AI face in 3D

What it does

A photo straight from the front does not show how far the nose, the lips and the chin stand out. The AI model makes a head in 3D from the picture the image model makes of your photos, with its shape and skin. The ped’s own head is shaped after it, and the skin is laid into its texture, on the face from the picture for 3D you approved. The ped keeps its model, its UV map and its eyes, so it blinks, makes expressions and can be changed with the face sliders as before.

The On this device mode is still there and does everything on your computer, without the photo leaving it. It shapes the face after the photo from the front, but not in depth.

Take good photos

Under Your photos you add one to five photos of the same person, from any angles. Ideally one straight from the front and a few from the side.

The AI model does not make the head from your photos directly. FLUX.2 at fal.ai first makes a clean picture straight from the front of all the photos, and only that picture goes to the AI model. Photos from the side help the image model get the nose, the chin and the profile right, and the measuring and the AI that compares the face with the photos use them too. So the face usually comes out more alike with more photos.

The AI model makes a better head from good photos:

• Ideally one straight from the front, with the camera at eye height.

• Ideally a few from the side, in the same light as the photo from the front.

• A neutral face with the mouth closed.

• Even light from the front, without hard shadows.

• Without glasses, a cap or anything else that covers the face.

• With the hair away from the forehead and the face.

• Only one face in the picture, ideally against a plain background.

Make the AI head

1. 01

   Select the head, press Face in the inspector and then Create face from photo. Choose the AI 3D (sends the photo) mode at the top of the panel.

2. 02

   Tick both boxes: that it is your own face or that you have permission to use it, and that the photos may be sent to fal.ai and Google (Gemini) and the picture for 3D to fal.ai and Tripo. You tick them every time.

3. 03

   Press Add photos under Your photos and choose one to five photos. You can choose several at once and add more afterwards. Each photo shows small with a button to remove it.

4. 04

   Press Make picture for 3D. The photos are sent to fal.ai, where FLUX.2 makes a clean picture straight from the front of them. It takes about 10 to 30 seconds and costs no credits.

5. 05

   The picture shows with the question Does it look like you?. If it does not look like you, press Make a new picture. You can make three pictures per try and ten an hour, and under the picture it says how many you have left. If you add or remove a photo, the picture goes, and you make a new one.

6. 06

   If you have added only one photo, the panel shows a yellow warning. Tick I understand, or add more photos, before you can start.

7. 07

   Press Make face (50 credits). Only the picture for 3D is sent to the AI model, and the credits are taken at once.

8. 08

   Wait while the head is made, about four minutes. The panel shows where the job is and how long it has run, and Cancel the job is there all along.

When you make a new ped from scratch, Face from photo is the first step of the base mode, if AI 3D is turned on for your account. There you do the same as here.

The face is shaped by itself

1. 01

   When the AI head has been fetched, the ped’s head is shaped after it by itself, and the skin is laid on. Then it is measured against the photos and compared with them (see below). With Include hair the hairstyle is chosen from the photos last. It takes a little while, and the panel shows the steps. Then the ped shows the new face.

2. 02

   Turn the ped and look at the profile. That is where it differs most from the On this device mode.

3. 03

   Choose what comes from the photo. The four boxes are ticked to begin with, and the ped changes at once when you untick or tick a box, with no new job.

4. 04

   Press Apply. The shape and the skin are kept as one step in the history, the hairstyle as a step of its own after it, and with Skin colour from the photo the body’s skin as a step last.

The face’s skin is laid from the picture for 3D you approved, since it shows the face as it should look. The shape comes from the AI head. Where the picture does not reach, the sides of the face, the jaw and the scalp, the skin comes from the AI head, in the picture’s skin tone so the two meet without an edge. If you open a saved AI head later, the picture is no longer there, and all the skin then comes from the AI head.

The skin becomes a layer of its own, Face from photo, in the texture editor, just as in the On this device mode, and the shape shows in the Face panel. The shape and the skin are one step, so Undo takes away both at once. The hairstyle and the body’s skin are steps of their own after them, so they go first, the body’s skin before the hairstyle.

Measured and compared with the photos

While the face is shaped, it is also measured against the photos. The eyes, the nose, the mouth, the chin and the jaw are read from the front on the picture for 3D and in profile on your photos from the side, at most one each way, and on the ped’s head. The face sliders are set so the measures match better. If the face from the front does not come nearer the picture, it is kept as it was shaped. A photo that is not read as a profile is not used for the measuring.

Then an AI from Google (Gemini) compares pictures of the ped’s head, from the front and from the side, with up to three of your photos, not with the picture for 3D. It judges only the face’s shape and features, not the hair, the skin colour or the light, and suggests changes to the sliders. It runs at most three rounds, and the version the AI found most alike is kept.

Above the choices it says how alike the AI found the face, for example Likeness: 7/10, and in one sentence what differs most. 10 is very alike. The score is the AI’s judgement and costs nothing extra.

If the comparison cannot be done right now, the panel says so, and the face is then set to match the measurements.

The four choices

The choices say what comes from the photo. They never change the face’s shape.

• Include hair: on, the game hairstyle you choose among the suggestions is put on, and no hair is painted on the scalp under it, since the hairstyle covers it. If there is no hairstyle to choose, the hair’s colour is painted on the scalp instead. Off, the scalp gets the ped’s own skin, and the hairstyle is left as it is.

• Include beard: on, the beard and the moustache from the photo come along. Off, the ped keeps its own skin there.

• Include eyebrows: on, the eyebrows from the photo come along. Off, the ped keeps its own eyebrows.

• Skin colour from the photo: on, the whole head gets the photo’s skin colour, and the whole body’s skin follows, the arms, the legs, the hands and the feet, in the editor too and on any ped. Only the skin changes, not the clothes. Off, the head keeps the ped’s skin colour, with the photo’s features and detail, and the body keeps its own.

Include hair also suggests three game hairstyles like the hair in the photos, and you choose one of them. How that works is in the next section.

The hairstyle from the photos

With Include hair an AI from Google (Gemini) suggests three game hairstyles like the hair in your photos, once the face is shaped and compared. It sees the library’s hairstyles as large numbered pictures of the hair and the face, each one only once, sorted by length: short, medium and long. It first describes the hair in the photos and judges only the hair: the length on top and at the sides, its volume, whether it is curly, wavy or straight, whether the sides and the back are cut short, and how the fringe falls. It also chooses the nearest of the game’s hair colours.

Under the box you see the three suggestions, the most alike first and chosen, each with its picture, its name and a sentence on why. Press another suggestion to choose it. Under the suggestions is the colour, with the game’s hair colours in a row where you can choose another.

Apply puts the hairstyle in the hair’s place, in the chosen colour and with the game’s own fade at the hairline, as a step of its own after the face, so Undo takes the hairstyle off first and keeps the face. You can change the hairstyle and its colour as usual. In the base mode it is free, and in the editor it costs as a new garment when you export.

If the AI cannot choose, or the photos are no longer there, the three hairstyles in the library most like the AI head’s hair by its shape are shown instead to choose among, in the nearest of the game’s hair colours, and the panel says so. Without the library no hairstyle can be chosen, and the panel says that too.

The AI head is kept on your device

Once the AI head has been fetched it is deleted at fal.ai. So it is kept in the browser on this device instead, until you press Apply or Throw away the AI head. It is never uploaded anywhere.

If you close the panel before you are done, the head is still there the next time you open Create face from photo on the same head. The face is then shaped again by itself, and it costs nothing more.

Throw away the AI head removes the kept head from the device. A new head costs a new try. Delete the project and the head goes with it.

Where the photos go

In this mode the photos leave your computer. They are sent through Pedify to fal.ai, where FLUX.2 from Black Forest Labs makes the picture for 3D. Only that picture is then sent through Pedify to fal.ai, which runs the AI model for 3D, and on to Tripo, which made the model. fal.ai and Tripo are outside the EU, and Google’s servers may be too.

Pedify does not keep the photos, nor the picture for 3D. They are passed on at once and never written to any database or log.

fal.ai keeps the picture and the head while the job runs. Pedify asks fal to delete them as soon as the head is fetched, and when the job fails or is cancelled. Anything still left at fal can only be removed by fal’s support.

How Tripo handles the picture is governed by Tripo’s terms. More about it is on the Privacy page.

For the comparison up to three of your photos are sent again, through Pedify, to Google (Gemini API), together with two pictures of the ped’s head. It happens right after the job and at most three times per job. With Include hair up to three of the photos are sent once more, together with pictures of the game’s hairstyles, so Google can suggest three hairstyles like the hair, at most twice per job. Pedify keeps neither the photos, the pictures nor Google’s answers.

What it costs

An AI head costs 50 credits per try. The credits are taken when you start, since every try costs real money at fal.ai.

If the job fails, takes too long or you cancel it while it waits in the queue, you get the credits back by themselves. If you cancel after the head has begun to be made, it costs all the same.

The credits are not paid back for a result you are not happy with, since the job has cost just as much at fal.ai. The pictures for 3D, the measuring and the comparison with the photos cost nothing extra.

The head then counts as paid, so when you export nothing more is taken for the face or the hair from the photo.