# Stillpoint

- **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:** [Pronita Ghimire](https://cerebralvalley.ai/u/hacker_pro)
- **GitHub:** https://github.com/prashma03/StillPoint
- **Demo video:** https://youtu.be/lVgmN3hmNZo?is=ZAuOIZyvwUQlpM1v
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4237

Stillpoint: Proof Without Seeing

As AI becomes increasingly capable of manipulating realistic images, an innocent photograph can be transformed into convincing, non-consensual intimate content. Yet investigating that manipulation can force victims to expose the very images that harmed them. Stillpoint addresses this gap through privacy-preserving verification, evidence collection, and proactive protection.

Rather than relying on facial recognition or asking human reviewers to examine sensitive imagery, Stillpoint investigates background geometry, camera perspective, spatial relationships, and visual feature correspondences. Automatic and manual masking help conceal sensitive regions before comparison.

Check lets users upload an original and a suspicious photograph, mask sensitive content, and examine geometric similarities. Automatically identified reference points can be added or removed to refine the comparison.

Proof brings the analysis to life through interactive 3D-inspired visualizations showing how structural features align. Users can also download PDF evidence reports documenting the findings and their limitations, without presenting geometric similarity as definitive proof of manipulation.

Protect introduces an experimental AI Protection Shield that applies image-level modifications to original photographs before sharing. Users can preview and download a shielded version. The goal is to explore defenses against unauthorized AI editing and regeneration, although effectiveness across different models still requires validation.

Discover is designed to help users investigate whether their photographs, including potentially manipulated versions, have appeared elsewhere on the public internet. Trust explains Stillpoint's approach to privacy, consent, transparency, and evidence integrity.

Stillpoint also includes a Meta-powered AI Guide integrated through the Meta Model API. The chatbot explains image verification, masking, geometric evidence, protective measures, and technical limitations in accessible language, so users don't need a computer vision background to understand the process.

Unlike tools that focus only on detecting deepfakes or removing harmful content after circulation, Stillpoint considers the entire experience: discovering suspicious imagery, investigating possible connections to an original photograph, concealing sensitive information, preserving supporting evidence, and exploring prevention before future misuse.

Our goal isn't to label every suspicious image as fake. It's to give people a clearer, more private way to examine the available evidence, recognize uncertainty, and decide what to do next.

Your image. Your consent. Your control.

Stillpoint: Proof Without Seeing.

## More from Meta Global AI Developer Hackathon

- [World Studio](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4234)
- [Recall](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4235)
- [genUIne](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4236)
- [The N Shadow](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4238)
- [Maple](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4239)
- [homebrain](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4240)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4237. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt