How do I use my own product catalog images instead of the AI agent's automatic wardrobe selections?

Override the invideo agent's auto-picked wardrobe by pulling the exact items from your product catalog and uploading them straight into the chat as reference images, then explicitly assigning each item to a specific character. Once you like the result, lock that character version by its version number so every later shot reuses those exact garments.

Start by opening the chat with the invideo agent on your project and pulling up its current wardrobe picks. When something doesn't match what you want from the catalog, drop the actual product images into the chat — close-up of the fabric, front, side, back, and a worn-on-person shot for each item. Sparse references produce drift; the agent needs the garment from multiple angles to render it consistently across shots.

In the same message, assign each uploaded item to a specific character by name and add styling notes ("give Tiara the linen mini dress from this reference, worn loose, sleeves rolled"). The agent treats this as an override of its own selection — you're not nudging its taste, you're replacing the pick. As one creative director on a fashion ad put it: "The agent picks some wardrobe options from the product catalog but I don't quite like the ones that it's picked, so I just pick some products from the product catalog myself and I upload them into the chat."

Once the agent regenerates the character with your chosen garment and you like a specific version, lock it by referencing its exact version number ("lock v3 of Tiara with this outfit"). Version-number locking is what stops the agent from drifting back to its earlier picks on the next shot — every subsequent storyboard frame, keyframe and video clip inherits that locked character + garment pair without re-prompting. For multi-character ads, lock each character sequentially, one at a time, before moving on to locations or video generation.

For scale: if you're doing this across a full catalog, upload your lookbook as a PDF into the project's context tab — every catalogue item at multiple angles, plus product-only and on-model shots. Once it's in context, the agent draws from your SKUs by default instead of inventing wardrobe, and you only have to do chat-level overrides on the exceptions. Brands with existing guidelines should upload their own decks rather than asking the agent to research — across documented productions, three context PDFs (visual guidelines, product catalogue, treatment note) were loaded once and reused across every ad in the project, with the second and subsequent ads producing 33% faster because the catalogue was already in memory.

One consistency tip worth doing upfront: name each uploaded reference file with the product ID, colorway and fabric type (e.g. LINEN-MINI-DRESS_off-white_linen.jpg). The agent reads filenames as structured metadata and matches them against brief requirements — the same trick that makes batch market localization work via language-named files works here for SKU routing.

If a shot is already generated and you only want to swap the garment in that frame, regenerate the keyframe image with your reference attached and the new garment assigned before spending video credits — image iteration is cheap, video iteration isn't.

Watch some of these to see what works for you:

The agent picks some wardrobe options from the product catalog but I don't quite like the ones that it's picked, so I just pick some products from the product catalog myself and I upload them into the chat.

— invideo's creative team, on overriding auto-selected wardrobe