Product Reference Preservation for AI Commerce Scenes

This product reference preservation for AI commerce example uses a host scene plus one isolated product photograph as its production input. Its target was a product-led frame that keeps both the host identity and object structure readable, and the saved canvas shows which parts of that target held up in the rendered frame or clip.

Canvas image: a product-led frame that keeps both the host identity and object structure readable.

A useful way to read this test

The source request focused on a host scene plus one isolated product photograph. The agent converted it into one visible goal: Assign the host image to scene continuity and the product image to shape, material, hardware, and color.

What changed on screen

The handbag version shows why a product needs its own detail list: handles, flap, lock, metal bars, stitching, and body proportions all contribute to recognition.

The visible target was a product-led frame that keeps both the host identity and object structure readable. For a related run, keep this production decision: Split reference authority by subject. A single image should not be asked to own both the person and an unrelated product.

Try the same decision path

1. Start with a host scene plus one isolated product photograph.
2. Write the visible constraint that matters most: Split reference authority by subject. A single image should not be asked to own both the person and an unrelated product.
3. Generate one reviewable result, inspect still frames as well as playback, and keep the saved canvas media as the production record.

The approach transfers to fashion commerce, beauty packaging, food presentation, or any host-led item comparison.

Open the source ArtArch Studio canvas to inspect the saved creative context.

Frequently asked questions

What should I prepare first?

Prepare a host scene plus one isolated product photograph. Use the clearest original files available and decide which one owns identity, product detail, or endpoint composition.

What should I inspect in the result?

Check the result against this visible target: a product-led frame that keeps both the host identity and object structure readable. Review framing, subject identity, props, text, and the first and last frames when video is involved.

Can I reuse the same setup?

Yes. Keep the canvas media and the decision that controlled it: Split reference authority by subject. A single image should not be asked to own both the person and an unrelated product.

Use the template card on this page to enter the experience page and try this workflow with your own references.