Product ad shot design

Inputs to prepare

• —A product photo or product reference. Several cases add a dedicated product lock so shape, colors, label, and logo stay fixed.
• —A model reference image if you want a consistent person, plus a wardrobe reference when clothing matters.
• —One benefit, one closing tagline, and the brand name.
• —Duration and aspect ratio, for example 15 seconds at 9:16 or 16:9.
• —If someone speaks, the exact lines, ideally one or two.

Method steps

1. 01Pick one benefit. The shampoo case sells only smooth, glossy hair; the soda case sells only post-workout relief.
2. 02Split the runtime by the second. For 15 seconds that is usually 4 or 5 beats: a 2 to 3 second hook on a product macro or a face, a middle that shows use (dispense, apply, crack open, put on), then the change or the feeling, and the last 2 to 3 seconds for the product hero shot, logo, and tagline.
3. 03Give each beat its camera language and light: macro, push-in, orbit, slow motion, match cut, studio or daylight.
4. 04Write consistency locks. Name the product shape and label text, the model's face and outfit, and say they must not change. With reference images, list each item.
5. 05Write sound and text on their own: exact dialogue, effects like the can crack and fizz, music tempo, and the second at which on-screen copy appears.
6. 06End with an avoid list. Common entries are plastic skin, distorted hands, extra text, cheap CGI, and a redesigned product.

Variables to swap

• —The product and its category
• —How the benefit is proven: before and after, texture macro, lifestyle montage, or the product transforming the scene
• —Production style: high-end studio commercial or handheld phone UGC with a talking head
• —Duration and aspect: 10 to 30 seconds, 9:16 or 16:9
• —Model, wardrobe, and setting
• —Closing copy and call to action

Common failures and limits

• —On-screen text and logos are unreliable. Many prompts ask for brand copy and, in the same breath, list extra text, fake logos, and logo distortion as things to avoid. Our retest data cannot show whether the rendered text was correct. (Source: prompt avoid lists; not verifiable in retests)
• —Replace placeholders yourself. The shampoo prompt says [BRAND NAME], and our retest submitted it unchanged, so fill in a real brand before you reuse it. (Source: prompt hash matches the retest)
• —Hands, faces, and products warp. Avoid lists keep naming distorted fingers, warped features, and redesigned products. Lock each item when you have references. (Source: prompt avoid lists)
• —Write the dialogue word for word. In the earbuds UGC retest the transcript matched the scripted lines exactly. In a few cases with no dialogue, the transcript still picked up unrelated words, and the data cannot tell whether that was generated noise or a transcription error. (Source: retest audio transcripts)
• —Length and resolution depend on the model. Prompts often ask for 15 to 30 seconds at 4K or 8K, while our retests ran at 768p and did not record output length. Check how long a single generation can be before you switch platforms. (Source: model limits; duration data unknown)