Person 7s
Loft
“I moved the chair twice. It looked worse. If someone sits in it, fine. I'm not guarding a chair.”
Asked: turn toward the red chair halfway through, then back to the lens.
New · Pruna AI · from 22 October
P-Video-2-Pro Avatar makes a still talk.
One portrait, plus your audio or a script. Pruna AI's talking-video model returns an MP4 where the lips and the expression follow the soundtrack, from one second to three minutes.
$0.02per second · 480p
$0.04per second · 768p
What it does
A talking video from one still, in one call.
Send a portrait and either an audio file or a script. The model returns a single continuous shot of that face speaking or singing, with the mouth and the expression following the soundtrack.
1s–3 min
The soundtrack sets the length. There is no duration to choose.
30 · 10
Built-in voices and languages, for when there is no recording.
24 fps
MP4 at 480p or 768p. The portrait sets the frame.
~1s
To generate each second of video at 480p, in Pruna's benchmark.
The portrait
Sets the face and the frame.
People, animals, anime, game characters, paintings and sculpture. A 3:4 still comes back at 768×1024, a 9:16 still at 768×1344 — the image decides the shape.
The soundtrack
Your file stays the voice.
Upload a recording — speech or a song — and it is kept in the finished MP4, not replaced. No recording yet: send a script and pick one of 30 voices in one of 10 languages.
The direction
Picture and delivery, apart.
A prompt for the picture: gesture, head turn, camera, what stays still. A separate delivery note sets tone and pacing and is never read aloud. A seed repeats a run.
Demos
Examples. Turn the sound on.
One still and one soundtrack each. Under each clip, what was said and what was asked of the picture.
Any character
A person, an animal, an anime still or a stone head. Whatever the portrait is, it talks.
Animal 5s
Fox
“I took the good cushion. The other one smells like the dog. That's the whole argument.”
Asked: flick one ear, tilt the head aside, then back. The fur and the jacket stay.
Anime · Japanese 5s
Anime
“セーブはボスの前。死んだあとじゃ遅いよ。今日はこれだけ。”
A talking still: the mouth carries the line and the frame holds.
Sculpture 7s
Moai
“They offered me eyes in the museum. I declined. I've managed without them.”
A face that never had eyes stays that way.
Direction
Gestures, head turns, standing up, pointing at a product. The scene can be part of the performance.
Product 7s
Kettle
“That's the kettle. One liter, the handle stays cool, and the spout is narrow, so you can actually aim it.”
Asked: gesture toward the kettle, then rest the hand. An ad can point at the product.
Painting · French 11s
Louis XIV
“Les enfants, à Versailles je me levais avant tout le monde. Pas par vertu. Si je dormais, la cour décidait sans moi.”
Asked: start seated, stand by the last word. A painted full figure can get up.
Painting · French 5s
Mona Lisa
“Vous restez longtemps. Le paysage, je le connais par cœur. Si vous partez, fermez doucement.”
Asked: uncross the hands and lift one. The landscape stays.
Scene · Japanese 4s
Rain
“終電、逃した。傘は店に置いてきた。まあ、いいか。”
Asked: rain falls, the neon flickers, a small turn of the head. The wet street and the face stay.
Delivery and language
Same still, same words, same voice. Only the delivery note changes. Then French and German.
Delivery 9s
Quiet
“We ship on Thursday. If the build is red, we slip it. I'm not going to pretend otherwise.”
Delivery note: quiet, slow, soft and private.
Delivery 7s
Bright
“We ship on Thursday. If the build is red, we slip it. I'm not going to pretend otherwise.”
Delivery note: bright, quick, amused — like telling a friend good news.
French 9s
Salon
“J'ai laissé la fenêtre ouverte pour la pluie. Ma sœur déteste ça. Moi j'aime le bruit, et je ne ferme pas.”
Calm, at 768p.
German 5s
Club
“Der Einlass sagt, ich bin zu früh. Ich bin pünktlich. Das ist sein Problem.”
Dry, at the door.
Songs, and one long take
Upload a song and the body goes with it. Or let a script run for fifty seconds in one shot.
Song 16s
Chorus
“The light is pink, the room is blue. I came in late, I'm staying too.”
Asked: sway and tip the head with the song, neon pulsing.
Song · Mascot 10s
Owl
A short playful chorus.
Asked: sing the whole time, beak opening with the words, and dance with the wings.
Song · Painting 10s
Degas
A short pop chorus.
Asked: sing from the first second and let the tutu and the raised arm move.
Long take · 480p 50s
Fifty seconds
“The soundtrack is a file you already have, or a script, and the video is as long as that soundtrack.”
Script only. The face and the room hold from open to close. The ceiling is three minutes.
Where it fits
A presenter, from a still.
Anything that needs a face to say something, without booking the face.
Product ads
A presenter who points at the product.
Variants per market and per audience from one still and a set of scripts, with no reshoot between them.
Localisation
Same face, every language.
Ten built-in languages, or upload the dub you already have. The identity holds across every version.
Characters
Mascots that speak and sing.
Anime, game characters, brand mascots, paintings and sculpture. If it has a face in the still, it can perform.
Explainers
Three minutes in one call.
Updates, onboarding, course modules. The length follows the audio or the script, up to three minutes.
Price
Billed per second of finished video.
The soundtrack sets the length, and the length sets the price. Voice, language and prompt rewriting do not change it, and there is no separate charge for the portrait or for text-to-speech.
| Output | 480p | 768p |
|---|---|---|
| Per second | $0.02 | $0.04 |
| 10 seconds | $0.20 | $0.40 |
| 1 minute | $1.20 | $2.40 |
| 3 minutes | $3.60 | $7.20 |
The lowest per-second price in Pruna's comparison of nine talking-video APIs, at 480p.
Faster than the clip runs. About 0.97 seconds of generation per second of video at 480p, and 2.2 at 768p, in the same benchmark.
Direct relationships with model providers. Our direct relationships with model providers help us secure better commercial terms for our clients, early access to new models, and access to models that are not publicly available.
On the invoice you already get. Same key, same monthly bill, split by project or team like the rest of the catalogue.
Inputs and limits
What it takes, and where it stops.
What you send
PortraitRequired. JPG, PNG or WebP, up to 20 MiB. It is the identity, and its shape sets the frame.
AudioSpeech or a song, up to 64 MiB. Under one second is rejected; over three minutes is cut at three. Wins over a script.
ScriptWhen there is no audio: the words, a voice and a language.
Picture promptGesture, camera, what stays still. Enriched from the portrait by default, or used exactly as written.
Delivery noteTone, pacing, accent, emotion. Short, and never spoken.
Resolution480p or 768p.
SeedRepeats a run. It does not change the person — the portrait does.
Worth knowing
Not 720p or 1080p.
No duration and no aspect-ratio setting. A still outside the model's range is cropped to fit.
One continuous shot. A second scene is a second call.
Long clips are assembled in parts. A change of room, or a wide static background, can show a seam.
Not text-to-video: a prompt on its own will not run it.
Built-in languages: English (US), English (UK), Spanish, French, German, Italian, Portuguese (Brazil), Japanese, Korean and Hindi — with 30 voices to speak them.
Get started
Run it on your Sociaro key.
P-Video-2-Pro Avatar is in the catalogue as pruna/p-video-2-pro-avatar from 22 October, 18:00 CEST. With a key already, it is a new model name in your call. Without one, tell us what you want to run and we come back with a key, usually the same day.