How do you use your existing UGC ads to train an AI agent's creative style?

Upload your existing UGC ads to the invideo agent as reference videos and tell it to study them before writing anything. The invideo agent transcribes each ad, detects every cut, extracts camera language, pacing, hook structure, and on-screen text conventions, and saves those style rules in the project's context tab — so every new generation follows your proven creative DNA.

Start by uploading your best-performing UGC ads directly into the project and instructing the invideo agent to learn their style before any scripting begins — in one documented production, the brief was literally to study how the brand's existing creator content is structured before writing a single word. invideo is an agentic video creation tool, so the reference analysis, style extraction, and generation all happen inside one project with shared memory.

1. Upload the ads and ask for a structural deconstruction. Have the invideo agent transcribe the script, break the ad into named beats with timestamps and emotional logic, and detect every cut — in one run it identified 9 cuts in a reference ad and extracted one representative frame per scene to guide recreation. Ask it to keep what worked structurally and adapt it to the new product.

2. Review the extracted style rules. The invideo agent organizes what it learned under structured production headings — Visual Standard, Pacing & Camera, cut rate, camera style, on-screen text, energy, and ending pattern — and stores all of it in the context tab, the persistent brain of the project. Once saved there, you never re-prompt the style: it applies automatically to every image and video generated in that project. If you also have formal brand guidelines, upload those PDFs alongside the ads rather than making the invideo agent infer everything from footage.

3. Lock the derived shot breakdown before generating. Ask the invideo agent to produce a new shot breakdown — cast, locations, shot sequence, durations, actions, on-screen text — derived from the reference ad's style, then review and explicitly approve it before any image or video credits are spent. Locking is what prevents the trained style from drifting once production starts.

4. Test the learned style with script variations. Brief the invideo agent to write three distinct script directions in different tones against the extracted style rules, pick one, then iterate conversationally. The training compounds within the session: in one production, by shot four the invideo agent had built enough taste context that a short note like "more cinematic, low angle" landed in the first or second try.

5. Watch for caption contamination. If your reference ads carry burned-in captions, exclude the reference video itself from generation prompts — otherwise the model reproduces the old captions in new outputs. Direct generations using the extracted shot breakdown and reference frames instead.

The payoff is cumulative: because the style lives in the context tab, the second ad in a documented session took 2 hours versus 3 for the first — 33% faster with zero setup — and teams running this way produce 4–5 UGC ads per 8-hour day at roughly $125 per ad. If a hook in your reference ads relies on precise timing (a match cut, a snap transition), verify the invideo agent can visualize it from text; if not, point it back to the exact reference clip before locking.

Watch some of these to see what works for you:

Now you could also upload your existing UGC ads here so that the agent can learn its style.

— invideo's creative team