A Shoot Has Controls. A Generation Has a Button

How it works

The photograph is scaled to about one and a fifth megapixels and repainted at a moderate denoise. The instruction describes a conventional studio setup in physical terms: one large soft key placed high and slightly to one side, a weaker fill opposite to open the shadows, and a clean gradient behind.

The product is held by name — shape, proportions, color, materials, label, and every marking on it.

The negative prompt covers both halves of the failure: the product changing, and the lighting being wrong in ways that read as amateur, such as harsh flash, multiple conflicting shadows, and blown highlights.

Each run gets a fresh seed, so two runs of the same file will differ.

What a light actually controls

The reason to describe lighting physically rather than as a mood is that each property maps onto something you can see in the result.

Size relative to the subject sets shadow softness. A softbox that is large as seen from the product wraps light around it and produces a soft shadow edge; a small source at a distance produces a hard one. This is why the same box gives different results at two distances.

Height sets where the shadow falls and how much of the top surface is lit. Raise the key and the shadow drops behind and below; lower it and the shadow stretches towards the camera.

The fill ratio sets how much detail survives in the dark side. No fill is dramatic and hides half the product. Too much fill is flat and hides the form.

Background separation is a third light or a gradient, and it exists so the product's dark edge does not disappear into a dark background.

A photographer changes one of these at a time. The value of this page is that it produces a plausible combination of all four; the limit is that it does so as a single package.

Generate and select, not adjust

The workflow this page implies is different from an editing workflow, and mismatched expectations are what make people frustrated with it.

An editing workflow converges. You look at what is wrong, you change the thing that causes it, and the image gets closer. Ten iterations produce one good result.

A generative workflow samples. You run it, you look, and if it is wrong you run it again, hoping for a better draw. Ten runs produce ten independent results, and you pick.

So treat this as a selection problem. Run it several times, keep the ones where the product is intact, then choose on lighting. Trying to nudge a specific run towards a specific improvement is the one thing the tool cannot do.

The ten-minute alternative

It is worth stating plainly, because for a lot of products it is better than anything on this page.

A window gives you a large soft source. A sheet of white card opposite gives you fill, and moving it changes the ratio. A sheet of paper curved up from the table gives you a sweep with a real gradient. A phone on any support gives you a stable camera.

That setup costs nothing, takes ten minutes, and produces a photograph of the actual product with real light on real materials — no invented labels, no drifting proportions, nothing to audit.

Use this page when the product is not available to photograph, when you are producing a variant of an existing shot, or when the shoot has already happened and the lighting is what needs fixing.

Publication gate

This page ships once the workflow has been run against a flat phone photograph of a labeled bottle, a product already lit in a studio, a highly reflective metal object, and a product with small printed type, with the label read character by character against the source in every case.