CheaperGen
Black Forest LabsOverview

FLUX.2 [klein] 4B AI

FLUX.2 [klein] 4B is Black Forest Labs' small, open (Apache-2.0) FLUX.2 model: sub-second text-to-image and editing with up to 4 reference images, at a fraction of a cent per image. Released January 2026. Try it online in the Studio or call it from code, up to 93% below the official price.

Developer
Black Forest Labs
FLUX family
Released
January 2026
Sizes
up to 2K
2 sizes
From
$0.001/image
1K

Try it

Try FLUX.2 [klein] 4B AI now

Runs on your balance at the price on the button; failed runs are refunded. Results are saved to your account.

Open the full-screen Studio

What it is

What is FLUX.2 [klein] 4B?

Real-time FLUX

klein 4B distills FLUX.2 into 4 steps, so images come back in well under a second — for live previews, batch jobs and interactive apps.

Edits too

Give it up to 4 reference images and describe the change; the output follows the input image.

Capabilities

What FLUX.2 [klein] 4B can do

Sub-second

4-step distilled

Fraction of a cent

Per 1 MP image

4 references

Image editing

Open weights

Apache-2.0

Examples

FLUX.2 [klein] 4B examples

Prompts you can load into the Studio and try.

  • Flat-lay of breakfast on a marble table, top-down, bright daylight

    Try it
  • Image Edit

    Make the jacket in image 1 bright yellow, keep everything else

    Try it

Use cases

What people use FLUX.2 [klein] 4B for

Images from a description
Illustrations, product visuals and concepts, up to 2K.
Editing your images
Change backgrounds, colours or outfits, or combine up to 4 images in one edit.
Any format
7 aspect ratios, from banners to vertical posts.

Versions

FLUX.2 [klein] 4B in the FLUX family

Every FLUX model we offer, newest first. FLUX AI covers the whole family.

FLUX models
ModelReleasedMax sizeFrom
FLUX 3 Image KleinOctober 20264K$0.0011/image
FLUX 3October 20264K$0.0317/image
FLUX 3 VideoAugust 2026720p · 5–20 s$0.1564/s
FLUX.2 [klein] 4BJanuary 20262K$0.001/image
FLUX.2 [max]December 20254 MP$0.0644/image
FLUX.2 [pro]November 20254 MP$0.0276/image
FLUX.2 [flex]November 20254 MP$0.046/image
FLUX.1 Kontext [pro]May 20251K$0.0348/image
FLUX.1 Kontext [max]May 20251K$0.0695/image

How it compares

FLUX.2 [klein] 4B vs similar models

FLUX.2 [klein] 4B vs FLUX.2 [pro]

FLUX.2 [klein] 4B

  • Lower price: $0.001/image vs $0.0276/image at 1K (97% less).
  • Higher maximum resolution: 2K vs 4 MP.
  • Newer model (released January 15, 2026).

FLUX.2 [pro]

  • Edits with up to 8 input images (vs 4).
  • More aspect ratios: 10 vs 7.

Where to use it

Three ways to use FLUX.2 [klein] 4B

In the Studio

No code: write a prompt, pick the settings, generate. The price is shown before you run.

Through the API

One key and one request shape for every model; poll the generation or get a webhook.

From an AI agent

Our MCP server lets Claude and other agents quote and generate for you.

FAQ

FLUX.2 [klein] 4B AI questions

FLUX.2 [klein] 4B is Black Forest Labs' small, open (Apache-2.0) FLUX.2 model: sub-second text-to-image and editing with up to 4 reference images, at a fraction of a cent per image. It was released in January 2026.

Black Forest Labs. FLUX.2 [klein] 4B is part of Black Forest Labs' FLUX family — see the FLUX AI overview.

Yes, in the CheaperGen Studio — no code and no subscription needed. A 1K image costs $0.001, up to 93% below the official price. Step by step: [how to use FLUX.2 [klein] 4B](/models/bfl/flux-2-klein-4b/how-to-use).

1K and 2K, and it edits images (up to 4 inputs).

FLUX.2 [klein] 4B: Lower price: $0.001/image vs $0.0276/image at 1K (97% less). Higher maximum resolution: 2K vs 4 MP. FLUX.2 [pro]: Edits with up to 8 input images (vs 4). More aspect ratios: 10 vs 7. Full comparison: [FLUX.2 [klein] 4B vs FLUX.2 [pro]](/compare/flux.2-klein-4b-vs-flux.2-pro).

Yes: one API key, POST /v1/run/{endpoint}, from $0.001/image. See the [FLUX.2 [klein] 4B API](/models/bfl/flux-2-klein-4b) page and its documentation.

FLUX.2 [klein] 4B is Black Forest Labs' open-weights model. CheaperGen provides API access to the open weights, run on GPU inference for open-source models, and does not own the model. Trademarks belong to their respective owners.