FLUX.1

What is FLUX.1?

FLUX.1 is the first generation of text-to-image models from Black Forest Labs, a German AI lab founded by researchers behind VQGAN, Latent Diffusion and Stable Diffusion. It launched on 1 August 2024 in three versions. FLUX.1 [pro] was the top-quality model offered through the BFL API and partners, FLUX.1 [dev] was an open-weight model distilled from it for non-commercial use, and FLUX.1 [schnell] was the fastest version, released under the Apache 2.0 licence. All three public models use a hybrid transformer design scaled to 12 billion parameters.

The family grew quickly after launch. FLUX1.1 [pro] followed with Ultra and Raw modes, FLUX.1 Tools added Fill, Depth, Canny and Redux in November 2024, and FLUX.1 Kontext brought in-context image editing in 2025, with an open-weight Kontext [dev] released on 26 June 2025. Black Forest Labs now labels FLUX.1 as its previous generation, behind FLUX.2 and FLUX 3, but several FLUX.1 endpoints remain on its API price list and the open weights are still on Hugging Face as of October 2026.

Key Features

• Three original variants: FLUX.1 shipped as [pro], [dev] and [schnell], each balancing quality, speed and licence terms differently. Teams can prototype locally on schnell or dev and move to paid API models when they need the highest output quality.
• Fast schnell generation: FLUX.1 [schnell] produces images in only 1 to 4 sampling steps, according to its model card. Black Forest Labs trained it with latent adversarial diffusion distillation, which makes it the quickest option for local testing and previews.
• Open weights on Hugging Face: FLUX.1 [dev] and [schnell] publish their weights on Hugging Face, with dev behind a gated licence agreement. Developers can run them through the Diffusers library or ComfyUI workflows on their own hardware instead of paying per image.
• FLUX1.1 pro, Ultra and Raw: FLUX1.1 [pro] upgraded the flagship model, and its Ultra and Raw modes target higher resolution and a more candid photographic look. All three remain on the BFL API at $0.04 to $0.06 per image as of October 2026.
• FLUX.1 Kontext editing: FLUX.1 Kontext edits images from text and image prompts together, supporting iterative changes and character consistency across scenes. Black Forest Labs offers Kontext [pro] and [max] through its API and Kontext [dev] as open weights.
• Fill inpainting and outpainting: FLUX.1 Fill repaints selected areas of an image or extends it beyond the original frame. The [dev] version is open-weight, while Fill [pro] costs $0.05 per image on the BFL API.
• Redux restyling: FLUX.1 Redux creates variations of an input image or restyles it with slight changes. Black Forest Labs released a [dev] adapter and integrated the [pro] version with FLUX1.1 [pro] Ultra.
• Wide partner availability: FLUX.1 models are reachable through the BFL API and partner hosts such as Replicate, fal.ai and Together AI. Local apps listed on the model cards include ComfyUI, Draw Things and DiffusionBee.

Pros

• Free commercial option: FLUX.1 [schnell] is released under Apache 2.0, so you can use it for personal, scientific and commercial work at no licence cost. That makes it a rare open model that small businesses can deploy without negotiating terms.
• Strong prompt adherence: Black Forest Labs says FLUX.1 matches leading closed models on output quality and prompt following. Directory reviews also praise its photorealism and human anatomy, which were weak points in older open image models.
• Low API prices: FLUX.1 API calls cost between $0.04 and $0.08 per image on the official price list. One credit equals one US cent, so budgeting for a batch of images is simple arithmetic.
• Self-hosting freedom: Open weights for dev, schnell, Fill, Redux and Kontext dev let you run models on your own GPUs. Your prompts and images then stay on your infrastructure, and you avoid per-image charges entirely.
• Large ecosystem: Years of community use have produced many LoRAs, ComfyUI workflows and tutorials built around FLUX.1 [dev]. Newcomers can usually find a ready-made workflow for the task they have in mind.

Cons

• Dev licence limits business use: FLUX.1 [dev] and Kontext [dev] weights fall under a non-commercial licence. Commercial self-hosting needs a paid licence from Black Forest Labs, and current self-serve plans list FLUX.2 models rather than FLUX.1.
• Previous-generation status: Black Forest Labs now groups FLUX.1 as its previous generation behind FLUX.2 and FLUX 3. New features and research effort go to the newer families, so FLUX.1 may fall behind on text rendering and realism.
• Some tools deprecated: FLUX.1 Depth and Canny are marked as deprecated on the official announcement page. Their open checkpoints remain on Hugging Face for reference, but the API versions are no longer supported.
• Technical setup required: Running FLUX.1 locally means handling a 12-billion-parameter model, GPU memory and tools such as ComfyUI or Diffusers. There is no simple consumer app, so non-technical users usually rely on third-party platforms.
• Unclear training data: Black Forest Labs has not disclosed the image data used to train FLUX models, according to Wikipedia. Brands with strict provenance rules may prefer models trained on licensed datasets.

Who is Using FLUX.1?

• Developers and startups: Teams call FLUX.1 through the BFL API or partner hosts to add image generation to their own products.
• AI artists and hobbyists: Creators run dev or schnell locally in ComfyUI and train LoRAs for personal styles.
• Designers and marketers: Creative teams use Kontext and Fill to edit product shots and extend images for different formats.
• Researchers: Academic groups study the open dev and schnell weights for image generation and editing research.

Pricing

FLUX.1 combines free open weights with paid API access priced per image (official prices, October 2026).

One directory quotes FLUX Pro from $0.055 per image and describes schnell as MIT licensed, but the official docs list $0.04 for FLUX1.1 [pro] and the model card names Apache 2.0. Third-party hosts such as Replicate and fal.ai set their own rates. Treat these figures as a snapshot and confirm current costs on the BFL pricing page before committing a budget.

What Makes FLUX.1 Unique?

FLUX.1 gives you the same model family across every level of control, from a free Apache 2.0 checkpoint on your own GPU to managed API endpoints. Most rival image generators are either closed services or open models without an official paid tier, whereas Black Forest Labs offers both for the same architecture. The later Kontext models added in-context editing that keeps characters consistent across scenes, and open Kontext [dev] weights let researchers inspect that capability directly. Even as a previous generation, it remains a reference point because so many community workflows still build on FLUX.1 [dev].

Compatibilities and Integrations

• API: BFL API endpoints for FLUX1.1 [pro], Ultra, Raw, Kontext [pro] and [max], and Fill [pro], billed in credits worth $0.01 each.
• Partner hosts: Replicate, fal.ai, Together AI and other providers listed by Black Forest Labs host FLUX.1 models.
• Local tools: Hugging Face Diffusers, ComfyUI, Draw Things and DiffusionBee run the open dev and schnell weights.
• Hardware: Kontext [dev] has TensorRT variants in BF16, FP8 and FP4 for NVIDIA Blackwell GPUs, according to Black Forest Labs.

FLUX.1 Tutorials

Pick your route first: hosted API for simplicity or open weights for control.

1. Decide whether you need commercial use, then choose schnell, dev or a pro API model accordingly.
2. For the API, create a BFL account, buy credits and generate an API key in the dashboard.
3. For local use, accept the licence on Hugging Face and download the weights for dev or schnell.
4. Load the model in ComfyUI or Diffusers and test short, descriptive prompts at a few step counts.
5. Use Kontext or Fill for edits, and compare results with FLUX.2 before starting a new project.

How We Rated It

Summary

FLUX.1 suits developers, AI artists and researchers who want strong image quality with the option to self-host. Its main limitation is licensing, because only schnell is free for commercial self-hosting and dev needs a paid licence. It is also a previous-generation family now that FLUX.2 and FLUX 3 exist. Start with schnell or the $0.04 API models to test your use case, then compare against FLUX.2 before building anything long term.