Make the thing you’re imagining
Phantasia turns a few words into a picture, a song, a voice or a short clip. Every result is made by an open model: one anyone can look up, study and run, credited by name on everything it makes.
How it works
- Describe it. A sentence is enough: “a cosy cabin in the snow at dusk”, “lo-fi piano for a rainy evening”.
- An open model makes it. Your request runs on a GPU in the cloud, usually seconds for a picture, about a minute for a song and a few minutes for a clip.
- Keep it, or share it. Everything you make is private until you choose to share it to the gallery. You can download it any time.
The models
PicturesSDXL-LightningCreativeML OpenRAIL++-M
SongsACE-Step 1.5MIT
VoicesKokoro-82MApache 2.0
Short clipsWan 2.2Apache 2.0- Turkmen voice (preview)Meta’s MMS-TTSCC-BY-NC 4.0 · non-commercial
We pick models whose licences allow everyday use, and say so plainly when one doesn’t: the Turkmen voice is a preview for non-commercial use only.
What we care about
- Open, and credited. No black boxes: every picture, song, voice and clip names the model that made it.
- Private until you share. Nothing you make is public unless you put it in the gallery, and you can take it out again.
- No one’s face without them. Pictures you upload to animate are checked first: photos with people’s faces are refused, so Phantasia is hard to use to fake a real person.
- Fair for everyone. It’s free while in beta, with one go per model, per person, per day, so there’s enough for all.
Phantasia is made by Feanor Labs Ltd.