The person is supposed to stand on the left; the headline needs space on the right. A long prompt full of style terms can still miss this simple requirement. The image looks appealing, but the workbench sits right behind the text. For a website, the problem is then not the beauty of the image but the wrong distribution of roles across the surface.
On 15 October 2026, Black Forest Labs introduced FLUX 3 Image Klein. Among other things, the announcement describes control via bounding boxes, that is, rectangles that mark out an area. They provide a concrete reason to formulate image briefs with more spatial precision. A box, however, does not yet say which action makes sense within its boundaries.
Three zones make the brief understandable
Our own example is a workshop announcement. The figure is sorting materials at a workbench. The illustration is meant to explain this action; the headline remains real page text. To do this, divide the imagined image area into a figure zone, an action zone and a calm transition area.
The figure must not become so small that the action is lost. The workbench needs enough space to remain recognisable as a place of work. The text zone should not also contain the most detailed object. These relationships matter more than an isolated demand for “more white space”.
Your own briefing note might read: figure on the left, looking towards the action in the middle, calm area on the right without additional objects. This is a spatial intention, not valid parameter syntax for any specific model. You will need to take the supported box format from the actual tool.
A bounding box does not replace meaning
A correctly placed person can still be performing the wrong action. Conversely, the action can be right while the proportions separate the figure from the object. A usable brief therefore names position and relationship together: who is doing what, how does it become visible, and which area has to stay calm?
For checking, the intention can be translated into three questions. Is the figure in the intended zone? Can the activity be recognised without additional explanation? Is there enough space left on the eventual page for easily readable text? An answer to the first question does not answer the other two.
A visual check like this is not a quality benchmark. It assesses a specific image for a specific use. A general model comparison would have to document considerably more tasks, conditions and actually generated results.
Why a crop is not a new composition
Reduce the imagined wide area to portrait format. The figure may still be visible while the workbench disappears. Or the object remains, but the line of sight leads out of the picture. Technically the crop fits; in terms of content, it loses the explanation.
So design the mobile version as a second arrangement of the same message. The action can move further up, the figure can stand smaller beside it and the real text can follow below. The aim is not to keep every element in the same relative position. It is to preserve the meaning under different space constraints.
For storing finished versions, KENJTEC MediaStudio ai offers a shared library and additional fixed crops, including 16:9 and 4:5. These create separate files under the original. They do not recompose the action. You create the deliberately designed second version in the image tool you are using and add it yourself. Images provided in MediaStudio can be accessed via their public file address. Do not upload confidential documents there; sensitive internal files stay in the BusinessCloud, which has separate permissions.
Open weights and good design are separate questions
The release describes an open-weights offering with licence terms. This does not imply blanket, unrestricted commercial use. Before actual use, the applicable terms are as much a part of choosing a tool as the supported control options.
Regardless of that, you can begin your next image brief with a sketch: three zones, one concrete action and a second mobile arrangement. This turns spatial control into an understandable design brief. Whether the model used fulfils it is then shown by the image that actually exists, not by the precision of the prompt alone.
Sources and status
Research and documentation as of 25 October 2026. FLUX 3 Image Klein.
