MUMBAI, India, Feb. 6 -- Intellectual Property India has published a patent application (202521121522 A) filed by Prof. Dr. Sandip Ramrao Shinde; Kedar Balaji Sathe; Aditya Anupkumar Sakhare; Samarveer Sushant Moray; and Prathamesh Shivaji Salokhe, Pune, Maharashtra, on Dec. 4, 2025, for 'chameleon-ai.'
Inventor(s) include Prof. Dr. Sandip Ramrao Shinde; Kedar Balaji Sathe; Aditya Anupkumar Sakhare; Samarveer Sushant Moray; and Prathamesh Shivaji Salokhe.
The application for the patent was published on Feb. 6, under issue no. 06/2026.
According to the abstract released by the Intellectual Property India: "A unified multi-task framework named Chameleon-AI for simultaneous blind face restoration, face inpainting, and video colorization with temporal coherence. The framework modifies the Stable Video Diffusion architecture by incorporating task-specific embeddings, unified latent regularization, facial prior learning through landmark prediction, and a self-referred refinement mechanism that ensures smooth temporal consistency across frames without requiring explicit optica! flow computation. Identity preservation is achieved through reference-based identity encoding, while shared latent representations enable all three tasks (restoration, inpainting, and colorization) to be performed efficiently within a single model. The unified design reduces computational overhead by approximately 67% compared to running separate task-specific models sequentially, while delivering superior quality and identity fidelity. The framework includes a practical web-based deployment that allows users to upload degraded videos and receive temporally coherent, high-quality restored outputs suitable for historical footage restoration, low-quality video enhancement, and professional video editing applications."
Disclaimer: Curated by HT Syndication.
