FLUX 3 Klein Text to Image

FLUX 3 Klein Text to Image

FLUX 3 Klein Text to Image is the fast, open-weights member of Black Forest Labs' FLUX 3 Image family, built for high-quality text-to-image generation with precise, prompt-driven layout control.

What is FLUX 3 Klein Text to Image?

FLUX 3 Klein is a 5.6B-parameter, step-distilled text-to-image model from Black Forest Labs. It turns a written prompt into a single image at up to native 4K (about 16 megapixels) across five resolution tiers and 15 aspect ratios. Its signature feature is bounding-box layout control: you tag elements in the prompt and append a JSON list of boxes on a 0-1000 grid, and the model renders each element inside its assigned region. That makes it a strong fit for posters, product grids, and any composition where placement matters, while staying fast and efficient enough for high-volume work.

Key Features

• •Bounding-box layout control — place named elements in exact regions using a JSON box list, each box written as [top, left, bottom, right] on a 0-1000 grid.
• •Native 4K output — generate up to about 16 megapixels across the 768sq, 1k, 1.5k, 2k and 4k tiers.
• •Accurate in-image text — render legible headlines, signage and labels, including non-Latin scripts such as Devanagari.
• •Broad stylistic range — photography, flat illustration, pixel art, stained glass, isometric 3D and more from one model.
• •Fast and efficient — a distilled 16-step model with JPEG or PNG output and an optional fixed seed for reproducibility.

Best Use Cases

FLUX 3 Klein suits design-adjacent work where layout and text matter: mock packaging, labelled diagrams, product flat-lays, posters and signage. Its native 4K output serves print and hero imagery with detail-dense close-ups, and its style range lets a single model cover a brand's photographic and illustrated needs. In testing, bounding-box placement was accurate across a five-element grid, and both Latin and Devanagari text rendered cleanly.

Prompt Tips and Output Quality

Write concrete nouns and one clear action, and name materials, styles and placement. To control layout, tag elements like <logo> and append a JSON box list; quote exact words in single quotes for signage. Use 2k for everyday work and 4k when fine texture must survive. Output stayed clean and artifact-free at both tiers in testing.

How to Use FLUX 3 Klein Text to Image

FLUX 3 Klein turns a prompt into a single image at up to native 4K, with bounding-box control over where each element lands. Write a clear, concrete prompt, choose a resolution tier, and set an aspect ratio — or let auto infer one from your words.

Getting the layout you want

The model's standout control is bounding boxes. Tag elements in the prompt with <id> markers, then append a JSON array of boxes, each written as {"id", "bbox": [top, left, bottom, right], "desc"} on a 0-1000 grid. Every tagged element renders inside its box, so you can pin a headline, place a product, or lay out each cell of a grid.

Recommended Settings

• •Resolution: 2k is the sweet spot (about 4 MP, ~15s per image). Use 4k (about 16 MP) for print or detail-dense close-ups; drop to 768sq or 1k for fast drafts.
• •Aspect ratio: auto reads the ratio from your prompt; set a fixed ratio (1:1, 4:5, 16:9) for a known canvas.
• •Steps: 16 is full quality for the distilled model and costs the same as fewer; lower only to iterate faster.
• •Output format and seed: jpeg for smaller files, png for lossless; fix seed to reproduce a result, leave it empty for variety.

Example prompts to try

Each example below ran end to end at the settings shown, and the layout, text and style carried over exactly as written.

• •Bounding-box layout (default), 2k, 1:1. Prompt: "Top-down flat-lay on a reclaimed-wood workbench, four components arranged in a clean grid with even daylight: a disassembled brass pocket-watch movement <watch>, a tight coil of bright copper wire <wire>, a teal printed circuit board <board>, and a fan of red-handled jeweler's screwdrivers <drivers>, with a small white index card <label> in the centre. [… JSON box list for watch, wire, board, drivers, label …]". All five tagged parts landed in their own quadrants and the centre card read BENCH KIT No. 3 — tag each element and append the box list to pin the whole layout.
• •Native 4K detail, 4k, 1:1. Prompt: "Extreme close-up of a hand-embroidered silk panel, gold metallic thread forming a peacock-feather motif over deep indigo fabric, a steel needle and a loose thread resting on the woven surface, even daylight." Individual thread fibres and the needle resolved on a 4096×4096 canvas — reach for 4k when fine texture must survive, and expect about 80s versus ~15s at 2k.
• •Multilingual in-image text, 2k, 4:5. Prompt: "A weathered hand-painted wooden shop sign above a spice merchant's stall in an old bazaar. The sign reads 'मसाला भंडार' in large ochre Devanagari letters, with 'SPICE HOUSE - EST. 1911' beneath in cream serif capitals. Burlap sacks of turmeric and dried red chillies sit below, even daylight." Both scripts rendered legibly — quote exact words in single quotes and keep each line short.
• •A named art style, 2k, 2:3. Prompt: "Stained-glass window of a flamenco dancer mid-spin, one arm raised and her ruffled skirt flaring into a wide arc, thick black lead lines separating panes of jewel-toned glass in crimson, amber and deep blue, strong backlight pouring through the coloured panes." The stained-glass look read clearly with lead lines and jewel panes — name the medium explicitly to lock the style in.

Further prompts the model is built for: "a vintage travel poster with a bold title", "an isometric 3D diorama of a tiny workshop", and "16-bit pixel art of a rainy night market".

Does it support bounding boxes?

Yes — append a JSON box list to the end of the prompt.

What is the maximum resolution?

Native 4K, about 16 megapixels.

Can it render text in images?

Yes, including non-Latin scripts.

Can it edit existing images?

Not on this text-to-image endpoint; use FLUX 3 Klein Image Edit (flux-3-klein-image-edit) for edits and up to ten references.

Are the weights open?

Yes, released under the FLUX Kommunity License.