An ordered prompt structure for product visuals in which order and priority of information matter more than prompt length or complexity. The slots, in fixed sequence, are:
This part of the source only elaborates the first three slots. The last four are named but not detailed.
The claim is that structure, not length, is the lever. Each layer exists to take away a freedom the model would otherwise use to drift. See prompt-layers-remove-model-freedoms. The source says a clear hierarchy shortens iteration, keeps corrections minimal and makes brand consistency predictable.
A running shoe with technical text on the heel and mid-sole was generated in two still-life environments, sandstone and terracotta. Zoomed in, the text stayed sharp, aligned, readable and faithful to the real product. A shortened prompt that kept the same hierarchy gave consistent results too, which the source takes as proof that order beats length.
Write the prompt in the fixed sequence above. When shortening, cut words but keep the order and the slots.
"The goal is not to create a complicated prompt. The goal is to create a structured prompt."
The same hierarchy is claimed to carry from stills into video: see typography-persistence-in-video-via-active-references.
The formula in order: composition first, exact identity, typography control, true color, lighting harmony, optical clarity, negative constraints.
The tutorial's central claim is that accuracy comes from the order and priority of information, not from the length of the prompt. Order tells the model what to treat as primary. composition-first-anchoring goes first because an unanchored prompt lets the model prioritize the object itself, and the overall aesthetic can become unstable.
The same formula was tested in two very different environments, sandstone and terracotta, with different materials and color balance. The structure held in both, which is offered as evidence that it does not depend on one scene.
The hierarchy applies to stills in Nano Banana Pro. For video in Kling 3.0 Omni, the references carry most of it, and the prompt shrinks to minimal-video-prompt-references-hold-identity.