Can AI replace every catalogue image?

Not without checking product requirements. Exact variants and materials may need photography or a mixed workflow.

What makes a useful comparison?

Matched products, originals, output format and revision tasks, recording attempts and accepted outputs.

Nano Banana and Midjourney for product imagery (2026)

Choose products that reveal different errors

Select clear packaging text, distinctive material and closely related variants as a practical sample, not a representative benchmark. Use identical originals and target formats.

Start simply, then add the desired setting. Compare colour, proportions, markings and material. Turning a matte surface glossy may make an attractive but inaccurate image.

Test a variant change

Change only the relevant colour or variant and inspect preserved geometry, perspective and background. Confused variants may require stronger naming and verification or original product cut-outs. Atmospheric quality does not establish catalogue reliability.

Record the actual model, version and access. Compare accepted output including correction time, keeping photography as an option.

Sample a catalogue before expanding

Choose challenges deliberately: label spelling, reflective surfaces and similar colour variants. Record why each item is included.

Assess against named approved references rather than memory, including proportions and cropped details.

Revise one item without altering others and include selection, correction and export time.

Use findings to define product-category rules, such as generated settings with original materials where needed. This is a decision format, not a declared Nano Banana or Midjourney result.

Keep product truth ahead of atmosphere. A glossy interpretation of a matte surface can be visually appealing but unsuitable for representing the item a buyer will receive. Compare markings, proportions and texture against approved images at the intended output size.

Start with a straightforward display before introducing elaborate settings. Then test a variant change while preserving the accepted product and background. If colours or packaging are confused, strengthen the reference-to-file mapping or use original product imagery. Record the model, access method and relevant settings for the trial, and judge total effort to accepted output rather than the best image selected from an undisclosed number of attempts.