FLUX.1 [schnell].

FLUX.1 [schnell] is an open-source image model from Black Forest Labs. Windpaint doesn't serve it yet, so there's no model id or price. Request early access to start with the live models.

Request early access Official model card

Status: Coming soon
Task: Text to image
Input: Prompt
Output: Images
Developer: Black Forest Labs
Parameters: 12B
Licence: Apache-2.0
Released: August 2024

What the model does.

Paraphrased from the official model card.

FLUX.1 [schnell] is the fast variant of Black Forest Labs' first FLUX release. It's distilled to make an image in one to four steps.

It's the only FLUX.1 model with an Apache-2.0 licence, so it's the one that can be used commercially without a separate agreement. It works with the FLUX.1 tooling in ComfyUI and Diffusers.

Coming soon to Windpaint.

FLUX.1 [schnell] isn't served, so it has no model id, no price and no request example here. When it goes live, this page gets its tiers, prices and request fields like the live models have.

Want this model? Request early access. We're adding models as we add capacity.

Request early access

Live for images today: Z-Image Turbo (text to image) and FLUX.2 [klein] 4B (text to image). See Image generation for what the API does now.

Is FLUX.1 [schnell] licensed for commercial use?

Licence: Apache-2.0.

Every model's licence

Yes. Apache-2.0 allows commercial use, including running the model behind a paid API, with no revenue or user caps. The other FLUX.1 open weights are under a non-commercial licence; [schnell] is the Apache-2.0 one.

Apache-2.0 per BFL's licence table. The other FLUX.1 open weights (dev, Fill dev, Canny dev, Depth dev and their LoRAs, Redux dev, Kontext dev, Krea dev) are under the FLUX.1-dev Non-Commercial License. The commercial-licensing and usage-tracking section of the README applies to those [dev] models. flux repo
The card's out-of-scope list rules out: illegal use, harming minors, verifiably false information meant to harm others, harmful PII, harassment, non-consensual nudity or illegal pornography, fully automated decisions that adversely affect legal rights, and large-scale disinformation campaigns. flux repo, FLUX.1-schnell.md
The Hugging Face repo is gated: you have to log in and agree to share your Hugging Face information before you can download the files. FLUX.1-schnell model card

This is our reading of the licence, not legal advice. Read the licence text before you rely on it.

FLUX.1 [schnell] hardware requirements.

Published by Black Forest Labs for running the weights yourself. These are the developer's figures, not Windpaint's.

Every model's published VRAM

Published VRAM: Not stated
Applies to: No VRAM figure on the card. Its example uses CPU offload to save VRAM.

The card gives no VRAM figure. Its diffusers example calls pipe.enable_model_cpu_offload() to save VRAM, with a note to remove it if you have enough GPU power. FLUX.1-schnell model card
BFL's repo provides TensorRT exports in BF16, FP8 and FP4 precision for its FLUX.1 text-to-image models. Its Gradio demo has an --offload flag that moves the model to CPU when it isn't in use. flux repo, text-to-image.md

FLUX.1 [schnell] native resolution.

What the developer's own code and examples use.

The reference CLI defaults to 1360 × 768 and rounds width and height down to multiples of 16. flux repo, cli.py
For the ONNX exports, height and width have to be between 768 and 1344. flux repo, text-to-image.md

Recommended settings for FLUX.1 [schnell].

Steps, guidance and defaults from the developer's card and code, for running the weights yourself.

guidance_scale=0.0, num_inference_steps=4, max_sequence_length=256 (card's diffusers example). FLUX.1-schnell model card
The reference CLI defaults to 4 steps for schnell (50 for dev), uses a 256-token T5 length for schnell, and its guidance setting applies to flux-dev only. flux repo, cli.py

FLUX.1 [schnell] limitations.

As the developer states them.

Not intended or able to provide factual information. flux repo, FLUX.1-schnell.md
It may amplify existing societal biases, may fail to match the prompt, and its prompt following is heavily influenced by prompting style. flux repo, FLUX.1-schnell.md

Questions.

Can I call FLUX.1 [schnell] on Windpaint today?

No. FLUX.1 [schnell] isn't served yet, so it has no model id, no price and no entry in the catalog endpoint. For images today, z-image-turbo and flux-2-klein-4b are live. There's no date for FLUX.1 [schnell].

Can I use FLUX.1 [schnell] commercially?

Yes. It's Apache-2.0, and the card says it can be used for personal, scientific and commercial purposes.

What's the difference between FLUX.1 [schnell] and [dev]?

The licence. [schnell] is Apache-2.0. [dev] and the other [dev]-based FLUX.1 weights are under the FLUX.1-dev Non-Commercial License, and commercial use needs a separate licence from BFL.

What settings should I use for FLUX.1 [schnell]?

The card's example uses 4 steps, guidance 0.0 and max_sequence_length 256.

What resolution does FLUX.1 [schnell] use?

BFL's reference CLI defaults to 1360x768 and rounds width and height down to multiples of 16.

How many parameters does FLUX.1 [schnell] have?

12 billion. It's a rectified flow transformer.

Is FLUX.1 [schnell] open source?

Yes. Its weights are licensed Apache-2.0, an OSI-approved open-source licence.

How do I ask for FLUX.1 [schnell]?

Request early access at app.windpaint.ai/register. We're letting people in gradually while we add models and capacity.

FLUX.1 [schnell].

Start yours.