PARALLAX

Inspiration

Making a useful commercial takes more than generating an attractive clip. Small creative teams also need audience research, a coherent concept, consistent characters, readable packaging and control over production decisions. Parallax brings those steps together around a real Super Leaf laundry-sheet campaign set in a Bangkok condo.

What it does

Parallax presents the campaign journey: a creative brief, saved Parallel research, three creative directions, storyboard reviews, an edit manifest and a finished 17.208-second commercial. Reviewers can inspect source links and historical Google Vertex AI video-generation records alongside the creative output.

The hosted version is a saved campaign demo. It displays completed project work; it does not run new searches or generate media. Its separate Guardian approval simulation illustrates a hypothetical $6.40 approval against a $20 budget, with committed and spent amounts shown separately. The simulation is resettable and spends no money.

How we built it

The original command-line prototype uses Google ADK and Gemini for creative and multimodal work, Parallel for research, and Google Vertex AI with Veo for motion generation. The retained research contains five queries and eight sources. Saved motion reports include model names, operation identifiers, source frames and latency measurements.

The interface uses React 19, TypeScript and Vite. Python/Pillow compositing preserves authoritative product artwork, and FFmpeg assembles the editorial output. The demo frontend edits were authored with Google Gemini CLI. The hosted frontend is static; the historical backend remains a separate prototype.

Challenges we ran into

Generated packaging could distort Thai text, and scenes could drift in face, wardrobe or physical interaction. Parallax addressed packaging fidelity through composition with the real artwork, while recorded visual audits exposed remaining problems. The demo preserves the historical still-audit failures; later motion reports provide a separate record of reviewed iterations.

Another challenge was making the demonstration match the actual project state. Saved campaign evidence, current browser interactions and future live production capabilities needed clear boundaries.

Accomplishments that we're proud of

• A completed 1080p, 24 fps campaign film with a documented edit manifest.
• A workspace connecting research, creative choices, storyboard review and output.
• Inspectable historical Parallel research and Vertex AI generation records.
• Product-label fidelity through authoritative artwork composition.
• A repeatable approval illustration that separates committed budget from actual spending.

What we learned

Creative generation becomes more useful when each decision has context and each output can be reviewed. Exact product artwork benefits from deterministic composition, visual QA should retain failures, and a demo should clearly distinguish saved evidence from live execution.

What's next for PARALLAX

Connect the interface to a reliable production backend, preserve request receipts automatically, and make campaign state consistent across research, generation and editorial stages. Future work also includes authenticated producer approvals, durable job tracking and budget controls around real provider calls. These are development goals, not capabilities of the current hosted demo.

Built With

• gemini
• google-adk
• google-cloud
• parallel
• react
• typescript
• vertex-ai
• vite