Hedra

Sweet spot: a creator who needs a one-off talking head from a single image — an illustrator giving voice to their character, an educator narrating with a custom mascot, a podcaster making a video edition with stylised art instead of webcam footage. Character-3 is the first model where this actually works at quality high enough to publish. The honest caveats. Source-image quality is the single biggest predictor of output quality — front-facing, well-lit, neutral-expression portraits work; oblique angles or busy backgrounds fail. Long clips drift; plan for 10–30 second segments stitched together rather than a single 5-minute take. Commercial-use rights only kick in on paid tiers, and the credit economy is real — heavy iteration on a single shot can burn 20+ credits. What to pilot. Pick three source images you actually want to use, generate a 20-second clip per image, and judge whether the lip-sync and expressiveness clears your bar. If yes, Hedra is irreplaceable for your workflow. If the result feels uncanny on your specific image style, pre-trained avatar platforms like HeyGen will be more consistent (at the cost of much less flexibility).

Hedra is a San Francisco-based AI video startup whose flagship is the Character-3 model, an audio-driven character animation system. Feed it a single still image (a photo, a portrait, an illustration, a 3D render) and an audio file (recorded voice, AI-generated speech, even a song), and Character-3 produces a video where the subject speaks the audio with lip-sync, head motion, and expressive eye and brow movement that holds up for 30+ second clips. Architecturally Character-3 is the company's third major checkpoint. Earlier Character-1 and Character-2 versions were lip-sync-with-some-head-motion. Character-3 introduced full-body awareness, gesture inference, and the foundation-model scaling that lets it generalise across photo-realistic faces, illustrated characters, animals, and stylised art. The 2026 product wraps the model in a web studio with a built-in voice library (ElevenLabs voices integrated), script-to-video flow, scene composition, and an export pipeline. Where it sits in the market: Hedra vs HeyGen vs Synthesia is the cleanest comparison. HeyGen and Synthesia are pre-trained avatar libraries — pick from a roster of stock or custom-cloned avatars and feed them scripts. Hedra is fundamentally different: any image becomes an avatar in seconds, no studio recording required. That makes it the right pick for one-off characters, illustrated speakers, period-piece historical figures, animal narrators, and anything where you need a single talking head you do not have a video clone for. HeyGen / Synthesia win when you need the same realistic clone across thousands of videos with strict consistency. Funding-wise Hedra has raised from a16z and is one of the most-watched 2025 video generation startups; the product is genuinely good and the model improvements between releases are visible.