Are AI images good enough for a consumer electronics listing? Partly

Electronics is the category where I have had the most mixed results. Some shots come out fine on the first pass. Others take eight or nine attempts and still do not ship.

The dividing line turned out to be predictable once I stopped treating the category as one thing.

Sort by surface, not by product

What matters is not whether it is a phone or a speaker. It is what the surface does to light.

Surface	Attempts to a usable image	Why
Matte plastic, fabric mesh, soft-touch coating	Low	Diffuse reflection, few constraints to satisfy
Brushed or anodised metal	Medium	Directional highlights have to stay consistent
Polished metal, chrome	High	Reflections should show the environment, and the model invents one
Glass, clear acrylic, screens	High	Transparency plus reflection plus what shows through

For the bottom two rows I now shoot rather than generate. The arithmetic is simple. If attempts multiplied by the time each one takes to review exceeds the cost of a shoot, generating is the more expensive option.

Screens are their own problem

A powered-on screen is the single most common defect I see. The model produces something screen-shaped with plausible-looking content, and that content is invented.

Two workable approaches.

Composite the real interface. Generate or shoot the device with the screen off, then place an actual screenshot into the display area. The screen content is then real by construction.

Or leave the screen off. A dark screen is honest and avoids the whole problem. Many category leaders do exactly this for the primary listing image.

Ports, buttons and text stay real

Anything a buyer will count or read has to come from the actual product.

Port count and type. Buyers check these against their own cables.

Button placement. Getting it wrong reads as a different model number.

Printed markings, model numbers, certification marks. These are frequently used to verify authenticity, and generated versions come out subtly wrong.

My rule is that AI handles background, lighting and scene. It does not handle anything a buyer would use to identify or verify the item.

What works well

Background replacement on an existing shot, which is most of the value in practice.

Consistent lighting across a set, so a category page looks coherent.

Scale references, meaning putting the device next to something of known size.

Colourway variants, but only when the physical product genuinely comes in that colour.

Frame sizes

The product image tool supports 1:1, 4:5, 3:4, 16:9 and 9:16 at 1K, 2K and 4K (product image page, checked 6 August 2026). For electronics I use 1:1 for the primary image and 3:4 where the listing allows a taller crop, because the extra height fits a scale reference without crowding the device.

Checklist

• Sort by surface behaviour before deciding to generate or shoot
• Composite real screenshots rather than generating screen content
• Never generate ports, buttons, markings or model numbers
• Keep colourways limited to what physically exists
• Compare attempt count against shoot cost before committing

Sources

Frame sizes and capabilities come from the TangyuanAI product image page and product overview, https://tangyuanai.vip/en/main-image and https://tangyuanai.vip/en/about , checked 6 August 2026. Surface categories and attempt counts come from my own production notes. Output depends on the uploaded material and the brief.