FLUX is the family of AI image models from Black Forest Labs (BFL), the German lab founded by the researchers behind Stable Diffusion. As of September 2026, FLUX.2 is the production line-up for image generation and editing, from the open, sub-second [klein] models to the web-grounded [max]. The newer FLUX 3 adds video with audio. It is a strong, affordable choice, but paid access is aimed mostly at developers.
Last updated: September 2026
What is FLUX?
FLUX is built by Black Forest Labs, founded in 2024 in Freiburg, Germany, by Robin Rombach (CEO), Andreas Blattmann and Patrick Esser. All three previously worked at Stability AI and helped create the Stable Diffusion models. BFL also has an office in San Francisco. In December 2025 it raised a $300 million Series B at a $3.25 billion valuation, led by Salesforce Ventures and Anjney Midha (AMP), with a16z, NVIDIA, General Catalyst, Canva and Figma Ventures among its investors.
The model line-up has changed a lot since this page first covered FLUX.1 Pro in August 2024:
- FLUX.1 (August 2024): [pro] via API, [dev] as non-commercial open weights and [schnell] under Apache 2.0. FLUX1.1 [pro] followed in October 2024, and Ultra and Raw modes in November 2024.
- FLUX.1 Kontext (May 2025): models that generate and edit images from both text and a reference image, in [pro], [max] and open-weight [dev] versions.
- FLUX.2 (November 2025): the current image family. [pro], [flex] and the 32-billion-parameter open-weight [dev] launched on 25 November 2025. [max] followed in December 2025, and [klein] arrived on 15 January 2026.
- FLUX 3 (July 2026): a multimodal model trained jointly on images, video and audio. FLUX 3 Video became generally available in August 2026, and the open-weight FLUX 3 Action robotics model was released in September 2026.
BFL’s documentation describes FLUX.2 as “fully supported for production image generation and editing.” It still lists the FLUX.1 models as a previous generation.
Key features of FLUX
Multi-reference generation and editing
FLUX.2 generates and edits images in one model, keeping characters, products and styles consistent without fine-tuning. According to BFL’s documentation, FLUX.2 [pro], [flex] and [max] accept up to 8 reference images through the API and 10 in the Playground. [klein] accepts up to 4.
Legible text and typography
BFL says FLUX.2 handles “complex typography, infographics, memes and UI mockups with legible fine text.” FLUX.2 [flex] is the variant tuned for text and fine detail, with adjustable steps and guidance. Microsoft promotes it on its Foundry platform for text-heavy design and UI prototyping.
High resolution and precise control
- Output up to 4 megapixels, in any aspect ratio
- Exact brand colours through hex codes in the prompt
- Structured JSON prompts for detailed control over a scene
- Prompts up to 32K text tokens
Grounded generation in FLUX.2 [max]
FLUX.2 [max] is BFL’s highest-quality image model. It can search the web for real-time context, so it can show current products or events without you supplying reference photos. BFL says it has the strongest prompt following and editing consistency in the FLUX.2 family.
Sub-second models you can run locally
FLUX.2 [klein] comes in 4B and 9B sizes. BFL says it generates or edits images in under a second on modern hardware, and the 4B model fits in about 13 GB of VRAM, so it runs on consumer graphics cards such as the RTX 3090 or 4070 and above. FP8 and NVFP4 versions cut memory use further.
FLUX 3 Video with native audio
FLUX 3 Video turns text or an image into clips of up to 20 seconds, with sound and lip-synced dialogue generated together with the picture. BFL’s launch post describes HD (720p) output with Full HD (1080p) through upscaling, and its API price list now also includes QHD and UHD tiers. BFL’s FLUX Tools add video upscaling and editing, billed separately.
Open weights
BFL publishes weights on Hugging Face, where FLUX.2 [dev], the [klein] models and FLUX.1 Kontext [dev] are available. This lets teams self-host, fine-tune and keep images on their own servers.
Benefits of using FLUX
- Low cost per image: hosted FLUX.2 starts at about 1.4 cents per image with [klein] and 3 cents with [pro].
- Choice of deployment: BFL’s API, partner platforms and creative apps, or your own hardware.
- Consistency for campaigns: multi-reference editing keeps a product or character the same across many images.
- Enterprise credentials: BFL lists ISO 27001:2022 and SOC 2 Type II certification on its website.
- One vendor for image and video: FLUX 3 Video is billed through the same API and credit system as FLUX.2.
FLUX use cases by role
- Marketers and e-commerce teams: product shots, retexturing and seasonal variations with consistent products and exact brand colours.
- Designers: posters, infographics and UI mockups with readable text, using [flex] or FLUX.2 inside Adobe Photoshop.
- Developers and product teams: image features in apps through the BFL API, or [klein] for fast, low-cost generation.
- Content creators and social media managers: short videos with sound from FLUX 3 Video.
- Researchers and enterprises with strict data rules: self-hosted open weights on private infrastructure.
How much does FLUX cost?
BFL has no subscriptions or seat fees on its API: you buy credits (1 credit = $0.01) and pay per image or per second of video. For FLUX.2, the first megapixel costs a flat rate and each extra megapixel adds to the price, so the figures below are starting prices for a roughly 1-megapixel image. Prices are in USD, as of September 2026, from BFL’s pricing documentation.
| Model | API price | Access |
|---|---|---|
| FLUX.2 [klein] 4B | from $0.014 per image | API, or free open weights (Apache 2.0) |
| FLUX.2 [klein] 9B | from $0.015 per image | API, or open weights (non-commercial licence) |
| FLUX.2 [pro] | from $0.03 (generation), from $0.045 (editing) | API only |
| FLUX.2 [flex] | from $0.05 per image | API only |
| FLUX.2 [max] | from $0.07 per image | API only |
| FLUX.2 [dev] | Free to download | Open weights (non-commercial licence) |
| FLUX.1 Kontext [pro] / [max] | $0.04 / $0.08 per image | API (previous generation) |
| FLUX1.1 [pro] / [pro] Ultra | $0.04 / $0.06 per image | API (previous generation) |
| FLUX 3 Video | $0.06/s (draft), $0.17/s (HD), $0.29/s (Full HD), up to $0.80/s (UHD) | API |
The BFL Playground gives 100 free credits to start, with no credit card required. Partner platforms set their own prices: fal, for example, charges $0.03 for the first megapixel of FLUX.2 [pro] and $0.015 for each extra megapixel. For commercial self-hosting of the non-commercial open models, BFL sells licence tiers with monthly image allowances through sales or its dashboard.
FLUX pros and cons
Pros
- Low per-image API prices, with FLUX.2 [pro] from $0.03
- Strong multi-reference editing and consistency across images
- Readable text and exact hex colour control
- Open weights you can self-host, including an Apache 2.0 model
- Widely available: Adobe, Krea, fal, Replicate, Microsoft Foundry and more
- Image and video on one API with pay-as-you-go billing
Cons
- No consumer subscription app: BFL’s own option is a pay-per-image Playground, so most non-coders use FLUX through partner tools
- Licences differ by model, and several open models are non-commercial
- FLUX 3 Image and the open FLUX 3 Dev are announced, but we found no general release as of late September 2026
- Per-megapixel pricing makes large or multi-reference jobs harder to estimate
- Frequent new versions and name changes make older guides and tutorials out of date
How to use FLUX
You do not need to code. The simplest routes are the BFL Playground or an app you already use.
- Try the Playground: go to playground.bfl.ai, create an account and use the 100 free credits. No credit card is needed.
- Or use FLUX inside another app: Adobe added FLUX.2 to Adobe Firefly (Text to Image, Prompt to Edit and Boards) and to Generative Fill in Photoshop on the desktop in December 2025. Krea and Freepik’s Magnific also offer FLUX.2, and Krea offers FLUX 3 Video. Mistral’s Le Chat uses Flux Pro and Ultra models for image generation.
- Write a specific prompt: describe the subject, setting, lighting, camera angle and style. Put any text you want in the image inside quotation marks and add hex codes for exact colours.
- Add reference images: upload product or character photos to keep them consistent, or upload an image and describe the change you want.
- Pick the model: [klein] for fast drafts, [pro] for most work, [flex] for text-heavy designs and [max] for the best quality or current-events context.
- For developers: get an API key from the BFL dashboard, or call FLUX through fal, Replicate or Microsoft Foundry. To self-host, download weights from Hugging Face and check the licence first.
FLUX alternatives
- Midjourney: pick it for bold, stylized, artistic images. V8.2 became its default model on 24 July 2026.
- Nano Banana: pick Google’s Nano Banana 2 (February 2026) if you want image generation and editing built into the Gemini app, Google Search and Google Ads.
- GPT Image: pick OpenAI’s model if you prefer to create and edit images by chatting in ChatGPT.
- Ideogram: pick Ideogram 4.0 for dense text and layout-driven design work. It is also an open-weight model.
Frequently asked questions
Is FLUX free to use?
Partly. The open-weight models, including FLUX.2 [klein] 4B (Apache 2.0) and FLUX.1 [schnell], can be downloaded and run for free, even commercially. FLUX.2 [dev] and [klein] 9B are free only for non-commercial use. The BFL Playground gives new users 100 free credits with no credit card, after which you pay per image. Hosted models such as FLUX.2 [pro] and [max] are pay-as-you-go.
Is FLUX.1 Pro still available?
The FLUX.1 generation is still listed on Black Forest Labs’ pricing page as a previous generation: FLUX1.1 [pro] costs $0.04 per image, FLUX1.1 [pro] Ultra and Raw $0.06, and FLUX.1 Kontext [pro] $0.04. For new projects, BFL’s documentation points users to FLUX.2, which it describes as fully supported for production image generation and editing.
What is the difference between FLUX.2 [pro], [max], [flex] and [klein]?
[max] is the highest-quality model and the only one with grounded generation that searches the web for current context. [pro] balances quality, speed and price for production work. [flex] exposes steps and guidance controls and is tuned for typography and fine detail. [klein] is the small, fast family (4B and 9B) that generates images in under a second and can run locally.
Can FLUX generate video?
Yes, through FLUX 3. BFL made FLUX 3 Video generally available on its API in August 2026 for text-to-video and image-to-video, with clips up to 20 seconds and native audio. It is billed per second of output, from $0.06 per second in draft mode. FLUX 3 Video is also live on partner platforms such as Krea.
Is FLUX 3 available for images yet?
Not generally, as far as we could verify. BFL announced FLUX 3 in July 2026 and said FLUX 3 Image would follow in early access. BFL’s August video post still lists FLUX 3 Image on its roadmap, and Krea says FLUX 3 image generation is coming soon, running FLUX.2 in the meantime. For images today, use FLUX.2.
Can I use FLUX images commercially?
The hosted API models are positioned for commercial production work; review BFL’s terms for specifics. For open weights, check the licence. FLUX.2 [klein] 4B and FLUX.1 [schnell] are Apache 2.0. FLUX.2 [dev], [klein] 9B and FLUX.1 Kontext [dev] use non-commercial licences, though the FLUX.2 [dev] and Kontext [dev] model cards say generated outputs may be used commercially. Running those models commercially requires a BFL licence.
Who owns FLUX?
FLUX is made by Black Forest Labs, a German AI lab founded in 2024 in Freiburg by Robin Rombach, Andreas Blattmann and Patrick Esser, former Stability AI researchers who helped create Stable Diffusion. In December 2025 it raised a $300 million Series B at a $3.25 billion valuation, led by Salesforce Ventures and Anjney Midha (AMP).
Verdict
FLUX has grown from a single 2024 model into a broad visual AI line-up covering images, video and robotics. FLUX.2 gives strong image quality, multi-reference editing and readable text at low per-image prices, with open weights for teams that need to self-host. FLUX 3 adds video with native audio. Developers and businesses building image features should shortlist it. Non-technical users will usually get more from FLUX inside Adobe, Krea or Magnific than from BFL’s own Playground, and should check each model’s licence before commercial self-hosting.
Sources
- FLUX API pricing (Black Forest Labs docs)
- Pricing and self-hosting licences (Black Forest Labs)
- FLUX.2 overview (Black Forest Labs docs)
- FLUX.2 models page (Black Forest Labs)
- FLUX.2 [max] model page (Black Forest Labs)
- FLUX.2: Frontier Visual Intelligence (BFL blog, Nov 2025)
- FLUX.2 [klein]: Towards Interactive Visual Intelligence (BFL blog, Jan 2026)
- FLUX.1 Kontext announcement (Black Forest Labs, May 2025)
- FLUX 3: Real World Models (BFL blog, Jul 2026)
- FLUX 3 Video, Part 1: Generation (BFL blog, Aug 2026)
- FLUX 3 Action (BFL blog, Sep 2026)
- Black Forest Labs blog index
- Black Forest Labs homepage
- FLUX Playground (Black Forest Labs)
- Black Forest Labs Documentation
- FLUX.2 [dev] model card (Hugging Face)
- FLUX.2 [klein] 4B model card (Hugging Face)
- FLUX.1 Kontext [dev] model card (Hugging Face)
- FLUX.1 [schnell] model card (Hugging Face)
- Black Forest Labs on Hugging Face
- Black Forest Labs unveils FLUX 3 (GlobeNewswire press release, Jul 2026)
- Black Forest Labs launches FLUX 3, in limited release to start (VentureBeat)
- Black Forest Labs makes FLUX 3 Video generally available (The Decoder)
- Black Forest Labs launches Flux.2 AI image models (VentureBeat)
- Black Forest Labs raises $300M at $3.25B valuation (TechCrunch)
- Flux (text-to-image model) (Wikipedia)
- Adobe Firefly adds new video tools and models, including FLUX.2 (Adobe Blog, Dec 2025)
- Image generation in Le Chat (Mistral Help Center)
- FLUX 3 on Krea
- Flux.2 is available on Magnific (Magnific blog)
- FLUX.2 [pro] on fal
- FLUX.2 [max] on Replicate
- FLUX.2 [flex] now available on Microsoft Foundry (Microsoft Community Hub)
- Midjourney Version documentation
- Introducing ChatGPT Images 2.0 (OpenAI)
- Nano Banana 2 (Google blog, Feb 2026)
- Ideogram 4.0 (Ideogram)
