What is a product reference sheet for AI video production and why do you need one?

A product reference sheet is a structured visual + written document you feed an AI video agent before generation: every angle of the product (front, side, back, close-up), a hand-holding-product shot for true scale, and notes on materials, finishes, and on-product text. It exists to prevent product drift across shots — so the same earring, sachet, or garment renders identically in every clip.

Build the sheet with these inputs, every time:

• Multi-angle product shots — front, side, back, and a tight close-up. For garments, add the fabric worn on a person. These give the agent enough visual reference to hold shape and detail across framing changes.

• A scale reference — at least one image of a human hand holding the product (and its packaging layers if relevant). Without this, the agent invents scale, and packaging or jewelry renders at the wrong size shot to shot.

• Material and finish notes — what the surface actually is (brushed metal, matte plastic, linen weave, pavé diamonds) in plain words. This is what stops a model from defaulting to a generic "AI product" look.

• On-product text and logos — exact spellings, fonts where you can describe them, and color hex codes. Text on packaging is where AI drifts fastest.

• Restricted elements — what the product should NEVER show (no alternate colorways, no competitor logos, no plastic-looking finish on a real fabric). Explicit don'ts catch the most common failure modes.

• A locked hero reference — one approved image of the product as it must appear. Every subsequent generation gets compared back to this.

Why you need it — the cost case. AI video iteration is where credits go: across documented productions, clip rejection sits around 85%, and a typical run generates 100+ images and clips to land 10 usable ones. The product sheet is what compresses that ratio. In one jewelry campaign, the same sheet held intricate product consistency across three ads (~$2,400 total for ~5,100 + 2,500 + 1,700 credits); a gummy supplement production used a hand-holding-sachet reference and reported it "solved product consistency at the source." The principle is simple: detail in the sheet upfront prevents expensive re-runs downstream.

How to use it inside the invideo agent. invideo is an agentic video creation platform where you load the sheet once into the project context and every sub-agent — storyboard, image, video — reads from it automatically. You don't re-attach the product on every prompt. The agent routes generations to the right model (GPT-Image-2 for on-pack text and design, Nano Banana for locking exact product detail, Seedance 2.0 reference-to-video for carrying product context into motion) while the sheet stays the single source of truth. Hridaye, invideo's creative director, frames the discipline behind it: "This is how you solve product consistency at the source."

Validate the sheet before you commit a full run: generate the same product at three focal distances — close-up, mid, wide — and confirm it holds. If it does, batch the rest of the shot list. If it drifts at one distance, fix the sheet (usually a missing angle or a missing scale reference) before spending more credits.

A product reference sheet, in short, is the document that turns AI video from "plausible-but-wrong" guesses into a repeatable, on-brand production.

Watch some of these to see what works for you:

This is how you solve product consistency at the source.

— Hridaye, invideo's creative director