Create an AI character you can keep.
Describe a person once and get a file, not a lucky render — the same face across photoreal, illustrated and 3D, into new scenes and outfits, and on into video whenever you need them to move.
"Mid-twenties, short copper-red curls, heavy freckles, a gold hoop in her left nostril, a pale scar through her right eyebrow. Green canvas utility jacket, collar up."
A character is something you keep.
One face, whatever the style asks for.
Photoreal for the campaign, illustrated for the series, stylised 3D for the game — and the freckles, the scar through the eyebrow, the gold hoop survive all three, because each one was generated from the same reference rather than from the same paragraph. A style is a treatment here, not a different person.
New scene, same person.
Different sweater, different hour, different country, still her. This is the part that decides whether a character is usable: a face that only works in the pose it was born in is a stock photo, and a face that survives a wardrobe change and blue-hour light is a cast member you can write for.
The reference is the asset.
Everything above came out of this one file. That is what you are actually making when you make a character: not an image you liked, but a handle you can pass to the next generation, the next campaign, and the next model — including the video ones, when she needs to move or speak.
Every model that can hold a character.
Built for visual inference.
Serving video is a different problem from serving text — a single request can saturate a GPU, and none of the tricks that made language models cheap apply. Hedra's engine was built for exactly that workload.
Whichever model you choose — ours or anyone else's — it runs on the same infrastructure, behind one key. Which matters more than usual for a character: the reference you pass to an image model this month is the reference you pass to a video model next month, without leaving the key you already have.
Generate with API or Agent
Build with the API
One key, every model. Keep the reference in your own store and pass it with each request.
Make one now.
No code — describe the character, save them, call them back into the next prompt.
FAQs
By being passed back in as a reference image rather than re-described. The reference-driven models hold identity across poses, outfits, scenes and styles — every image on this page was generated from one portrait that way.
Browse all models →