Kiss.photography · v0.1 private alpha

Reconstruct the versionof you that wins.

An AI system that doesn't edit your photos. It generates the photographs you would have taken if every variable had aligned.

Built on FLUX · trained on Lambda Labs · identity-preserving · off-chain generation

Reconstruction

The output is not an edit. It is a reconstruction.

Reconstructed output: professional portrait with controlled studio lighting and sharper focal planeInput reference: casual candid portrait used as identity anchor
Scrub
Input → 32 candidates generated → ranked → top 6 selected
Pipeline

How it works

01Upload

Ingestion normalizes color profiles, pose metadata, and lighting cues. Identity is embedded once and frozen as a hard constraint for every downstream candidate.

02Generate

FLUX emits 16–32 photoreal candidates per pass. Each is a full frame, not a patch edit. The spread is deliberate variance, not random noise.

03Rank

A composite score orders outputs for aesthetic coherence, predicted engagement, identity lock-in, and pairwise diversity. You rank what matters; the model enforces the guardrails.

04Deploy

You receive a ranked set ready for profiles, decks, and launch surfaces. Outputs ship with lineage metadata so teams know what was generated versus archived.

Specifications

System surfaces without the brochure language

system_architecture.jsonread-only
{  "diffusion_backbone": "FLUX family · latent flow matching",  "identity_preservation": "Frozen embedding · landmark regularization · texture coherence",  "ranking_signals": "Aesthetic · engagement · identity lock · diversity",  "training_infrastructure": "Lambda Labs GPU clusters · deterministic job graph",  "deployment_stack": "Isolated generation workers · signed delivery URLs · optional chain attestations"}
Ranking

Composite score, explicit weights

R = + + +

w₁Aesthetic

Composition, color discipline, and print-grade sharpness scored against a calibrated reference set. Not “beauty.” Fitness for context.

w₂Engagement

Predicted attention and click-through shaped by historical performance surfaces. Calibrated conservatively to avoid meme drift.

w₃Identity

Embedding distance, landmark stability, and texture coherence. The constraint that keeps the output attached to a single human.

w₄Diversity

Penalty on near-duplicate poses and redundant backgrounds across the ranked set. You receive a spread, not a clone strip.

Weights are not magic. They encode which environments you plan to enter and which failure modes you refuse to tolerate.

Allocation

Credits before retail pricing

Standard

120 credits / month

~30 full reconstructions (4 candidates / credit)

  • Single identity lock per seat
  • Standard ranking profile
  • Email support · 48h SLA
  • Signed delivery links
Most selected

Pro

420 credits / month

~105 full reconstructions (4 candidates / credit)

  • Three identity locks with shared policy
  • Custom weight presets per channel
  • Expedited support · 12h SLA
  • Batch export to DAM

Power

1200 credits / month

~300 full reconstructions (4 candidates / credit)

  • Org-wide policy graph
  • Dedicated ranking review UI
  • Private throughput reservation
  • Optional on-prem routing

Dollar amounts finalize at general availability. Current tiers communicate capability density, not a public price sheet.

We are not in the business of making people look better. We are in the business of predicting which version of a person performs best in the environments where they will be evaluated.

— Hated By Many LLC, founding document

Admission is limited. The waitlist is not.

One working email. No marketing drip. We ship an invite when the throughput graph says we can keep promises.

We will not email you until your invitation is ready.
FAQ

Direct answers