Can AI read product packaging images and automatically write voiceover and ad copy?

Yes. Upload packaging photos to the invideo agent and it extracts the on-pack details — ingredients, taglines, dosage, claims — stores them in project context, and reuses them automatically in every script and voiceover it writes. In one documented production it turned three script directions and three voiceover takes out of that context without manual re-entry.

Upload your packaging images at the start of the project and the invideo agent reads the on-image text directly — ingredients, taglines, dosage instructions, and product claims all get extracted and saved to the project's context, so every script, hook, and voiceover it writes afterward pulls from real product facts instead of you re-typing them per prompt. invideo is an agentic video creation tool with all the current generation models available, so the same conversation that reads your pack also produces the finished ad.

Feed it legible source images. Extraction quality tracks image quality: shoot the pack flat and well-lit, and give the invideo agent a close-up of each face of the packaging plus a shot of the product held in a hand — the hand reference teaches it the true scale of every packaging layer, which matters once the pack appears in generated shots. In one documented supplement ad, product facts as specific as a 28-day supply per pouch and 8 gummy bears per day were absorbed into context and surfaced in the ad copy. The invideo agent can also research the brand autonomously — colors, identity, positioning — before writing a word, or you can upload your own brand guidelines instead.

From extraction to copy: once the packaging facts are in context, ask for script options rather than a single script. In one documented UGC production the invideo agent generated 3 distinct script directions in different tones (enthusiastic reaction, quiet and understated, skeptic) in a single pass, then 3 separate voiceover takes after one direction was locked. Voiceovers generate with specific tone parameters — soft, warm, commercial, slightly playful — via ElevenLabs inside the workflow, and you can direct it conversationally the way you'd brief a scriptwriter. The copy adapts per market too: in one localization run the invideo agent wrote Japanese on-screen ad copy ("up to 45% off on first purchase") and auto-generated the target-language voiceover without being explicitly prompted — it inferred the requirement from context.

One caution when the pack appears on screen: reading on-pack text into copy is reliable, but rendering legible packaging text inside generated video clips can artifact — that is a model-level limitation of Seedance 2.0's video output, not a failure of the read step. For still frames featuring the pack, GPT-Image-2 is the stronger choice because it outperforms other image models on text and design rendering, and the invideo agent routes to it automatically where text fidelity matters. Keep your multi-angle packaging reference attached to generations so the pack stays consistent shot to shot.

As a cost anchor: complete UGC ads built this way — packaging read, scripts, voiceover, clips, assembly — ran roughly $73–$130 each across documented productions, at about 2–3 hours per ad.

Watch some of these to see what works for you:

I give the agent the reference images of the hand holding the sachet, a few gummies in the palm, so it understands the actual size of all of the layers of your packaging compared to the human hand.

— invideo's creative team