Research at Darwin
Our goal is simple: to make video editing so intuitive, fast, and powerful that it feels interactive. The AI video & audio research we pursue at Darwin is at the heart of this vision.
Research
Darwin Evo, our SOTA video-to-video foundation model is the core of Darwin’s magic. We develop pipelines that enable real-time, seamless transformations of video elements using a combination of open-source and proprietary models. From face-swapping and full body replacements to object-based edits and text-driven changes, our tools open up endless possibilities for interactive video. Whether it’s altering characters, replacing objects, or making frame-perfect edits, our systems deliver unmatched responsiveness and fidelity.
Audio is the soul of storytelling, and our audio-to-audio research ensures that it can be just as dynamic and interactive as video. We’ve built pipelines that enable seamless dialogue-to-dialogue transformations, allowing creators to change tone, language, and even delivery style with incredible accuracy. Lip-syncing technology further ensures that every change looks natural and believable, down to the movement of a character’s lips.
To truly unlock the full potential of interactive experiences, we’ve developed the next frontier: World and Character Grounding. Drawing inspiration from retrieval-augmented generation (RAG), we’ve designed pipelines that anchor our models to specific worlds, characters, and IPs. Whether it’s generating Iron Man with flawless fidelity to his Marvel universe or maintaining a cohesive aesthetic for custom environments, our research delivers unparalleled immersion.