SDXL Lightning
Stableby ByteDance
Progressive distillation of SDXL from ByteDance. Available in 1-step, 2-step, 4-step, and 8-step variants via LoRA or full UNet checkpoints. Achieves near SDXL quality in as few as 2-4 steps — significantly faster than SDXL's standard 25-50 steps.
VRAM requirements, GPU fit, and setup notes for SDXL Lightning, including 8GB/12GB fit guidance where relevant. Recommended runtimes: ComfyUI and Diffusers support. Best download size: ~6.9 GB at FP16.
- 2-4 step generation — near real-time on consumer GPUs
- Progressive distillation preserves SDXL quality
- Available as LoRA or full UNet checkpoint
- Compatible with existing SDXL ecosystem
Your hardware
Detecting...
Image Quality Benchmarks
Measured quality metrics for SDXL Lightning outputs.
How often humans prefer this model's output (0-100%)
Visual quality and composition rating (5-9 scale)
VRAM Requirements by Resolution and Precision
Compare which GPUs can run SDXL Lightning at different precisions. FP8 uses less memory than FP16 when available, and the grade shows how comfortably each GPU handles the workload.
FP16 (full precision)
| Resolution | VRAM Required | RTX 4090 24GB | RTX 3060 12GB | RTX 4060 8GB | MacBook Pro M4 Pro 24GB |
|---|---|---|---|---|---|
| 512×512 | 7.1 GB | S● | S● | A● | S● |
| 768×768 | 7.3 GB | S● | S● | A● | S● |
| 1024×1024 | 7.5 GB | S● | S● | B● | S● |
Optimization Tips
ControlNets available
Add guided generation with 3 adapters (+1.2 GB VRAM each)
Rich LoRA ecosystem
Customize style, characters, and quality with community LoRAs
Run with Python
from diffusers import StableDiffusionXLPipeline
import torch
pipe = StableDiffusionXLPipeline.from_pretrained(
"ByteDance/SDXL-Lightning",
torch_dtype=torch.float16
)
pipe.to("cuda")
image = pipe(
prompt="your prompt here",
num_inference_steps=4,
height=1024,
width=1024,
).images[0]
image.save("output.png")Get started
Setup instructions for running SDXL Lightning locally
1. Download the model
Get the checkpoint from HuggingFace
2. Place in:
ComfyUI/models/checkpoints/3. Launch ComfyUI
python main.pyMemory Breakdown
VRAM allocation at 1024×1024 on RTX 4090 24GB (24 GB)
Estimated Generation Time
Time per image at 1024×1024, 28 steps, FP16.
Sample Outputs
Available Formats, Downloads & Setup
Download SDXL Lightning in the precision that matches your GPU. Lower precision usually means less VRAM pressure, while higher precision keeps more quality.
| Formato | Precisión | Tamaño | Proveedor | |
|---|---|---|---|---|
| safetensorsRecomendado | FP16 | 6.9 GB | official | Descargar |
ControlNet Support
3 ControlNets available for SDXL Lightning. ControlNets add guided image generation (edges, depth, pose) at the cost of extra VRAM.
Canny Edge
+1.2 GB VRAMEdge detection for structural guidance. Full SDXL ControlNet compatibility.
Depth Map
+1.2 GB VRAMDepth-based spatial control for maintaining 3D composition.
OpenPose
+1.2 GB VRAMHuman pose control for character positioning and body language.
LoRA Ecosystem
Large EcosystemFull SDXL LoRA ecosystem compatible. Lightning itself available as LoRA (200MB) for use with any SDXL checkpoint.
Browse all LoRAs on CivitAIRelated Workflows
You might also like
Frequently asked questions
FAQ — SDXL Lightning VRAM, Runtimes & Fit
How much VRAM does SDXL Lightning need?
SDXL Lightning (2.6B parameters) requires approximately 7.5 GB of VRAM at FP16 precision for standard 1024×1024 image generation. If you want a lighter setup, lower precisions like FP8 can reduce memory use when available.
Can I run SDXL Lightning on an 8GB GPU?
SDXL Lightning usually needs more than 8GB for comfortable local use. Check the VRAM table above for the exact resolution and precision trade-off.
Does SDXL Lightning work in ComfyUI and Diffusers?
SDXL Lightning is marked for ComfyUI and Diffusers support in our catalog, so those are the runtimes we recommend first for local setup. If your workflow uses another front end, check the model's available formats and workflow notes above before downloading.
Can I run SDXL Lightning on RTX 4090?
Yes, the RTX 4090 (24 GB VRAM) can run SDXL Lightning comfortably at FP16. Expected generation time is around ~1.5s per image at 1024×1024.
Does SDXL Lightning support ControlNet?
Yes, SDXL Lightning has 3 ControlNet adapters available: Canny Edge, Depth Map, OpenPose. Each ControlNet adds roughly 1.2 GB of extra VRAM.
Does SDXL Lightning have LoRA support?
Full SDXL LoRA ecosystem compatible. Lightning itself available as LoRA (200MB) for use with any SDXL checkpoint. The LoRA ecosystem for SDXL Lightning is rated as "large". Each LoRA adds roughly 0.2 GB of extra VRAM.
How fast is SDXL Lightning?
On a reference GPU (RTX 4090 24GB), SDXL Lightning generates a 1024×1024 image in approximately ~1.5s at FP16 with 28 inference steps. Faster GPUs with higher memory bandwidth will produce images more quickly.
About SDXL Lightning
See also