SDXL Turbo
Real-time text-to-image via adversarial diffusion distillation — one step, not fifty.
About
SDXL Turbo, released November 28, 2023, is Stability AI's distillation of SDXL 1.0 using Adversarial Diffusion Distillation (ADD): adversarial plus score-distillation objectives that let it synthesize high-fidelity images in a single inference step, down from the ~50 steps SDXL 1.0 typically needs. It skips classifier-free guidance entirely, which cuts memory use, and in practice 2-4 steps give the best detail for complex prompts. Stability AI's own human evaluations had it beating StyleGAN-T++, OpenMUSE, IF-XL, SDXL, and LCM-XL at one step. It runs on consumer GPUs and kicked off the whole real-time diffusion wave. The honest tradeoffs: single-step outputs are less detailed than full-step SDXL, and it launched under a non-commercial research license — personal and research use only, not commercial products.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Stability AI-reported: human evaluators preferred 1-step outputs over StyleGAN-T++, OpenMUSE, IF-XL, SDXL, and LCM-XL (provider-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Released Nov 28, 2023 per Stability AI announcement; ADD technique with research paper.; Open weights; runs on consumer GPUs; kicked off the real-time diffusion wave.; Stability AI-reported human evals beating StyleGAN-T++, OpenMUSE, IF-XL, SDXL, LCM-XL at 1 step.; Limitations: less detail than full-step SDXL; launched under non-commercial research license..