In short
- AI staging models compose each photograph independently; they do not build a 3D model of the room and re-photograph it.
- Holding the room type, style and seed constant across a set of angles makes them share a palette and a furniture language.
- A shared seed biases the result but does not determine it: in one test of three identical requests, two came back closely matched and one did not.
- Individual pieces of furniture will not line up exactly between viewpoints.
- When one frame strays, re-running it on the same seed usually brings it back toward the rest of the set.
Photograph a living room from the doorway and again from the window, stage both, and the two results will usually disagree. The sofa is a different shape; the rug has moved; one has a floor lamp and the other does not. Agents notice, and so do buyers flicking between photos.
This is not a defect in any one product. It follows from how generative staging works, and it is worth understanding before promising anyone a perfectly consistent set.
Each photo is its own composition
A staging model receives one photograph and an instruction — a room type, a style — and composes furniture into that image. It infers depth and perspective from the photograph in front of it. It does not know that another photograph of the same room exists, and it does not reconstruct the room in three dimensions and then place a camera at each of your positions.
So two photographs of one room are, to the model, two separate rooms that happen to look alike. There is nothing tying the sofa in the first to the sofa in the second.
What a seed is
Generative models start from randomness and refine it into an image. The seed fixes where that randomness starts. Give a model the same photograph, the same instruction and the same seed and it will tend to make the same choices; change the seed and it explores a different result.
Across different photographs, a shared seed cannot make the model draw the same objects — the inputs differ, so the outputs must. What it does is nudge the choices in the same direction: the same kind of sofa, the same palette, the same density of furniture. A set staged on one seed reads as one scheme rather than as several unrelated rooms.
What we measured
We sent three identical requests — same photograph, same instruction, same seed — to the staging model at once. Two came back closely matched. The third had a different sofa, rug and layout. We reproduced it through Vesta and directly against the model's own interface, so it is how the model behaves rather than anything in our pipeline.
That is the honest shape of it: a shared seed moves a set from matching by chance to matching most of the time. It does not move it to always.
How Vesta handles a set
- Multi-angle staging renders every frame of one area in a single pass, with the room type, style and seed held identical across all of them.
- The seed is shown on screen, so a set can be re-run on the same draw rather than starting over.
- You choose the scope: re-render only the frames still queued, which never touches work you have accepted, or re-render the whole area so every frame shares the seed. Both counts and both prices are shown before you commit.
When one frame strays
Re-run that frame on the same seed. Because the seed biases rather than determines, a second pass often lands closer to the rest of the set. If a set keeps disagreeing, try a new seed for the whole area; some draws simply suit a room better than others.