01 /Subject-preserving hair synthesis

PiHair Ultra 9000

Hair,resolved.

Change the hair.
Keep the person.

PiHair Ultra 9000 isolates the geometry of hair from the identity beneath it. Controlled style variation. Consistent subject conditioning. Every strand in context.

Inside the engine

AI-generated portrait study of the same adult woman with a straight dark bob, shoulder-length waves, and a short textured pixie cut. — FIG. 01 / CONTROLLED STYLE VARIATIONONE SUBJECT. THREE CONDITIONS.

Shared portrait condition / Independent style fieldsIllustrative study. AI-generated imagery.

Synthesis enginePiHair Ultra 9000

Reference generation0.3 seconds^*

Inference runtimeNPU-9 / v2.11

02 /Architecture

A separate system
for every strand.

Hair has its own geometry, its own light response, and its own relationship to the face. Ultra 9000 treats those constraints as a dedicated synthesis problem.

01 / EncodeFFE

Follicular Field Encoding

Maps orientation, density, and volume into a continuous style field. Global shape and local strand behavior remain independently addressable.

INPUT → DIRECTIONAL STYLE FIELD

02 / PreserveIBL

Identity Boundary Lock

Maintains a persistent subject envelope around the face, ears, and neckline. Hairline transitions resolve against the original portrait condition.

CONSTRAINT → SUBJECT ENVELOPE

03 / ResolveSSR

Strand-Space Refinement

Reconciles flyaways, fine edges, and overlapping strands with the scene’s lighting. Detail is resolved at the boundary, where it matters most.

OUTPUT → CONTEXTUAL HAIR DETAIL

Design principle

Subject conditioning persists across variations. The style field changes; the portrait remains the reference.

03 /Inference layer

Small latency.
Fine distinctions.

The PiHair NPU-9 Accelerator schedules field synthesis and boundary refinement within a shared execution context. A resident subject condition keeps successive variations close to the metal.

0.3seconds / generation

*Reference profile: one 512 × 512 output, warm runtime, resident portrait condition. Excludes initial conditioning, transfer, and export. Timing varies by deployment profile.

Engine: PiHair Ultra 9000
Acceleration layer: PiHair NPU-9 Accelerator
Runtime release: v2.11
Conditioning mode: Persistent subject envelope
Execution profile: Resident / single subject
Refinement policy: Boundary-adaptive
Output annotation: Embedded generation signature

RUNTIME SERIES / NPU-9REVISION 2.11

04 /Output provenance

An identifiable output.

Each export carries a compact generation signature identifying the synthesis engine, accelerator release, and elapsed generation time.

OUTPUT SIGNATUREPLAIN TEXT / EXAMPLE

PiHair Labs
PiHair Ultra 9000
PiHair NPU-9 Accelerator v2.11
Generation time: 0.3 seconds. AI-generated image.

05 /Technical notes

A few finer points.

What does subject-preserving mean?

The portrait supplies a persistent identity condition. Style synthesis is constrained to a hair-aware region, with boundary refinement handling transitions around the forehead, ears, and shoulders. Occlusions and extreme pose changes can require a fresh conditioning pass.

What does the NPU-9 version identify?

NPU-9 names the acceleration layer. The v2.11 suffix identifies its runtime release, including field scheduling, resident-condition handling, and export annotation behavior. It is separate from the Ultra 9000 engine designation.

How is generation time measured?

The reported interval begins when a style condition enters a resident inference session and ends when the synthesis buffer is complete. Source conditioning, file encoding, and transport are measured separately. The 0.3-second reference applies to the warm 512 × 512 profile described above.

Is a public model or API available?

Interested in using Ultra 9000 on your website? to discuss availability and pricing per 100 images.

Specialized systems. Deliberate boundaries.

PiHair Labs / Computational hair imaging