AI product photo with a hand holding it: packshot plus pose

To make an AI product photo with a hand holding the item, send the real packshot as the first reference, a photo of the grip or pose as the second, and tell the model to keep the product identical while it draws the hand. On Sume, Nano Banana Pro takes up to 10 references and GPT Image 2.5 up to 16.

Choose the references

The packshot is the truth about the product: its shape, color, label and finish. The pose photo only supplies the grip. Pick a pose with the same size of object so the fingers wrap at the right scale. Remove branding from the pose photo so the model does not copy it.

Write the prompt as a job order

Name each image and what it is for. Say what must not change: the product shape, label text and color. Then say what to add: a natural hand, the skin tone, the cuff, the light direction and the background.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"google/nano-banana-pro","prompt":"Image 1 is the product: keep its shape, color, label and finish identical. Image 2 shows the grip only. Photograph a right hand holding the product from image 1 using the grip from image 2, soft window light from the left, plain warm gray background.","aspect_ratio":"4:5","input_references":[{"type":"image_url","image_url":{"url":"https://example.com/packshot.png"}},{"type":"image_url","image_url":{"url":"https://example.com/grip.png"}}]}'

Check what hands break

Count fingers, check the knuckle direction and look at where the fingers meet the product edge, because contact points are where errors cluster. Then check the product against the packshot: label text, logo, seams and cap shape. If the hand is right but the product drifted, re-run from the original packshot with a stricter preserve list.

Fix with a mask when only one area is wrong

If only the thumb is wrong, switch to GPT Image 2.5 and use mask_url over the hand region, with the instruction to keep everything outside the mask identical. See mask_url edits. Nano Banana models do not take mask_url and return 400, so pick the model by the fix you need.

Use it responsibly

A hand on a product is a staged scene, not a record of a real person using it. If the listing platform asks for disclosure of AI-generated imagery, label it. Keep the original packshot as the main image where a platform asks for a plain product view.