Imago Obscura: An Image Privacy AI Co-pilot to Enable Identification and Mitigation of Risks
summary
Imago Obscura is an intent-aware copilot that takes a user’s natural-language sharing intent plus visual annotations, runs a vision model and a multimodal LLM to flag risks, then uses segmentation and an image generation pipeline to apply nine obfuscation techniques (e.g., inpainting and generative replacement, blurring). Evaluated with 15 participants using four personal images each, it raises risk awareness and cuts perceived risk substantially (expression capture stays stable) while raising authenticity and misuse guardrail concerns.
AI-written summary · about
chapters (8)
- 0:00Introduction to Imago Obscura
- 0:23What is Imago Obscura
- 1:28Why this tool was built
- 2:53How the system works
- 5:27Obfuscation strategies
- 7:12User study findings
- 9:36Limitations and future work
- 10:18Q&A on live streaming
related talks
15mEnvisioning and Mitigating Privacy Risks for Consumer-Facing AI Product Concepts...
09mImageAlly: A Human-AI Hybrid Approach to Support Blind People in Detecting...
40mRene van Veldhuijzen: FOMO Sapiens and the Rise of the AI Empire
36mArt in the Age of Algorithms
55mAI Showdown: The Good, The Glitchy, and The Risky
44mThe uncanny mirror seeing ourselves in AI
01h 07mMinimizing Bias in AI-Generated Images
07mVision: Human-as-the-Unit Privacy Management with AI Agents