Finished in Unreal Early access
Our Unreal automation conforms the MetaHuman face to your head, overrides the skin with the person's own maps, puts on the recommended grooms and picks a body that fits the neck.
Photo → MetaHuman
One photo in. A finished MetaHuman out.
FaceNinja fits a MetaHuman head to the person in your photo, paints their own skin onto it, recommends the hair, brow and beard grooms that match, and fits the neck and shoulders. You get a character, not a bald mesh you still have to finish.
Live 3D · our engine
01 How a head is made
From a holiday snapshot to a rigged head in four steps.
Every image below came out of one run of our engine on one synthetic photo. Nothing was touched up by hand.
Your photo
Laughing, glasses on, side light from a window. One photo is enough; more photos of the same person help.
Neutral reference
We redraw the person facing the camera, mouth closed, glasses off, in flat light, and check it is still the same person before any 3D work.
Fitted MetaHuman head
Three independent 3D models shape the head; we fit MetaHuman topology to their consensus and project her skin at 4K.
Grooms and body
Hair, brows, lashes and beard are matched to MetaHuman grooms, with colours. The neck and shoulders follow hers.
02 Why FaceNinja
Other services stop at the head. We keep going.
Hair, brows and beard from the photo
We read length, texture and colour off the photo and name the MetaHuman grooms that match, with Melanin, Redness and Whiteness values ready to type in.
A neck that belongs to the face
The template's slim default neck is scaled to the person's, and the trapezius line is raised or lowered to match. Broad necks stay broad.
52 ARKit blendshapes, refitted
Every expression shape is transferred onto the new face, so anything that speaks ARKit, from iPhone face tracking to web trackers, drives the GLB straight away. Try them above before you download.
An API you already know
Create a head, wait for it, request a file, fetch a short-lived signed link. If you have built against KeenFace, the steps are the same, so moving over is a small change.
Measured in public
We are building an open benchmark on consenting test people: shape error in millimetres and likeness scores, side by side with other services. Status
03 Hard photos on purpose
Four awkward photos. Four clean heads.
Laughs, glasses, a turned head, a full beard, hair over the ears, harsh light: the things that break photo-to-3D. These four people do not exist; we generated the photos to test the engine without using anyone's face.
Harsh light, deep wrinkles
Light baked into the photo is removed; the wrinkles stay as relief in the normal map.
Glasses and a big laugh
The smile is not baked into the shape and the glasses leave no marks.
Full beard, head turned
The jaw under the beard is recovered; the beard itself becomes a groom recommendation.
Hair over the ears
Ears hidden by hair come from the template; the face is hers, ready for a long-hair groom.
04 Groom recommendations
We tell you which hair to put on.
A MetaHuman head is bald by design: hair, brows, lashes and beards are grooms. We look at the photo and pick them for you.
- Named grooms for hair, eyebrows, eyelashes, moustache and beard, with the next-best alternatives.
- Colour values for MetaHuman's Melanin, Redness and Whiteness sliders, with swatches to judge against.
- A recipe sheet in every download, showing what we measured in the photo and why.
- A hair preview in the browser, so you can judge the likeness with hair before you open Unreal.
Colours are approximate: exposure and white balance in the photo are not undone.
Hair_S_Coil, Beard_S_Full, Mustache_S_Full, Eyebrows_M_Shaded, each with its colour.Template neck
Fitted
05 Neck and shoulders
The body starts where the head ends.
Stick a thick-necked face on a slim template neck and the character looks pasted together. We measure the neck in the 3D shapes and fit width, front and the trapezius line, then recommend a body type that joins it without a seam.
How the neck fit works
06 For developers
Four calls from photo to file.
Upload photos, wait for the head, ask for the format you need, download it from a signed link. Long polling and server-sent events, so you never have to guess when it is done; webhooks for your backend.
- GLB with or without ARKit shapes, FBX, OBJ, STL, PLY, MetaHuman ZIP
- Idempotency keys, so a retry never creates or bills a second head
- Failed heads are refunded automatically
shellbash + curl
# 1. create a head from one or more photos of the same person curl https://api.faceninja.net/v1/heads \ -H "Authorization: Bearer $FACENINJA_KEY" \ -F photos=@me.jpg -F photos=@me-side.jpg # 2. wait for it (returns as soon as it is done, or after 60 s) curl "https://api.faceninja.net/v1/heads/$HEAD?wait=60" \ -H "Authorization: Bearer $FACENINJA_KEY" # 3. ask for a format, 4. download from the signed URL it returns curl https://api.faceninja.net/v1/heads/$HEAD/files \ -H "Authorization: Bearer $FACENINJA_KEY" \ -d format=metahuman_zip
Simplified. The reference has every field and error.
07 Benchmark
Numbers, not adjectives.
We are measuring FaceNinja against other photo-to-MetaHuman services on people who agreed to take part, and we will publish the method, the photos we are allowed to show, and every result, including the ones we lose.
— mmShape error against a reference scan, face region
—Likeness score: face recognition similarity to the photos
— minTime from upload to a finished head
—Test people, each with written consent
The benchmark is running. This section fills in when the first results are ready.
08 Pricing
$15
per finished head
Start with 1 free head when you sign up. Then buy credits for one head or a whole cast; no subscription.
Bring a photo. Leave with a character.
Your free head is waiting. A head takes about 5 to 10 minutes today.