# TestBench

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Muhamed-Kanapiya Zhaksylyk](https://cerebralvalley.ai/u/qazaqbiz)
- **GitHub:** https://github.com/muhamed-kanapiya/testbench
- **Demo video:** https://youtu.be/cK27J4VeE38
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3652

TestBench is a creative experiment studio for performance marketers, small ecommerce brands and agencies. It turns one product photo into a structured set of ad creatives with an explicit test variable.

Creative tests become difficult to interpret when the background, product layout and message all change together. TestBench lets the user choose what to vary and keeps the other creative layers consistent within each format.

The workflow starts with a product image and a reviewed cutout. Users can compare three backgrounds, three product composition presets, three headlines or three artwork CTA texts. Meta models handle segmentation, background generation, copy alternatives and an editable test hypothesis. A deterministic Canvas renderer combines these outputs with the approved product, brand and offer layers.

The studio includes logo and offer overlays, Instagram Feed, Facebook Feed and Stories simulations, local autosave and Undo. Users export nine PNGs: three variants at 1080 x 1350, 1080 x 1080 and 1080 x 1920. The ZIP also includes the source product layer, optional logo and an experiment manifest recording the variable, variant settings, asset provenance and product SHA-256. Validation blocks duplicate variants and text overflow.

The prototype has been tested with live Meta requests and browser exports. Its purpose is to help small teams prepare controlled creative experiments. It does not launch ads or predict a winner. Campaign testing and outcome measurement happen outside TestBench. Composition tests compare position and scale together, and placement previews are simulations. Morrow is a fictional demo brand.

https://youtu.be/cK27J4VeE38

## More from Meta Global AI Developer Hackathon

- [Thiba Mahlezana](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3649)
- [DreamTeam](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3650)
- [ArborPulse](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3651)
- [IndexReel](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3653)
- [EvalVerse](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3654)
- [FlowSME](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3655)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3652. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt