Use Black Forest Labs' fast FLUX.2 Klein 4B and 9B in your browser, with no GPU and nothing to download. Draft from a prompt or edit your own photos on Oeneo AI, and see the credit cost before every run.
Klein generates and edits in one model. With no photo attached it works from your prompt alone; attach 1 to 4 photos and the same prompt box edits them instead.
Text to image
Set the image count to the maximum and one run returns several takes on the same brief, so you can compare layout, viewpoint and color before you commit. Keep the strongest one and attach it as a reference to refine it.
Multi-reference edit
Use it to dress a person in a new outfit, place them somewhere new, or keep one character across several scenes by reusing the same portrait. Check faces and garment details against the originals, since similar items can swap features.
Single-photo edit
Upload one scan, name the damage to remove and the colors you expect, and say what must stay the same. Generative restoration can alter faces and small details, so compare the result with the original before you print or share it.
Style to photo
A rough sketch, an anime frame or a game screenshot sets the shape and angle; your prompt sets the materials, light and backdrop. Use artwork you made or have rights to, and expect fine lines and any lettering to change.
Both sizes take the same prompts, photos and settings here. 4B is the cheapest way to explore. BFL positions 9B as its higher-quality small model, and its edit price rises with each photo you attach. FLUX.2 Pro is the full-size step up. Credits are per image, and the generator shows the exact quote before you run.
| Compare | FLUX.2 Klein 4B | FLUX.2 Klein 9B | FLUX.2 Pro |
|---|---|---|---|
| BFL positioning | Smallest, fastest Klein | Flagship small model | Full-size FLUX.2 model |
| Pick it for | Cheap batches of drafts | Edits that hold detail | Photoreal finals |
| Text to image, 1K / 2K | 1 / 4 credits | 2 / 5 credits | 6 / 9 credits |
| Edit with 1 photo, 1K / 2K | 2 / 8 credits | 6 / 12 credits | 9 / 12 credits |
| Edit with 4 photos, 1K / 2K | 2 / 8 credits | 12 / 18 credits | 18 / 21 credits |
| Reference photos | Up to 4 | Up to 4 | Up to 8 |
| Images per run | 1 to 4 | 1 to 4 | 1 |
| Open weights | Apache 2.0 | FLUX Non-Commercial | None, API only |
| Use it | Switch to 4B above | Selected above | FLUX.2 Pro page |
The words you type are everything Klein works from, so a short prompt gets a generic image. These habits matter most.
The subject first
Words near the start carry the most weight. Open with the main subject and action, then add the setting, light and style.
Which photo does what
Call your uploads image 1 to image 4 and give each a single role, such as the person from image 1 in the jacket from image 2.
What must stay
For edits, list what to keep (face, pose, framing, colors) as well as what to change.
Negative phrasing
FLUX.2 has no negative prompt, and writing "no people" can still bring in a crowd. Describe what you want instead, such as an empty street.
Editing your last edit
Every pass redraws the whole image, so color and detail drift with each round. Go back to the original and put all the changes in one prompt.
Complex poses in one try
Hands and tangled poses are a known weak spot. Simplify the pose, attach a photo that shows it, or generate 4 and keep the best.
Example edit prompt: The woman from image 1 wears the red quilted jacket from image 2 and sits on a washing machine in the laundromat from image 3. Keep her face, hair and earrings unchanged. Bright afternoon window light.
Each model has its own page and runs in the same workbench on one credit balance.
New
Black Forest LabsBFL's newest model, with up to 4K output and 10 reference images, for finished pieces.
View model
Black Forest LabsPhotoreal products, materials, camera and lighting
View model
New
GoogleConsistent characters, product placement, exact short text
View model
New
IdeogramRefine image details with source-guided edits and readable text.
View modelRemove the background, upscale, vectorize or denoise any result, all on the same credit balance.
Transparent PNG or WebP with clean hair and product edges, ready for a listing or a new composition.
2× / 4× / 8× with recovered texture instead of smeared pixels, so a 1K draft can become a print.
Logos, icons and flat illustrations become editable SVG that scales to any size.
Remove grain and compression artifacts from low-light photos and scans while keeping real detail.
FLUX.2 Klein, written FLUX.2 [klein] by Black Forest Labs, is the small, fast branch of the FLUX.2 family, released on January 15, 2026. It comes in 4B and 9B sizes, and one model handles text-to-image, photo edits and composing from up to 4 reference images. Oeneo AI runs both distilled sizes online.
Yes. Choose Klein 4B or 9B in the generator on this page and it runs in the cloud, with nothing to download or install. Running the weights yourself takes a capable graphics card; BFL's own table lists about 20 GB of VRAM for the 9B.
You can try it free: new accounts get 10 credits, enough for 10 Klein 4B images at 1K. After that every image costs credits, listed per size in the table above. Images made on the free plan are public, and paid plans can keep them private.
The two sizes ship under different licenses: the 4B weights are Apache 2.0, while the 9B weights use the FLUX Non-Commercial License. Those licenses govern downloading and running the weights yourself. On Oeneo AI you use Klein through a hosted service, and what you may do with your images is set by our Terms of Service and your plan.
No. FLUX.2 [dev] is a larger open-weights model released under a non-commercial license, and Oeneo AI does not offer it. Klein is BFL's smaller, faster line. For a full-size FLUX.2 model on this site, use FLUX.2 Pro.
No. Oeneo AI runs the distilled 4B and 9B with BFL's default settings, so there are no LoRAs, Base models, step or seed controls, or mask painting. To change one area, name it in the prompt and list what should stay.
An edit redraws the whole image, so clothing, colors or even faces can shift. Name what must stay, edit from the original rather than a previous result, and compare the two side by side. In community comparisons, 9B usually holds detail better than 4B.
Complex poses and hands are a known weak spot, and people running Klein locally report extra fingers and twisted limbs in prompt-only images. Attaching a photo that shows the pose works better than describing it. For a finished portrait, FLUX 3 Image may suit better.
Short words can work, but BFL's model card warns that text may come out inaccurate or distorted, so check every letter. For posters, menus or anything text-heavy, use a model built for typography.
Choose Klein when you need to edit or combine photos or want 2K output. On Oeneo AI, Z-Image Turbo makes 1K images from a prompt only. For quick text-only drafts both are low-cost, so try the same prompt on each and compare the look.
The generator above opens on Klein 9B. Attach photos to edit them, or switch to 4B when you want many low-cost drafts. The credit quote updates before every run.
