Image: Black Forest Labs
FLUX 3 Image is the Black Forest Labs model that lets you tell an image generator where things go. Instead of hoping “headline top left, face on the right” lands somewhere sensible, you draw boxes. For anyone who builds thumbnails, title cards, storyboard frames or key art, that is the difference between a lucky render and a layout you can plan. Here is a workflow you can follow today, based on the official pages and one detailed creator guide.
What FLUX 3 Image Is
According to the Black Forest Labs release notes, FLUX 3 Image uses one endpoint for generating and editing, so “the prompt decides whether to generate or edit.” It supports bounding boxes in the prompt, local edits that leave untouched areas alone, up to ten reference images and output up to 4K across 15 aspect ratios. API pricing per image rises with resolution: $0.041 at 768 square, $0.048 at 1K, $0.100 at 2K and $0.607 at 4K.
Black Forest Labs also says on its model page that it works best on compositions with “many parts in strict relationships”: typeset layouts, collages, panel grids, editorial spreads and crowded scenes. That is a fair description of most video-adjacent design work.
Step 1: Pick an Aspect Ratio First
Choose the ratio before you write anything. For a YouTube thumbnail or a 16:9 title card, pick 16:9. For a vertical social cover, pick 9:16. The official page describes a three-step flow: select the aspect ratio, define the elements in boxes, then write a scene description that ties them together. The canvas is a 0 to 1000 grid on both axes whatever the ratio, so the numbers mean the same thing on a wide frame and a tall one.
Step 2: Draw the Boxes
Boxes are written as four numbers, in the order top, left, bottom, right. The Morphic guide shows the shape of a row: an id, a bbox and a description, for example "id": "swimmer_1", "bbox": [560, 380, 820, 620], "desc": "A swimmer in white cap mid-stroke". The scene caption then refers to the element by its id. Black Forest Labs says the model receives your boxes verbatim, with identical ids and coordinates.
You do not have to type coordinates by hand. The model page describes an agent-assisted mode where a language model turns a one-line prompt into a layout caption and an element table, and the boxes stay editable before you generate. Use that to get a rough layout, then nudge the boxes yourself. The browser playground is at flux-tools.bfl.ai/precise-editing, per the official page.
Step 3: Write the Prompt in Plain Order
The Morphic guide to FLUX 3 Image recommends describing medium, subject, setting, light and framing in that order, and using positive wording: “empty street” rather than “no cars.” Its sample prompt is a good template: “An editorial photograph of an empty public reading room: tall oak shelves along both walls, long tables with green-shaded lamps, low winter sun falling through high arched windows in long beams. Wide shot at eye level, 35mm, quiet and warm.”
Note how little of that is style jargon. It names a place, a light source and a lens. That also happens to be how a cinematographer briefs a shot.
Step 4: Use References for Continuity
Up to ten reference images can go in. Per Morphic, you refer to them by position (image 1, image 2) and give each a role: the setting, a product whose shape should stay, a material, or a subject whose likeness should hold. Black Forest Labs lists the tokens as ref_image_0 through ref_image_9. For a storyboard, put your hero frame in as the setting and your character as the subject, and the next frame inherits both.
Step 5: Edit Instead of Re-Rolling
This is where a title card workflow saves time. Morphic’s editing recipe is to send the image with a written instruction, name the target element, specify the change (a hex code for a color, quoted text for words) and list what stays the same. The official page says you can change multiple elements in one generation round without disturbing the rest. So fix the typo in the headline and recolor the lamp in a single pass, rather than regenerating the whole frame.
Step 6: Draft Small, Then Render Final
Morphic recommends drafting at 768 pixels square while you develop the prompt, then rendering finals at 2K or 4K. One warning from that guide is worth taping to your monitor: “A new resolution is a new generation, so the composition can shift between a draft and the final.” Lock your layout with boxes, and check the final before you build on it.
Where It Falls Short
Two honest limits. First, a 4K render costs several times more than a 2K one at the listed rates, so 2K is the sensible default for thumbnails. Second, I did not find an independent test of how closely it follows tightly packed boxes on very dense layouts. Black Forest Labs says the upsampler may suggest additions that only persist if they are captioned, so read your caption carefully.
Bringing It Into Your Edit
Export the final, drop it into Premiere Pro, Resolve or Final Cut Pro as a still, and treat it like any graphic. A generated still is perfectly clean, which can look pasted onto real footage. Add a slow push or pan, then a light layer of film grain across both the still and the footage so they share one texture. A real 35mm scan works better than a digital noise filter for this. Grainzilla’s free grain downloads are ProRes 422, not compressed H.264, so the fine detail survives your export.
Boxes make the layout yours. The grain and the grade are what make it look like it belongs in the film.
Sources: Black Forest Labs, API Release Notes: FLUX.3 Image (Oct. 1, 2026); Black Forest Labs, FLUX 3 Image model page; Morphic, “Flux 3 Image: the complete guide”.
