SDXL-Lightning

ByteDance lightning-fast SDXL distillation. 4-step text-to-image at near-real-time latency.

By ByteDance

Pricing: $0.003 per request

Overview

SDXL-Lightning by ByteDance is a lightning-fast SDXL distillation. 4-step text-to-image at near-real-time latency, suited to interactive UIs and rapid creative iteration.

Key capabilities

• **$0.003 per image at 1024x1024**, fixed per-call pricing. About 333 runs per $1.
• **Multi-region failover**, Runflow routes SDXL-Lightning across multiple regions. If one region has an outage, traffic moves to another automatically.
• **4-step inference**, distilled for sub-second to low-second response on a single GPU.
• **SDXL base**, full Stable Diffusion XL 1.0 compatibility.

SDXL family

| Model | Details |

|---|---|

| **SDXL-Lightning** (this model) | 4-step lightning-fast SDXL distillation |

| **SDXL base** | Stable Diffusion XL 1.0 reference model |

Technical specifications

| Spec | Details |

|---|---|

| Architecture | SDXL distilled with adversarial diffusion |

| Inference Steps | 4 (default), up to 12 |

| Inputs | prompt, negative_prompt, seed, width, height, guidance, steps |

| Output Format | JPEG |

Examples

• a cyberpunk city street at night, neon signs glowing

Frequently asked questions

What is SDXL-Lightning?

SDXL-Lightning by ByteDance is a 4-step distillation of Stable Diffusion XL. Near-real-time text-to-image, suited to interactive UIs and rapid creative iteration.

How much does SDXL-Lightning cost on Runflow?

SDXL-Lightning is $0.003 per image at 1024x1024, fixed per-call pricing. About 333 runs per $1.

What's the latency of SDXL-Lightning?

The 4-step distillation is built for sub-second to low-second response on a single GPU. Typical runs return in around 1 to 2 seconds on Runflow's API.

Can I use SDXL-Lightning commercially?

Yes. Runs made through Runflow's API are licensed for commercial use, including embedding outputs in customer-facing products.

Do I need to manage GPUs to run SDXL-Lightning?

No GPU management is required. Runflow handles inference infrastructure across multiple regions and routes requests automatically. You hit a single REST endpoint, you get a result.

Related models

• FLUX 1.1 [pro] Ultra, Ultra-resolution variant of FLUX 1.1 [pro] with 4MP+ outputs and richer detail. Drop-in upgrade when the standard FLUX 1.1 [pro] resolution is the bottleneck.
• FLUX 1.1 [pro], Black Forest Labs' production-grade FLUX 1.1. State-of-the-art prompt adherence and photoreal quality at competitive per-image pricing.
• GPT Image 2, GPT Image 2.0, OpenAI's latest image model, is capable of creating extremely detailed images with fine typography.
• FLUX.2 [klein] 9B, Text-to-image generation with FLUX.2 [klein] 9B from Black Forest Labs. Enhanced realism, crisper text generation, and native editing capabilities.

Discoverable surfaces

• Dispatch endpoint: POST https://api.runflow.io/v1/models/sd-xl-lightning/runs
• Per-model spec (markdown): https://app.runflow.io/models/sd-xl-lightning/llms.txt
• Docs page: https://docs.runflow.io/models/sd-xl-lightning
• Public OpenAPI spec: https://docs.runflow.io/api/openapi.public.json
• Agent skill (start here): https://www.runflow.io/.well-known/agent-skills/runflow/SKILL.md