Table of Contents Show
For those who don’t know the tool: KeenFace is KeenTools’ image-to-3D-head service, available through a browser-based Studio and a cloud REST API. It builds on the company’s FaceBuilder research to turn facial photographs into textured head geometry for character creation, digital doubles and further production work.
A familiar name with another useful idea
Some developers have a habit of turning the jobs we quietly dread into tools we actually want to open. KeenTools belongs in that category. We have followed its FaceBuilder and GeoTracker developments, and KeenFace has us leaning towards the screen again. Apparently, we are quite easily turned by a promising head.
The proposition is practical: supply photographs, receive a textured head and spend more of the working day developing the character. KeenTools has the reconstruction down to minutes. We have not timed it ourselves, but we are keen to put it through its paces. Yes, keen (I know, bad punning. I lasted almost two paragraphs before that happened). Getting to the sculpting sooner is an appealing prospect, especially when our own contribution to the reference photography is likely to be a face that clearly needs more sleep.
Start with the reference you have
The character-creation workflow accommodates photographs, archival images and AI-generated character references. Multiple angles help establish head shape, while a single image remains an option when reference is limited.
KeenFace supplies initial likeness and proportions as a base for further sculpting – or just as the “first pass” at the face replacement. KeenTools describes continuing in ZBrush or Mudbox, or using the generated shape to wrap a studio’s own topology. That is an attractive role for the service: get the first head onto the artist’s desk, then let the artist decide where the character goes. We would quite like to get our hands on that starting point, preferably before becoming emotionally attached to a badly placed cheekbone.
For more deliberate digital-double work, Scan Reconstruction returns an FBX scene with solved cameras and focal lengths. Artists can continue refining likeness against matched views in Maya or another compatible DCC. Having the reference cameras accompany the head is a welcome courtesy to whoever receives it next. That person might be us, and future us appreciates the occasional act of kindness.
Capture still deserves some affection
The capture guidelines specify two to fifteen photographs for Scan Reconstruction. Include a frontal view and turned views from both sides, approximately 40 degrees from the front. Keep the subject still, the expression neutral and the lighting even and indirect. A synchronised multicamera rig is the preferred setup.
The face should be sharp, large enough in frame and fully visible. Tie long hair back, uncover the ears and remove glasses where possible. Frames and reflections obscure useful facial geometry; hats and hoods can interfere with head shape. Nobody ever thought of that, surely :P Deep shadows, blown highlights and changing expressions also make reconstruction harder. The reconstruction may be automated, but the subject still needs to cooperate. For our own test, holding a neutral expression while resisting a dreadful head-related pun may prove the more demanding part of capture.
Consistency is a feature worth getting excited about
KeenFace delivers heads in MetaHuman topology, including neck and shoulder geometry. Vertex order and UVs remain consistent between sessions, supporting texture reuse and repeatable downstream processing. That consistency is especially useful when the job involves more than one character. A predictable mesh gives pipeline tools a stable starting point, rather than inviting them to discover a new interpretation of the human head every Monday Morning.
The generated PBR set contains 4K base colour, normal, roughness and specular maps. Texture cleanup removes hair and image artefacts. The launch announcement also describes a bald base head ready for custom grooming, with dedicated grooming features planned for later updates. A clean starting point, then, accompanied by a fairly decisive haircut. We can always add the artistic temperament with the groom.
KeenFace also generates 51 ARKit-compatible blendshapes adapted to each head’s geometry. These cover brows, eyelids, cheeks, jaw and lips, supporting combined expressions, keyframe animation and ARKit-based facial capture. Plenty of scope for raising an eyebrow at our own reference photographs.
Two texture routes, with different intentions
For scanning integrations, KeenFace combines reconstruction, fitting to MetaHuman topology and texture generation in one workflow. Teams can project and blend captured photographs onto the head or generate PBR maps, with optional cleanup. Photo projection preserves the captured appearance; a PBR workflow prepares surface information for subsequent material and lighting work. They serve different purposes, and having the choice is more useful than silently deciding on the artist’s behalf. The scanner integration route exposes reconstruction and processing through the cloud API, allowing capture software to submit images and retrieve meshes and maps.
Into MetaHuman, with the actual steps attached
The dedicated MetaHuman export packages a head template and textures into an archive. The documented route uses MetaHuman Creator’s Conform from Template workflow, followed by adding the base colour and normal maps. After assembling the MetaHuman, the included SRMF texture supplies generated specular and roughness information to its skin material in Unreal Engine. The export includes an FBX head template, 4K base colour and normal maps, and the packed 4K SRMF texture.
Scan Reconstruction additionally combines photogrammetric reconstruction with fitting to the target topology. Its texture options include photo-projected or PBR output, with cleanup enabled or disabled. This is the sort of integration detail we like to see: an identifiable route from reconstructed head to the next application, with the necessary material steps spelled out. We have spent quite enough time staring at exports and hoping the next application shares our interpretation of what we thought that is going to be…
The technical small print earns its section
The API reference puts useful boundaries around the workflow. Default accepts 1–15 images, while Scan accepts 2–15, uploaded as files rather than image URLs or base64. Scan offers PBR and Raw output: Raw retains photographed lighting and gaps, whereas PBR cleanup can fill gaps, remove hair and generate eye textures. You can estimate focal lengths per image or share them across the set, or supply them manually as 35 mm-equivalent values from 1–2000 mm. I struggle to see why anyone would use a 2-meter lens for Portraits.
API downloads offer GLB, OBJ and MetaHuman ZIP, with different contents to account for downstream. ARKit targets require GLB, and the set lacks tongueOut. OBJ retains quads. The MetaHuman ZIP contains a head template without eyes, teeth or blendshapes; the rigged character still needs assembling.
Normal maps for GLB and OBJ use OpenGL +Y, while the MetaHuman export uses DirectX −Y. Data maps need non-colour interpretation. GLB and OBJ geometry is Y-up and faces +Z, with uncalibrated absolute scale. These are small details until a perfectly respectable face arrives looking inside-out or at an unexpectedly architectural size.
Creation and file preparation run asynchronously. The /model endpoint returns JSON; clients need to follow retry-after, then start the signed download within 20 seconds. Idempotency keys handle retries. Studio and API share credits, although neither can open heads created in the other. Raw GLB also defaults to metallic materials, so set metallicFactor to zero for a non-metallic preview. An important distinction if the brief calls for a person rather than a commemorative coin.
Room for an entire cast
The API also opens a route to player likenesses and character customisation. KeenTools thinks about submitting player photographs, generating a head and combining it with a game’s bodies, hairstyles, clothing and accessories.
The game team retains responsibility for character assembly, storage and use. KeenFace handles head generation within that flow. Consistent topology and UVs support reuse of processing and material setups across characters, which is exactly where repeatability starts becoming more interesting than a single impressive preview.
A head start for €15
KeenFace’s current pricing page displays €0.10 per credit, excluding VAT. A head costs 150 credits in either Studio or the API, making that €15 at the displayed rate. New accounts receive 450 credits, enough for three generations.
Downloads are included, failed heads are free and credits do not expire. Studio lists GLB, OBJ, PLY, STL and FBX mesh exports, alongside textures and expression and ARKit blendshapes. The API lists GLB, OBJ, MetaHuman ZIP, ARKit blendshapes and camera data for Scan heads.
We want to try this with our own reference set, follow the head through export and see how it behaves when the expressions start moving. KeenFace brings together likeness, geometry, textures and a documented handoff, which gives us rather more to get excited about than a handsome preview. Before adopting it for a project, we would check likeness, deformation, material interpretation and scale against representative assets. For now, consider our curiosity thoroughly engaged. Three starter generations should at least establish whether we can get our own head into the pipeline without letting it go to our head. (I know, I am punning. Sorry!)
| Product | KeenFace |
| Developer | KeenTools |
| Access | Browser-based Studio and cloud REST API |
| Reconstruction | Default and Scan Reconstruction |
| Geometry | MetaHuman topology |
| Surface maps | 4K base colour, normal, roughness and specular |
| Facial shapes | 51 ARKit-compatible blendshapes |
| Billing | Pay-as-you-go credits |
| Current displayed price | €0.10 per credit, excluding VAT; 150 credits per head |
| Starter allowance | 450 free credits |
Topics
Modeling462 postsStart with a point, end with a world. Follow modelling tools and techniques for building characters, props and environments, with an eye on shapes that will survive the production pipeline.
Rigging148 postsGive digital characters their strings, joints and room to perform. Explore rigs, deformation and animation controls that help a carefully built model bend, move and express a little personality.
Texturing64 postsEvery surface has a story, even the suspiciously clean ones. Explore texture creation, painting and asset workflows that add grain, wear and believable detail to digital characters and environments.
Article details
- Published
- Author
- Bela Beier
- Reading time
- About 8 min
- Word count
- 1,488