Flux

Flux is a family of text-to-image AI models developed by Black Forest Labs (BFL), a frontier AI lab founded in 2024 by former Stability AI researchers. Text-to-image means the model generates visual output from a natural language description — you write a prompt, and Flux produces an image. Flux has become the leading open-weights image generation model family, with production integrations at Adobe, Microsoft, Freepik, and Picsart, and is widely used by developers building image generation into their own products.

How Flux works

Flux is built on latent diffusion — a technique where the model learns to generate images by training on the relationship between text descriptions and visual content, then working backwards from noise to a coherent image at inference time. Here is how the pieces fit together.

• Prompt encoding: Your text prompt is converted into a numerical representation using a text encoder. Flux uses a hybrid architecture with both a standard text encoder and a T5 model, which gives it stronger language understanding and better prompt adherence than earlier diffusion models.
• Latent denoising: The model generates a compressed representation of the image in latent space — a lower-dimensional mathematical space — rather than working directly at pixel level. This makes generation faster and more computationally efficient.
• Decoding to pixels: Once the latent image is refined, a decoder converts it back into a full-resolution image you can see and use.
• Model variants: BFL offers several tiers. FLUX.2 [max] is the highest quality, slowest, most expensive option. FLUX.2 [klein] is optimised for speed and interactive use. Open weights versions like FLUX.1 [dev] and FLUX.1 [schnell] can be self-hosted and fine-tuned. FLUX Kontext adds image editing capabilities on top of generation.

You access the closed models through the BFL API, which charges per image. Open weights models can be downloaded from Hugging Face and deployed on your own GPU infrastructure.

What you can build with Flux

• Product image generation for e-commerce: Generate photorealistic product shots on different backgrounds, surfaces, or in lifestyle contexts without a photo shoot. FLUX Kontext lets you take an existing product image and place it in new settings automatically.
• Character-consistent asset pipelines: Flux maintains visual identity across multiple generations of the same character — useful for game studios, animation teams, and content creators who need a consistent person or creature across many images.
• Virtual try-on systems: FLUX VTO lets fashion and retail teams generate images of a garment on different body types from a single catalogue photo, dramatically reducing the cost of showing sizing diversity.
• Custom fine-tuned image models: Developers fine-tune FLUX.1 [dev] on their own brand assets, product photography, or illustration style using the open weights, then deploy a model that generates on-brand images reliably at scale.
• Image editing pipelines: Using FLUX Kontext, teams build automated pipelines that take product images and apply consistent edits — background replacement, lighting adjustments, localised changes — without manual editing work.
• In-app AI generation features: Companies like Gamma and OpenArt embed Flux via the BFL API to offer image generation directly inside their products, without training or hosting a model themselves.