A new system from Peking University improves on the state-of-the-art for neural face-editing, offering more faithful expression manipulation and more disentangled editing of facets such as hair and eye color, among other attributes.
New research could improve the ability of augmented reality (AR) systems to convincingly insert synthetic objects into scenes, by studying the currently complex ways that they are matted and occluded, and simply ‘guessing’ what the best results would be.
A new project offers a way to cleanly transfer motion into synthesized video, promising a new level of realism and verisimilitude – if the code is ever actually released.
A new collaboration between the University of California, Stanford, and NVIDIA has proposed a system capable of turning people into neural representations in real time, using Neural Radiance Fields (NeRF) and Generative Adversarial Networks (GANs).
Deepfakes and AI-edited faces are often too ‘perfect’, which can help deepfake detection systems to identify them. But a new academic collaboration has devised a way to inject real-world ‘flaws’ into fake or altered images so that they are much harder to distinguish from real photos.
A new text-to-video system backed by NVIDIA can create temporally consistent, high-res videos by injecting video knowledge into the static text-to-image generator Stable Diffusion – and can even let users animate their own personal DreamBooth models. However, the system does not seem likely to get a public release, as the ongoing retrenchment around the data that fuels image synthesis ramps up under legal pressure.
Adobe has invented a way of injecting people’s identities into Stable Diffusion as custom characters that out-competes former methods such as DreamBooth and Textual Inversion, while running at 100x the speed of those former methods. However, all indications are that the system is intended for corporate use, or as an adjunct service to Adobe’s emerging and IP-friendly generative services, such as Firefly.
Across people, and particularly across cultures, facial expressions are not always accurate indicators of how we are feeling. How then can we create AI systems that can accurately change facial expressions without descending into anime-style generalizations about facial disposition? One new project aims to accomplish this by using human evaluation for image pairs – a rare approach to a topic that is going to be of increasing interest as the quality and diffusion of neural facial synthesis grows.
A new method uses Generative Adversarial Networks (GANs) to not only offer high quality facial editing, but even to convincingly change the orientation of the subject’s face in respect to the original video. Additionally, it does face-swapping (deepfakes) very well.
A new system offers better recognition of faces that are very small in a surveillance video or CCTV stream, which could increase the likelihood of more accurate and reliable identifications in court cases, as well as being a benefit to the VFX industry.
