Official launch partner · Pruna AI · from 5 November

P-Video-2-Pro Replace puts a new person in the shot.

Send a video and one or two photos. Pruna AI's character replacement model returns the same scene with the person from the photo: same motion, same camera, same sound, from one second to two minutes.

pruna/p-video-2-pro-replace · 1s–120s · 480p–4k · 24 fps

$0.025per second · 480p
$0.05per second · 768p

Launch price until 17 November, 18:00 CET: half the list rate. After that, $0.05 a second at 480p and $0.10 at 768p.

You need the rights to, or the consent of, every person whose face you use.

SourceOutput · 768p
The photo sentBoxing, 16:9. Fast punches, seen from the side.
What it does

One video in, a new person in it.

Send a video and one or two photos of people. The model puts the person from the photo in place of the person on screen and leaves the rest as it was shot: the motion, the camera, the background, the timing and the sound.

1–120s

The output is as long as the source, up to two minutes. There is no duration to set.

1 or 2

People replaced per call, one photo each. images takes one or two.

~12s

To make each second of video at 768p, about 10 s at 480p, per Pruna.

1:4–4:1

Vertical, 16:9 and cinema sources come back in their own shape.

The scene

Shot as it was.

The camera, the framing, the background, the other people and the timing come from the source. Back views, turns, profiles and fast motion are handled.

The sound

Kept, and the lips follow it.

The source soundtrack goes into the output, and the new person's mouth moves with it. The voice stays the voice of the source, so pick a photo that fits it.

The cast

One photo per person.

With two photos and no instruction, they go to the people on screen from left to right. A short instruction_prompt picks another order, or one person in a group.

Demos

Examples, source and output.

Pruna AI's own clips: the source video and the result play on top of each other. Drag the handle to compare. The photo sent is under each clip. All are at 768p.

For the feed

Vertical clips for ads, fashion and creators. A 9:16 source comes back 768×1344.

SourceOutput
Photo sent: the new person1 photo · no instruction
UGC and ads 7s

Talking to camera

One creator clip, a new face for every market. The speech, the lip sync and the product in hand are kept.

SourceOutput
Photo sent: the new look1 photo · no instruction
Fashion 10s

Runway walk

The whole person changes, outfit included. The walk and the camera stay.

SourceOutput
Photo sent: the 3D character1 photo · no instruction
Character 10s

Fitness coach, as a 3D character

A real person swapped for a stylized character or a mascot. The character keeps its style.

SourceOutput
Photo sent: the new host1 photo · no instruction
Creator 8s

Vlogger

A handheld creator clip with a new host. The output is the person in the photo, not a blend.

Who changes

Two photos for two people, or one sentence to pick one person in a group.

SourceOutput
Photo 1: the new left dancerPhoto 2: the new right dancer2 photos · no instruction
Two people 10s

Two dancers

Two people replaced in one call. With no instruction, photo 1 goes to the left and photo 2 to the right.

SourceOutput
Photo 1: goes to the man on the rightPhoto 2: goes to the woman on the left2 photos · instruction
Two hosts 8s

Podcast

“Replace the woman on the left with the woman from image 2, and the man on the right with the man from image 1.”

The photos are sent in the opposite order. The instruction says which goes where.

SourceOutput
Photo sent: the new barista1 photo · instruction
One of a group 10s

Café

“Replace only the barista standing behind the counter with the person in the image. Keep the customers in the foreground unchanged.”

Only the barista changes. The customers in front stay as they are.

Sound and long takes

The soundtrack is kept and the lips follow it. Turn the sound on.

SourceOutput
Photo sent: the new singer1 photo · no instruction
Music 8s

Singer with a guitar

Singing, strumming and the crowd behind. The song stays and the lips follow it.

SourceOutput
Photo sent: the new chef1 photo · no instruction
How-to 8s

Chef

Hands, a knife and talking to camera, kept in sync.

SourceOutput
Photo sent: the new presenter1 photo · no instruction
Long take 24s

Presenter

One take, 24 seconds. The new face holds from the first frame to the last.

Where it fits

The take is right, the person changes.

Footage you already have, re-cast without a reshoot.

UGC and ads

One clip, a face per market.

Shoot one creator clip, then re-cast it for each market or audience. The script, the timing and the product in hand stay.

Fashion

The whole look changes.

Outfit included. The walk, the set and the camera stay as they were shot.

Characters

Mascots and stylized hosts.

Swap a real performer for a 3D character or a brand mascot. The character keeps its style.

Training and how-to

Reshoot without the reshoot.

Re-cast a presenter, a chef or a coach on takes up to two minutes long, with the original voice.

Price

Billed per second of finished video.

The resolution sets the rate. Until 17 November, 18:00 CET, every rate is half the list price.

Per second480p768p1080p2k4k
Launch, to 17 Nov$0.025$0.05$0.06$0.07$0.09
List$0.05$0.10$0.12$0.14$0.18

A 10-second clip is $0.25 at 480p, $0.50 at 768p or $0.90 at 4k at the launch price.

Rounded up to the whole second of returned video. It is as long as the source, so the source sets the bill. A source longer than 120 seconds is cut to 120.

1080p, 2k and 4k are made at 768p and upscaled with P-Video-Upscale, so they take longer: about 2.5 minutes for 10 seconds at 1080p and 4 minutes at 4k.

One photo or two, an instruction or none, sound kept or not: the price is the same.

Pruna's own list price. The launch row is half of it for jobs finished by 17 November, 18:00 CET. If Pruna changes the list price, the price here follows.

On the invoice you already get. Same key, same monthly bill, split by project or team like the rest of the catalogue. All prices: docs.sociaro.com/static/pricing.

Inputs and limits

What it takes, and where it stops.

What you send

videoRequired. A URL to the source: MP4, MOV or WebM, 1 to 120 seconds, up to 8K, between 1:4 and 4:1. The motion, the scene and the sound come from it. A longer video is cut to 120 seconds.
imagesRequired. One or two photo URLs, one per person to replace. A third photo is ignored.
instruction_promptOptional. Who to replace, or which photo goes to which person, in plain words. Leave it empty when there is one person on screen.
resolution480p, 768p, 1080p, 2k or 4k. Default 768p. 1080p and above are upscaled from 768p.
save_audioOn by default: the source soundtrack goes into the output. Turn it off for a video with no sound.
ignore_audioOff by default. Turned on, the model does not listen to the soundtrack and the lips do not follow it. With save_audio on, the sound is still added.
seedOptional. The same seed repeats a run, to compare settings on one clip.

Worth knowing

You need the rights to, or the consent of, every person whose face you use.
No prompt for the picture, no duration and no aspect-ratio setting. The output is a 24 fps MP4, as long as the source and in its shape.
With two photos and no instruction, the photos go to the people on screen from left to right.
The voice stays the voice of the source. Pick a photo that fits it.
Expressions and lip movements follow the source. When the new person looks very like the one on screen, the model hides the source face so the new identity comes through; the expression then comes from the pose and the audio.
The full reference, with the request and the polling call: the model card and the generations API.
Output size at 768p
Source shapeOutput
16:91344×768
9:16768×1344
21:91536×640
Generation time, in Pruna's tests
ClipResolutionTimeList price
8s768p~1 min 45 s$0.80
10s768p~2 min 30 s$1.00
24s768p~6 min$2.40
120s480p~18 min$6.00
10s4k~4 min$1.80

From the request to the finished video, one request at a time.

Get started

Run it on your Sociaro key.

Sociaro is one of Pruna AI's official launch partners for this model. P-Video-2-Pro Replace is in the catalogue as pruna/p-video-2-pro-replace. 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.