other
2.6B parameters
Restricted
Reviewed May 2026

SDXL Turbo

2.6B SDXL backbone trained with Adversarial Diffusion Distillation (ADD), producing photorealistic 512px images in a single forward pass. Designed for real-time, interactive text-to-image.

License: stabilityai-non-commercial-research-community·Context: 0 tokens
BLK · VERDICT

Our verdict

OP · Fredoline Eruo|VERIFIED MAY 29, 2026
unrated

Still the speed benchmark — nothing matches ADD for one-step latency at 512px. Held back by the Stability non-commercial license; for production use look at FLUX.1 [schnell] or SD-3.5-medium.

Overview

2.6B SDXL backbone trained with Adversarial Diffusion Distillation (ADD), producing photorealistic 512px images in a single forward pass. Designed for real-time, interactive text-to-image.

Strengths

  • 1-step sampling at 512x512 — sub-100ms on an RTX 4090, real-time on M-series Macs
  • ADD distillation: outperforms the 50-step LCM-XL and matches base SDXL on user-pref studies (per paper)
  • Fits in <8GB VRAM at FP16 — runs on most consumer GPUs
  • Mature SDXL tooling: ComfyUI nodes, diffusers pipeline, ControlNets carry over

Weaknesses

  • Stability non-commercial research license — no commercial use without a Stability membership
  • Native 512px; pushing past degrades quickly vs FLUX or SD3.5
  • Single-step trade-off: less fine detail, weaker prompt adherence than full SDXL
  • Lags FLUX/SD3.5 on text rendering and anatomy

Quantization variants

Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.

QuantizationFile sizeVRAM required
Q4_K_M1.4 GB2 GB

Get the model

HuggingFace

Original weights

huggingface.co/stabilityai/sdxl-turbo

Source repository — direct quantization required.

Hardware that runs this

Cards with enough VRAM for at least one quantization of SDXL Turbo.

Compare alternatives

Models worth comparing

Same parameter band, plus what's one tier above and below — so you can decide what actually fits your hardware.

Frequently asked

What's the minimum VRAM to run SDXL Turbo?

2GB of VRAM is enough to run SDXL Turbo at the Q4_K_M quantization (file size 1.4 GB). Higher-quality quantizations need more.

Can I use SDXL Turbo commercially?

SDXL Turbo is released under the stabilityai-non-commercial-research-community, which has restrictions for commercial use. Review the license terms before using it in a product.

What's the context length of SDXL Turbo?

SDXL Turbo supports a context window of 0 tokens (about 0K).

Source: huggingface.co/stabilityai/sdxl-turbo

Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.

Related — keep moving

Before you buy

Verify SDXL Turbo runs on your specific hardware before committing money.