Pollo AI and Seedance 2.0: build a reviewable ad-testing workflow

Prepare references and the fictional product

The 18 May 2026 recording tests Pollo AI and Seedance 2.0 with Solara SPF50, a fictional sunscreen. Luke starts with trending Reels ideas and creates a product image before generating ads. Treat the source’s quality judgement as an account of these examples, not evidence that synthetic product claims are true.

1. Use Viral Reels Scout for the demonstrated inspiration route: https://github.com/lukejbyrne/viral-reels-scout.
2. Retain the original prompt document: https://docs.google.com/document/d/1D56ZqTPAc3Ii2dLSxs7ZT4LEjtHcBNv83KNTTjV-cog/edit?usp=sharing.
3. Follow product-image creation at 3:21, checking the packaging and wording before using that image as video input.

Compare the seven ad styles manually

The styles are UGC, product promo, review, unboxing, explainer, native product ad and direct response. The original Pollo route is https://bit.ly/3QX1G0z. The description promotes an unlimited offer there; check current terms rather than treating that phrase as an enduring entitlement.

1. Follow image-to-video setup at 4:22 and the seven-style comparison at 5:17.
2. Inspect each result for product consistency, spoken claims, skin details, sound and understandable movement.
3. Keep generated reviews and presenters clearly distinct from real customer testimony.
4. Save the prompt and selected output together so a revision can target a specific weakness rather than restart the entire comparison.

Scale a checked creative, not an unchecked prompt

At 9:20 the sequence moves to the Pollo API with Codex; Meta Ads MCP appears at 17:16. The description explicitly favours manually testing prompts before scaling. Good-looking imagery, music and AI voices can support a production workflow without proving that an advertisement converts.

1. Choose a reviewed format and define which inputs should vary in the next batch.
2. Compare API outputs with the approved manual example and inspect any changed product details.
3. Keep campaign setup, publishing permissions and any later performance measurement separate from generating the files.

The takeaway

The reproducible asset is a checked prompt-and-output pair; automation should preserve its useful constraints rather than multiply mistakes.