a negative prompt?

You can, but at CFG 1 it barely acts. Put everything you need in the positive prompt.

Länkar och källor

Modeller i den här guiden

Skriven av
Captain

Relaterade guider

Modeller
Hyper-SD XL: one-to-eight-step SDXL with a flexible LoRA
Hyper-SD from ByteDance accelerates SDXL checkpoints to as few as one step. Learn which variant to pick, the CFG and sampler rules, and how to combine it with checkpoints and LoRAs without losing quality.
Av Captain
2026-10-02
Modeller
Stable Diffusion XL 1.0: prompts, settings and what it still does best
How to get the most out of the official SDXL 1.0 base model: native resolutions, CFG and sampler settings, the refiner, negative prompts and ideas for styles where SDXL still shines.
Av Captain
2026-10-01
Modeller
Smooth Detail Booster: cleaner shading and more detail on Illustrious and Pony
DigitalPastel's Smooth Detail Booster LoRA adds detail and smooth shading to anime checkpoints without changing the art style. Learn the weight range, the prompts that help it, and how to pair it with quality tags.
Av Captain
2026-09-29
Modeller
Playground v2 1024 Aesthetic: a second opinion to SDXL
Playground v2 is an SDXL-architecture model trained from scratch for aesthetics. Learn how it differs from SDXL, its CFG and step preferences, and the painterly and realistic prompts where it beats the original.
Av Captain
2026-10-02

Fler guider

Modeller
FLUX.1 Dev: how to prompt it, best settings and creative ideas
A practical guide to FLUX.1 Dev by Black Forest Labs: natural-language prompting, guidance and step settings, LoRA stacking, text rendering and prompt ideas that play to its strengths.
Av Captain
2026-09-30
Modeller
Stable Diffusion 1.5: the classic model, prompted properly
Stable Diffusion 1.5 is still worth knowing: the right resolution, CFG and sampler, how to use its huge library of LoRAs and embeddings, and creative prompt ideas suited to a 512-pixel model.
Av Captain
2026-10-02
Modeller
FLUX.2 Dev: prompting the newest Black Forest Labs model
FLUX.2 Dev brings larger prompts, better text, stronger realism and multi-reference editing. This guide covers the turbo workflow, guidance and resolution settings, prompt structure and ideas that use its new strengths.
Av Captain
2026-10-03
Modeller
Qwen-Image: long prompts, perfect text and bilingual posters
Qwen-Image is the model to reach for when the words in the picture matter. Learn how to write its long descriptive prompts, render English and Chinese text accurately, set CFG and steps, and explore layout-heavy creative ideas.
Av Captain
2026-09-29
ComfyUI
ComfyUI nodes: documentation and troubleshooting
Plain-language documentation for the ComfyUI nodes people use most: what each node does, every input and output, how to wire it, the settings that matter, and the errors it throws with the fix for each one.
Av Captain
2026-06-02
Modeller
Waifu Diffusion: Danbooru-tag prompting for a classic anime model
Waifu Diffusion is the anime base model that taught a generation how to prompt with Danbooru tags. This guide covers tag order, quality tags, negatives, the right resolution and ideas for clean anime illustrations.
Av Captain
2026-10-01