by Stability AI
Adversarial distillation of SD 1.5 for single-step image generation. Only 0.86B UNet — the smallest and fastest Stable Diffusion variant. Quality is lower than SD 1.5 but generation is nearly instant. Ideal for real-time interactive use.
VRAM requirements, GPU fit, and setup notes for SD Turbo, including 8GB/12GB fit guidance where relevant. Recommended runtimes: ComfyUI and Diffusers support. Best download size: ~2.5 GB at FP16.
Your hardware
Detecting...
Measured quality metrics for SD Turbo outputs.
How often humans prefer this model's output (0-100%)
Visual quality and composition rating (5-9 scale)
Compare which GPUs can run SD Turbo at different precisions. FP8 uses less memory than FP16 when available, and the grade shows how comfortably each GPU handles the workload.
| Resolution | VRAM Required | RTX 4090 24GB | RTX 3060 12GB | RTX 4060 8GB | MacBook Pro M4 Pro 24GB |
|---|---|---|---|---|---|
| 512×512 | 3.5 GB | S● | S● | S● | S● |
| 768×768 | 4.2 GB | S● | S● | S● | S● |
| 1024×1024 | 5.2 GB | S● | S● | S● | S● |
ControlNets available
Add guided generation with 3 adapters (+0.7 GB VRAM each)
Rich LoRA ecosystem
Customize style, characters, and quality with community LoRAs
from diffusers import StableDiffusionPipeline
import torch
pipe = StableDiffusionPipeline.from_pretrained(
"stabilityai/sd-turbo",
torch_dtype=torch.float16
)
pipe.to("cuda")
image = pipe(
prompt="your prompt here",
num_inference_steps=1,
height=512,
width=512,
).images[0]
image.save("output.png")Get started
Setup instructions for running SD Turbo locally
1. Download the model
Get the checkpoint from HuggingFace
2. Place in:
ComfyUI/models/checkpoints/3. Launch ComfyUI
python main.pyVRAM allocation at 1024×1024 on RTX 4090 24GB (24 GB)
Time per image at 1024×1024, 28 steps, FP16.
Download SD Turbo in the precision that matches your GPU. Lower precision usually means less VRAM pressure, while higher precision keeps more quality.
| Format | Precision | Size | Provider | |
|---|---|---|---|---|
| safetensorsRecommended | FP16 | 2.5 GB | official | Download |
3 ControlNets available for SD Turbo. ControlNets add guided image generation (edges, depth, pose) at the cost of extra VRAM.
Edge-based structural guidance. SD 1.5 ControlNet compatible.
Human pose estimation for character control.
Full SD 1.5 LoRA compatibility
Frequently asked questions
SD Turbo (0.86B parameters) requires approximately 5.2 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.
SD Turbo usually needs more than 8GB for comfortable local use. Check the VRAM table above for the exact resolution and precision trade-off.
SD Turbo 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.
Yes, the RTX 4090 (24 GB VRAM) can run SD Turbo comfortably at FP16. Expected generation time is around 500ms per image at 1024×1024.
Yes, SD Turbo has 3 ControlNet adapters available: Canny Edge, Depth Map, OpenPose. Each ControlNet adds roughly 0.7 GB of extra VRAM.
Full SD 1.5 LoRA compatibility The LoRA ecosystem for SD Turbo is rated as "massive". Each LoRA adds roughly 0.1 GB of extra VRAM.
On a reference GPU (RTX 4090 24GB), SD Turbo generates a 1024×1024 image in approximately 500ms at FP16 with 28 inference steps. Faster GPUs with higher memory bandwidth will produce images more quickly.
See also