How We Keep a Product Consistent Across 100 AI-Generated Shots

A single beautiful frame proves almost nothing. Any model in 2026 will hand you one. The commercially interesting question is what happens at shot forty, when the client has approved the look and the deadline has stopped being theoretical, and the product has quietly grown a millimetre of shoulder.

Here is how we run a hundred-shot set, in the order it happens.

1. Lock the product before the set exists

The first artefact is a master plate: one clean, flatly lit frame of the product, verified against the packaging artwork at full resolution and then frozen. Nothing gets generated until that file has a version tag on it. Every campaign frame afterwards is generated from the plate rather than from a prompt describing the product.

The reason this comes first is arithmetic. A drift discovered at shot four costs one regeneration. The same drift discovered at shot ninety costs eighty-six.

2. Write the set once, in words

Before any frame, the set becomes a document. It is short and it is boring and it is the reason the hundredth shot belongs to the same world as the first.

• —One key light direction, stated as a clock position. Not "soft natural light".
• —One colour temperature, stated in kelvin.
• —One surface and one background material, named.
• —A lens language: focal length range, and whether depth of field is shallow or deep. Pick one and hold it.
• —A list of what is never in frame. Hands, competitor products, seasonal props, anything with text on it that is not the product.

3. Batch by difficulty, not by deliverable

The instinct is to render in the order the shot list is written. Do the opposite. Run the hardest shots first: the ones with legible printed type, the ones where the product is held, the ones at an angle the plate does not cover well.

Hard shots tell you your real acceptance rate within the first hour, while there is still time to change the approach. Easy shots tell you nothing and make you optimistic, which is worse than useless on a fixed budget.

4. Gate every frame, mechanically

Four tests, applied to every frame, with a pass and a fail rather than a discussion.

Those numbers are the reason a hundred-shot set is a three-hundred-render job. They are also why we agree a credit ceiling per asset before a batch starts: when a shot has consumed its ceiling, the run halts and asks for a decision rather than quietly spending its way to a number nobody approved.

5. Never patch. Return to the plate.

A failed frame goes back to its source file and is regenerated with a tighter constraint. It does not go to a retoucher. This is the rule people break under deadline pressure and it is the rule that decides whether the set holds.

A patched frame passes review in isolation and still does not sit correctly beside its neighbours, because the drift that produced it is still in the pipeline and will produce the next twelve frames the same way. The retouching bill then grows silently until it exceeds the generation budget, which is how a cheap production becomes an expensive one without anyone deciding to make it so.

6. Review the set, not the shot

The final gate is a contact sheet. All hundred frames, small, on one screen. Drift that is invisible at full size is obvious at thumbnail scale, because the eye stops reading each image and starts comparing them.

Anything that jumps out at thumbnail scale gets regenerated even if it passed every individual gate. This catches the class of failure no per-frame test can: the shot that is correct and still wrong.

What this costs, honestly

Three hundred renders to ship a hundred assets means your true cost per accepted asset is three times the per-render price, before any human time. Add the operator hours, the plate build and the gating, and a set like this lands where it lands. The number is not the point. The point is that it is knowable in advance, and a studio that has run this method before can tell you the ratio at quote time rather than discovering it at invoice time.