The short answer
FLUX is a family of image and video models from Black Forest Labs, a German AI company. Its newest image model, FLUX 3 Image, reached the API on 1 October 2026 and adds bounding boxes for exact layouts, edits with up to ten reference images and output up to 4K, although custom sizes are capped at 4 megapixels when you use references. You can use FLUX in the BFL Playground or API on a pay-as-you-go basis, or run FLUX 3 Image Klein's open weights yourself.
| Made by | Black Forest Labs, Germany |
|---|---|
| Free plan | No free hosted plan; pay as you go. Open-weight models are free to download |
| Models | FLUX 3 Image, FLUX 3 Image Klein, FLUX 3 Video, FLUX 3 Max, FLUX.2 |
| Best at | Photorealistic images, precise layouts and edits, developers adding images to products |
| Shows sources | No, though FLUX 3 Image can search the web to get real things right |
| Team plans | Enterprise contracts and self-hosting licences |
| Your data | Inputs and outputs may train models unless you email an opt-out; enterprise contracts differ |
What FLUX is in November 2026
FLUX is a family of image and video models from Black Forest Labs, an AI company with labs in Freiburg, Germany, and San Francisco. Black Forest Labs says its founding team includes pioneers of Latent Diffusion and Stable Diffusion, and TechCrunch describes its co-founders Robin Rombach, Patrick Esser and Andreas Blattmann as former researchers who helped create Stability AI's Stable Diffusion models. Unlike Midjourney or ChatGPT, FLUX is mainly sold to people who build with it: developers call it through an API, companies license it to run on their own servers, and companies such as Envato build their own products on it. You can also try it directly in the BFL Playground in your browser.
The current generation is FLUX 3, announced on 23 July 2026 as one model family covering images, video, audio and robot control. The pieces arrived in stages. FLUX 3 Video came first, with clips of up to 20 seconds and native sound, and gained 2K and 4K output on 10 September. FLUX 3 Image reached the API on 1 October 2026, and FLUX 3 Image Klein, a smaller version with open weights, followed on 15 October. On 22 October BFL added exact output sizes and more accurate single-image edits.
The previous generation, FLUX.2, is still available in [max], [pro], [flex] and [klein] versions, and BFL describes FLUX.2 [flex] as specialised for typography and small details.
Where FLUX is genuinely strong
Photorealistic images
BFL's own guides focus heavily on realistic photography, with lens and lighting language for macro shots, cinematic outdoor scenes and documentary photos. In our view, if you need an image that could pass for a real photo shoot, FLUX is one of the first tools to try. Describe the subject, setting, light and framing, and the model renders the whole scene.
Control over layout
The standout feature of FLUX 3 Image is bounding boxes. You add a list of boxes to the end of your prompt to say exactly where each element goes, such as the headline at the top, the product on the right and the logo in the corner. The same boxes let you edit one element, recolour it, move it or remove it, while everything else in the image stays exactly as it was. For marketing teams who need many variations of one approved layout, this is a real time saver.
Editing with references
FLUX 3 Image accepts up to ten reference images in one request. You refer to each by its position, for example "put the bottle from image 2 on the table in image 1". That makes it practical for product placement, consistent characters and restyling a photo without starting over.
Getting real things right
By default FLUX 3 Image searches the web before it draws, so a prompt about a real landmark, event or product is more likely to come out accurately. You can switch this grounding off for faster results that use only your prompt.
Open weights
FLUX 3 Image Klein has 5.6 billion parameters, supports the same features as FLUX 3 Image and runs on consumer graphics cards down to an RTX 3090. Its weights are on Hugging Face under the FLUX Kommunity License. If your company cannot send images to an outside service, this is the route.
Where FLUX falls short
- No free plan for the hosted service. The Playground and API are pay as you go. You buy credits, where one credit is one US cent, and the Playground costs the same as the API.
- Built for builders. There is no chat assistant that helps you refine an idea, as there is in ChatGPT or Gemini. You write prompts, and for the best results you learn a little about the API settings.
- Licences need care. The open weights are free to download, but commercial self-hosting is sold through BFL licences such as Builder, Platform and Professional. Read the terms before you ship anything.
- Training on your inputs. See the privacy section below.
How to get better images from FLUX
- Write like a photographer. Describe the subject first, then the setting, the light, the camera angle and the lens. "Glass of lemonade on a marble counter, morning window light, shallow depth of field" beats "nice lemonade picture".
- Say what you want, not what you do not want. BFL's prompting guide recommends replacing negative prompts with positive descriptions.
- Use hex codes for brand colours. BFL recommends hex colour prompting for precise colour matching, which helps keep campaign images on brand, but check the result.
- Put text in quotation marks. When an image needs words, such as a poster headline, quote the exact text, as BFL's editing guide recommends, so the model renders it as written.
- Use boxes for layouts. If the position of elements matters, add bounding boxes instead of describing positions in words.
- Preview small, finish large. Generate at 768 by 768 to explore, then render the version you like at 2K or 4K (up to 4 megapixels if you use reference images). Prices are set by resolution tier, from 768 by 768 up to 4K.
- With Klein, keep the prompt upsampler on. BFL warns that FLUX 3 Image Klein gives noticeably weaker results without it.
FLUX privacy settings and your images
Black Forest Labs' privacy policy, revised on 1 August 2026, says it may use your prompts, uploaded images, tasks and generated outputs to train and improve its models, after de-identification. There is no switch in the Playground. To opt out, email privacy@blackforestlabs.ai with the subject line "Training Opt Out" and your account details. Customers on enterprise contracts are covered by their own agreements instead, and BFL lists ISO 27001:2022 and SOC 2 Type II certification.
If you work with unreleased products, client photos or images of real people, send the opt-out email before your first upload, negotiate an enterprise contract, or run the open weights on your own hardware.
Which way of using FLUX suits whom
- Marketers and designers: the Playground. Buy a small amount of credit, test FLUX 3 Image on a real brief and compare it with your current tool.
- Developers: the API, starting with FLUX 3 Image for quality or FLUX 3 Image Klein for low cost per image.
- Companies with strict data rules: the open weights with a self-hosting licence, or an enterprise contract.
- Not for: casual users who want a free, chat-style image maker. ChatGPT, Gemini or Canva are easier starting points.
Check BFL's pricing page for current rates. To compare FLUX with the best-known alternative, read Midjourney vs FLUX, see our pick of the best AI for images, or start with Learn FLUX.
Prompts to copy
Make a realistic product photo
A glass of fresh lemonade on a white marble counter, morning window light from the left, shallow depth of field, 50mm lens, a sprig of mint beside the glass, photorealistic, background in #1E4FD8.
You get: A studio-quality product shot in your brand colour
Edit one thing and keep the rest
Change the background of image 1 to cobalt blue (#0047AB). Keep the product, its label, its shadow and the lighting exactly as they are.
You get: A clean background swap without touching the product
Place a product into a scene
Place the bottle from image 2 on the wooden table in image 1, standing upright near the window. Match the warm afternoon light and add a soft shadow on the table.
You get: A believable composite from two references
Design a poster with exact text
A minimalist event poster, bold sans-serif headline "SUMMER SPRINT 2027" at the top, a single lemon slice centred on a bright yellow background, small text "Berlin, 12 June" at the bottom, flat graphic style.
You get: A poster with the words rendered as written
Create a consistent campaign set
Using image 1 as the style reference, create a new scene in the same colours, lighting and mood: [describe new scene]. Keep the same camera angle and the same product size in frame.
You get: A new image that matches your approved look
Generate catalogue images with a script (API)
Write a Python script that reads [catalogue.csv] and, for each product, sends a request to the FLUX 3 Image API with this prompt template: "[product name] on a [surface], soft studio light, 3/4 angle, plain [colour] background, photorealistic". Read the API key from an environment variable, wait for each result, and save each image with the product ID as the file name.
You get: A script for ChatGPT, Claude or a coding agent that produces a full set of catalogue images through the FLUX API