FLUX.1 [schnell] — implementation by Spacelike AI

Distilled rectified-flow transformer · 12 B parameters · Black Forest Labs

FLUX.1 [schnell] is Black Forest Labs' timestep-distilled image generation model, designed to produce high-quality output in just 1–4 sampling steps — roughly an order of magnitude fewer than comparable undistilled models. The 12 B parameter rectified-flow transformer backbone is shared with FLUX.1 [dev].

Released in August 2024 under the permissive Apache 2.0 license, schnell has become one of the most widely deployed open-weight image models, powering hosted endpoints at Cloudflare, fal, Replicate and others.

Specification

Architecture

Rectified-flow transformer (double + single stream DiT)

Parameters

12 B (backbone)

Training objective

Rectified flow · timestep-distilled

Native resolution

1024 × 1024 · up to 2 MP supported

Text encoders

T5-XXL + CLIP-L (frozen)

Sampler shown

Euler · 4 steps · guidance 1.0

License

Apache 2.0 (fully open)

Release

August 1, 2024

Checkpoint

flux1-schnell.safetensors

Client: Tenstorrent Inc. — implementation · performance optimization.

implementation · Hugging Face · Vendor announcement

Live sample on the Spacelike AI home page.

Sample images on this page are licensed under CC BY 4.0 — reuse with attribution to Spacelike AI and a link back to spacelike.ai.