Understand FLUX.2 before you write a prompt.
FLUX.2 is Black Forest Labs’ image generation and editing family — several hosted tiers plus open weights — built around multi-reference consistency, legible text and photoreal detail at up to 4 MP.
Capabilities and specifications on this page come from Black Forest Labs’ official FLUX.2 pages. Pricing, quotas and regional availability are set by the vendor and change without notice.
What FLUX.2 renders
These are Black Forest Labs’ own FLUX.2 examples, taken from the official model page. Together they cover what the model is sold on: reference-driven composition, photoreal people, readable text and product-grade detail.
Example images belong to Black Forest Labs and are reproduced with credit: FLUX.2 official model page
What FLUX.2 is
FLUX.2 is the current generation of Black Forest Labs’ image models. It is a family rather than one checkpoint: a hosted tier for production work, a controllable tier for typography and fine detail, a fast tier for iteration, and an open-weight tier you can run yourself.
A family, not one model
FLUX.2 [max], [pro], [flex], [dev] and [klein] trade quality, speed, control and licensing against one another. Choosing FLUX.2 means choosing a tier.
Consistency is the product
Multi-reference control keeps characters, products and styles stable across many generations — the property that makes an image model usable for a whole campaign rather than a single picture.
Text that survives production
Black Forest Labs documents complex typography, infographics and UI mockups as reliable outputs, which is what makes the model viable for ads, packaging and interface concepts.
Two ways to ship it
Call the hosted tiers through the API or the browser playground, or download the open-weight [dev] build and run it on your own hardware.
FLUX.2 variants at a glance
Black Forest Labs documents FLUX.2 as several tiers. The table summarises what the official pages say each one is for; it is vendor positioning, not a benchmark.
| Variant | What the vendor says | Choose it for |
|---|---|---|
| FLUX.2 [max] | Top-tier generation and editing, strongest prompt following, grounded generation with real-time web context. | Maximum quality when cost is not the constraint. |
| FLUX.2 [pro] | State-of-the-art quality at maximum speed; the production-grade generation and editing tier. | High-volume production where quality and turnaround both matter. |
| FLUX.2 [flex] | Maximum precision, adjustable steps and guidance, specialised for typography and small details. | Text-heavy layouts and fine control over the render. |
| FLUX.2 [klein] | Fast and efficient, tuned for rapid iteration and prototyping. | Drafts, exploration and speed-to-quality trade-offs. |
| FLUX.2 [dev] | Open weights: frontier performance you deploy and customise yourself. | Self-hosting, fine-tuning and research. |
Variant names and positioning are Black Forest Labs’ own. A third-party API may expose only some of them — kie.ai, for example, documents FLUX.2 [pro] and [flex] text-to-image and image-to-image endpoints.
What FLUX.2 does well
Every claim below is documented on the official FLUX.2 pages or in the published API reference. Figures the vendors do not publish — success rates, per-image cost — are left out rather than estimated.
Multi-reference control
Reference up to 10 images in one request, with the character consistency Black Forest Labs markets as its strongest.
Photoreal detail
Fabric, skin, architecture and lighting rendered close enough to photography that the vendor describes the gap with real images as closing.
Production-ready text
Complex typography, infographics and UI mockups render with legible small text and a layout that holds.
Colour and mark accuracy
Brand colours can be specified by hex value and logos reproduced exactly rather than approximately.
Spatial reasoning and world knowledge
Objects sit where the prompt puts them, lighting behaves plausibly, and scenes follow real-world logic.
Documented specifications
Compiled from Black Forest Labs’ official FLUX.2 pages and the published kie.ai API reference. Where a figure differs between the vendor and a reseller, both are shown.
- Developer
- Black Forest Labs
- Current family
- FLUX.2
- Variants
- [max] · [pro] · [flex] · [dev] · [klein]
- Maximum output
- Up to 4 MP
- Aspect ratios
- Any; the API documents 1:1, 4:3, 3:4, 16:9, 9:16, 3:2 and 2:3
- Reference images
- Up to 10 (kie.ai documents a maximum of 8)
- Prompt length
- 3–5000 characters (kie.ai); up to 32K text input tokens (vendor)
- Reported speed
- Sub-10-second generations, per the vendor
- API resolution
- 1K and 2K (kie.ai)
- Input images
- JPEG, PNG or WebP, up to 10 MB each (kie.ai)
- Open weights
- FLUX.2 [dev] only
- Pricing
- Not published here; set per vendor
Documented access channels
FLUX.2 is reachable in several ways, and they are not interchangeable: the hosted tiers and the open weights differ in licence, control and who you pay.
Official playground
Black Forest Labs hosts FLUX.2 in a browser playground, including the multi-reference and editing demos linked from the model page.
OpenOfficial API
The production route: call the hosted tiers from your own application. Black Forest Labs publishes pricing and rate limits on its own site.
OpenSelf-hosted open weights
The [dev] tier is the only one whose weights are published, so it is the only tier you can deploy on your own infrastructure.
FLUX.2 questions people search for
What is FLUX.2?
FLUX.2 is the current generation of image generation and editing models from Black Forest Labs, released as several hosted tiers plus an open-weight build.
Is FLUX.2 free?
Not as a whole. The open-weight [dev] build can be run on your own hardware and resellers advertise free trial credits, but the hosted tiers are paid and Black Forest Labs publishes its own pricing. This page deliberately quotes no prices because they change.
Which FLUX.2 model should I use?
Start with [pro] for production output, move to [flex] when typography or fine detail decides the result, use [klein] when iteration speed matters most, and choose [dev] when you need to self-host.
How many reference images can FLUX.2 take?
Black Forest Labs documents up to 10 references per request. The kie.ai API reference documents a maximum of 8 input images, so the ceiling depends on the channel you call.
Can FLUX.2 render readable text?
Yes — text rendering is one of the capabilities Black Forest Labs explicitly markets, alongside infographics and UI mockups. Small type is where image models usually fail, so test it with your own copy.
Is FLUX.2 open source?
Only the [dev] tier carries open weights. [max], [pro], [flex] and [klein] are hosted models, so their weights are not published.
Is FLUX.2 better than GPT Image 2.5?
Neither vendor publishes a head-to-head comparison, so any ranking is inference. The concrete difference is delivery: FLUX.2 offers an open-weight tier and adjustable controls, while GPT Image 2.5 is API-only. Our comparison article covers the documented differences.
Latest FLUX 2 Pro articles
Browse every published FLUX 2 Pro article, from introductions to practical guides and developer documentation.
-
FLUX.2 Prompt Guide: The Structure Black Forest Labs Documents
The four-part FLUX.2 prompt structure Black Forest Labs documents, its word-order and length rules, text-rendering guidance, and two published prompts.
Read article -
FLUX.2 vs Nano Banana Pro: What Each Vendor Actually Publishes
FLUX.2 and Nano Banana Pro compared only on what Black Forest Labs and Google publish, row by row, with the gaps each vendor leaves open.
Read article -
How to Use FLUX AI: A FLUX.2 Getting-Started Guide
Where Black Forest Labs runs FLUX.2, how a first image comes together, one worked prompt with references and hex colours, and what the vendor leaves out.
Read article -
FLUX.2 API: Getting Started with Black Forest Labs
How to get a Black Forest Labs key, call api.bfl.ai with a working curl and Node script, and read the FLUX.2 limits the official pages actually publish.
Read article