Black Forest Labs

FLUX 1 Schnell

Name: FLUX 1 Schnell
Author: Black Forest Labs

FLUX 1 Schnell is an open-weight text-to-image generation model from Black Forest Labs, distilled for few-step sampling to produce images in a small number of inference steps.

Contact providers for pricing

Compare Prices

API Pricing

No pricing data available for this model at the moment.

Prices updated daily. Last check: Sep 24, 2026

Model Details

General

Creator: Black Forest Labs
Modalities: Text

Capabilities

Tool Calling: No
Open Source: No
Aliases: black-forest-labs/flux-schnell, FLUX.1 [schnell], FLUX.1 [schnell] FP8

Strengths & Limitations

Strengths

Step-distilled for few-step sampling, so a single image needs far fewer denoising passes than standard diffusion sampling
Open weights, so the model can be self-hosted rather than accessed only through a vendor API
Served by multiple inference providers, allowing price and latency comparison across hosts
FP8 quantized builds are available from some providers, lowering VRAM requirements for local deployment
Low per-image latency makes it practical for interactive prompt iteration and chat-embedded image generation
Supported by mainstream open tooling including Diffusers and ComfyUI, with LoRA and fine-tuning ecosystems built around the FLUX.1 architecture
Part of the FLUX.1 family, so prompts and workflows transfer to higher-fidelity siblings

Limitations

Distillation for speed trades away some fidelity and fine detail compared with the non-distilled FLUX.1 variants
Few-step sampling gives less room to tune quality by raising step counts — pushing steps higher yields limited gains
Text rendering inside images and dense compositional prompts remain error-prone, as with most text-to-image models
No image editing or inpainting interface by default; it is a text-to-image generator unless paired with additional pipeline components
Output quality varies noticeably between providers depending on quantization (for example FP8 builds) and default sampler settings

Key Features

•Text-to-image generation from natural language prompts

•Step-distilled sampling optimized for a small number of inference steps

•Open-weight release available for self-hosting

•FP8 quantized variant offered by some inference providers

•Configurable output resolution and aspect ratio

•Seed control for reproducible generations

•Available through both hosted APIs and local tooling (Diffusers, ComfyUI)

•Part of the FLUX.1 model family from Black Forest Labs

About FLUX 1 Schnell

FLUX 1 Schnell (styled FLUX.1 [schnell]) is a text-to-image generation model released by Black Forest Labs as part of the FLUX.1 family. Within that family, Schnell is the speed-oriented variant: it is a step-distilled model designed to generate an image in a handful of denoising steps rather than the dozens typically required by standard diffusion sampling. It sits alongside the other FLUX.1 releases from the same lab, which target higher fidelity at the cost of more compute per image. As an image generation model, Schnell is evaluated on prompt adherence, image quality at a given step count, and generation latency rather than on context window or token throughput. It accepts a text prompt and typical image-generation parameters — output resolution and aspect ratio, seed, and number of inference steps — and returns raster images. The model is distributed with open weights under a permissive license, which is why it appears both as a hosted API on many inference providers and as a self-hosted checkpoint in local tooling such as ComfyUI and Diffusers. An FP8 quantized variant is also served by some providers to reduce VRAM requirements and speed up inference further. In practice, Schnell is used where images are generated at volume or in interactive loops: prompt exploration, thumbnails and placeholder art, chat-integrated image generation, and batch asset pipelines. Because pricing for image models is generally per image rather than per token, the low step count is the main lever on cost and wall-clock latency. Users who need maximum fidelity or licensing terms suited to a specific commercial workflow typically compare it against the other FLUX.1 variants; see the pricing table on this page for what each provider charges.

Common Use Cases

FLUX 1 Schnell suits workloads where image volume or responsiveness matters more than squeezing out maximum fidelity: rapid prompt exploration before committing a final render to a heavier model, chat assistants and apps that generate images inline while a user waits, thumbnail and social-post generation, concept sketching and moodboarding, and batch asset pipelines that produce many variants per prompt. Its open weights also make it a common choice for teams that need on-premise or air-gapped image generation, or that want to fine-tune and attach LoRAs to a FLUX-architecture base. Projects that need the highest possible detail for print or hero imagery generally reach for the non-distilled FLUX.1 variants instead and reserve Schnell for the iteration stage.

Frequently Asked Questions

How much does FLUX 1 Schnell cost to run?

Image models are usually billed per generated image, and the rate varies by provider, output resolution, and whether you use a standard or quantized deployment. Self-hosting shifts the cost to GPU rental or owned hardware instead. Check the pricing table on this page for current per-provider rates.

What is FLUX 1 Schnell best used for?

High-volume or interactive image generation — prompt iteration, in-app or in-chat image creation, thumbnails, concept art, and batch variant generation — where low latency per image matters more than maximum fidelity.

How does Schnell differ from the other FLUX.1 models?

Schnell is the step-distilled, speed-oriented member of the FLUX.1 family. It produces an image in a small number of inference steps, while the other FLUX.1 variants use full sampling and generally deliver more detail per image at higher compute cost per generation.

Can I run FLUX 1 Schnell on my own hardware?

Yes. The weights are openly released and the model is supported by common open tooling such as Diffusers and ComfyUI. An FP8 quantized variant is also available from some providers, which reduces VRAM requirements relative to the full-precision checkpoint.

Does increasing the number of steps improve output quality?

Only marginally. The model is distilled specifically for few-step sampling, so quality plateaus quickly as steps increase — unlike non-distilled diffusion models, where more steps typically keeps refining the image.

Does FLUX 1 Schnell support image-to-image or inpainting?

It is primarily a text-to-image generator. Some hosted providers and local pipelines expose additional modes by wrapping the model in extra pipeline components, so check the specific provider's API documentation for what it exposes.