What FLUX is
FLUX is a family of text-to-image models made by Black Forest Labs, a company based in Freiburg, Germany. Robin Rombach, Andreas Blattmann and Patrick Esser, who had worked at Stability AI, founded it in 2024. You write a prompt, a sentence or two that describes the picture, and FLUX draws it. Some FLUX models also edit an image you give them.
FLUX.1 arrived in August 2024 as three kinds: Schnell, Dev and Pro. Wikipedia dates FLUX.2 to 25 November 2025 and the FLUX.3 series to 23 July 2026, with a smaller FLUX.3 Klein following in October. Its table lists FLUX.2 Klein 4B, from January 2026, as a distilled model for few steps under Apache 2.0, and that is the one the free Space below runs.
There is no single FLUX app. The models run on the Black Forest Labs website, in tools such as ComfyUI, on Hugging Face and on partner platforms.
Use is freemium. Several models are free to download, and Black Forest Labs runs free demo Spaces on Hugging Face. Its own API charges by the picture with no plan, as the image generator lesson read its pricing page on 6 November 2026.
What FLUX is good for
- A free first picture in a browser: the FLUX.2 [klein] 4B Space makes a 1024 by 1024 picture from a short prompt, in 4 steps by default.
- Photorealistic scenes and people: Wikipedia reports that Ars Technica's 2024 test put FLUX.1 Dev and Pro close to DALL-E 3 on following the prompt and close to Midjourney 6 on photorealism.
- Words inside the picture: Wikipedia says FLUX.2 claimed better typography, and lists FLUX.2 Flex as the typography model, sold through an API.
- Editing a photo with words: FLUX.1 Kontext, from May 2025, takes both text and images. Adobe added Kontext Pro to Photoshop's generative fill in beta in September 2025.
- Editing and reference images: the FLUX.2 [klein] 4B Space has an optional Input image(s) section, and the FLUX 3 Image Klein Space takes up to 10 reference images.
- Running on your own computer: open-weights models run in ComfyUI, which the ComfyUI lesson sets up.
When to use FLUX, and when to pick another tool
| Your job | Use FLUX? | Or try |
|---|---|---|
| A free first picture with no sign-up | Yes. The FLUX.2 [klein] 4B Space opens in any browser | Gemini or ChatGPT, inside a chat you already use |
| A poster or sign with exact words | Possible, since FLUX.2 claimed better typography, but read every letter | Ideogram, which is built around readable words |
| Changing one thing in a photo | Yes, with a Kontext or Klein model that takes a picture and an instruction | Gemini or ChatGPT, which edit inside the chat |
| Pictures that stay on your computer | Yes, with open weights in ComfyUI. Check the license of the model first | Other open models in the same app |
| A short video clip | Not in the free Space. FLUX 3 Video is a Black Forest Labs product, which its launch video lists at up to 20 seconds | Kling or Hailuo AI |
| A written answer or code | No | ChatGPT, Claude or Gemini |
How to use FLUX free in your browser, step by step
These steps use the FLUX.2 [klein] 4B Space that Black Forest Labs runs on Hugging Face. Spaces change over time, so the steps stick to the parts that stay put: the Prompt box, the Run button and the settings under it.
- Open the Space in your browser: huggingface.co/spaces/black-forest-labs/FLUX.2-klein-4B. Its heading names FLUX.2 [Klein] 4B and Apache 2.0. Nothing to install, and you can use it signed out.
- If the page says the Space is starting, wait until the Prompt box appears.
- Click the Prompt box, the field that says "Enter your prompt", and type one or two sentences: the subject, the style and the light. The examples at the bottom of the page fill the box for you.
- Leave Mode on Distilled (4 steps), the default, for speed. Base (50 steps) switches the settings to 50 steps and takes longer.
- Press Run beside the Prompt box, or press Enter.
- Wait for the picture to appear on the right of the page. The Space queues for a shared GPU and loads the model, so a first run can take longer than the next ones. A progress bar shows while it works.
- Save the picture with the download button in the corner of the result, or right-click it and save the image.
- To change the next picture, open Advanced Settings and change one setting, such as Width or Height. Both default to 1024. Press Run again.
- To edit your own photo, open Input image(s) (optional), add the photo, and describe the change in the Prompt box, such as "change the sky to sunset orange". Width and Height follow the photo's shape.
Open the Space and click the Prompt box, the field that says "Enter your prompt". No account is needed.
Describe the picture in plain words, with any text it should show, then press Run.
Leave Mode on Distilled (4 steps) for a quick picture. Base (50 steps) takes longer.
Wait while it runs. The picture appears on the right, here the mug with its words spelled right.
Point at the picture and press the download arrow at its top right to save it.
Screenshots of Black Forest Labs' FLUX.2 [Klein] 4B Space on Hugging Face, used without an account, taken on 6 November 2026.
The tutorials at the top of this page show FLUX in other apps. If a name on your screen differs from the one here, follow the one you see.
An example to try
Paste this prompt into the Prompt box and press Run. It names the subject, the style, the colors and one short word to print.
A lighthouse on a green cliff at sunrise, flat illustration, soft pastel colors, a small wooden sign by the path that says "Open"- A lighthouse stands on a green cliff with a pale sunrise sky behind it.
- The style is a flat illustration in soft pastels, not a photo.
- A small sign by the path may show the word Open. If a letter is wrong, press Run again, since each run uses a new seed while Randomize seed is on.
- The picture is square, 1024 by 1024, unless you change Width and Height under Advanced Settings.
- To keep a result you like, open Advanced Settings and clear Randomize seed. The Seed slider holds the number the last picture used.
Tips for better results
- Say the subject first, then the style, the light and the shape. One or two sentences are enough to start.
- Change one thing per run. If you change the subject and the style together, you cannot tell which one fixed the picture.
- Keep Mode on Distilled (4 steps) while you test wording. Choosing Base (50 steps) sets 50 steps and a higher Guidance scale, so each picture takes longer.
- Prompt Upsampling in Advanced Settings asks a vision-language model to rewrite your prompt. It is off by default, so leave it off while you test your own wording.
- Put words you want printed in double quotes, and keep them short.
- Write prompts in English first. Wikipedia, citing a 2025 Gigazine test of FLUX.1 Kontext, says it understood English well and Japanese poorly.
- For many reference images, open the FLUX 3 Image Klein Space from the same Black Forest Labs account. Its page says it takes up to 10.
- Sign in to a free Hugging Face account before a long session. Its daily GPU quota is larger than a signed-out visitor's.
What to know before you rely on it
- The free Space shares a GPU. Hugging Face's ZeroGPU page gives signed-out visitors 2 minutes of GPU time a day and free accounts 5 minutes, and less quota left means a lower place in the queue.
- Treat the Space as a public demo. Keep private text and client material out of it.
- License depends on the model. Hugging Face's model cards list FLUX.1 [schnell] and FLUX.2 [klein] 4B under Apache 2.0, FLUX.1 [dev] and FLUX.2 [dev] under non-commercial licenses, and FLUX 3 Image Klein under the Flux Kommunity License.
- The image generator lesson reads the Klein license as allowing commercial output only below a yearly revenue limit. Read the license itself before a picture goes into an ad or onto a product.
- Wikipedia notes that Black Forest Labs has not published the data it trained on. Check the terms of the model you use.
- Model names move fast. If the Space or a model name has changed, follow the names on the Black Forest Labs Hugging Face page.
- A picture of a real person needs that person's consent.
Learn more about FLUX
- 1Best AI image generator in 2026: which one to use for each jobWhich AI image generator to use: ChatGPT, Gemini, Adobe Firefly, Ideogram, FLUX, Canva and Recraft compared on free plans, prices, limits and rights.
- 1AI image prompts: how to get the picture you meantAI image prompts that work: say what the picture is for, name the subject, setting, style, framing and light, quote any words, then refine one change at a time.
- 4ComfyUI tutorial: install Comfy Desktop and make your first imageA ComfyUI tutorial for beginners: install Comfy Desktop, open the text-to-image workflow, download its model, make a first image and learn the nodes.
- 4Hugging Face Spaces and Gradio tutorial: build a demo and host itA Gradio tutorial for Hugging Face Spaces: build a demo with gr.Interface, run it locally, then host it on a Space with its requirements, secrets and ZeroGPU.
Frequently asked questions
Is FLUX free?
Partly. Black Forest Labs' FLUX.2 [klein] 4B Space on Hugging Face is free to open within Hugging Face's daily GPU limits, and several FLUX models are free to download. Its own API charges by the picture, so the free routes are the Space or your own computer.
Who makes FLUX?
Black Forest Labs, a company based in Freiburg, Germany. Robin Rombach, Andreas Blattmann and Patrick Esser founded it in 2024 after working at Stability AI, and FLUX.1 came out in August 2024.
Can I use FLUX pictures commercially?
It depends on the model. FLUX.1 [schnell] is under Apache 2.0 on its Hugging Face card, while FLUX.1 [dev] has a non-commercial license. Wikipedia says a commercial license for Dev can be had from Black Forest Labs. Read the license of the exact model.
Which FLUX should I start with?
FLUX.2 [klein] 4B in its Space is the quickest start, with a Prompt box, a Run button and an Apache 2.0 license. The newest open image model is FLUX 3 Image Klein from October 2026, which has its own Space with a Generate button and more options.
Can FLUX put readable words in a picture?
Often, for short words. Wikipedia says FLUX.2 claimed improved typography, and Black Forest Labs sells FLUX.2 Flex as its typography model. Read each letter before you share a picture, and try Ideogram if words are the whole job.
Can I run FLUX on my own computer?
Yes, for the models with open weights. Wikipedia lists ComfyUI, Recraft Studio and Stable Diffusion WebUI Forge as ways to run them, and the ComfyUI lesson installs the first. Your graphics card needs enough memory for the model you pick.
Why is my first picture slow?
The Space waits for a shared GPU before it draws. Hugging Face's ZeroGPU page says remaining quota decides your place in the queue, so a signed-out visitor waits longer than a signed-in one.
Sources
- Wikipedia: Flux (text-to-image model) https://en.wikipedia.org/wiki/Flux_(text-to-image_model)
- Hugging Face Space: black-forest-labs/FLUX.2-klein-4B https://huggingface.co/spaces/black-forest-labs/FLUX.2-klein-4B
- Hugging Face: black-forest-labs/FLUX.1-schnell https://huggingface.co/black-forest-labs/FLUX.1-schnell
- Hugging Face: black-forest-labs/FLUX.1-dev https://huggingface.co/black-forest-labs/FLUX.1-dev
- Hugging Face: black-forest-labs/FLUX.2-dev https://huggingface.co/black-forest-labs/FLUX.2-dev
- Hugging Face: black-forest-labs/FLUX.2-klein-4B https://huggingface.co/black-forest-labs/FLUX.2-klein-4B
- Hugging Face: black-forest-labs/flux-3-image-klein https://huggingface.co/black-forest-labs/flux-3-image-klein
- Hugging Face Space: black-forest-labs/flux-3-image-klein https://huggingface.co/spaces/black-forest-labs/flux-3-image-klein
- Hugging Face Hub: Spaces ZeroGPU https://huggingface.co/docs/hub/spaces-zerogpu
- Black Forest Labs: FLUX pricing https://bfl.ai/pricing
- Black Forest Labs on YouTube: Introducing FLUX 3 https://www.youtube.com/watch?v=uqQ_wvhAlJY
- Black Forest Labs on YouTube: FLUX.2, Frontier Visual Intelligence https://www.youtube.com/watch?v=E-LUKgMaLpA
- pixaroma on YouTube: Flux 3 Klein in ComfyUI https://www.youtube.com/watch?v=ntxxhO9H3bY
- Dan - Smart Tutorials on YouTube: Flux AI Tutorial https://www.youtube.com/watch?v=D1tcNzot2HI