On-device processing
Inference and prep on your workstation or studio box.
REC · local perimeter
Image AI that never leaves the machine
Fine-tunes, batch jobs, and generation on local GPUs. Frames stay inside your studio — not a third-party train log. Built around RTX 5090-class throughput.
local GPU only no train log export weights stay on-box faces never leave LAN
RTX 5090-classstudio throughput on private iron
Darkroom modeinference without egress by default
Air-gap pathdocumented offline train loops
Studio-dark by design
Architecture assumes a private local pipeline first. Cloud is an exception you choose — never the only path labeled “private.”
No cloud-only path
We refuse architectures that require upload to stay “private.”
Custom models
Brand and product LoRAs trained where the data already lives.
Studio-grade GPU
Real throughput on high-end local cards — including RTX 5090-class.
What we set up
Studio
Local gen stack
Comfy / A1111 / private stacks hardened for a team.
Train
Brand model training
Style and product LoRAs without shipping the set offsite.
Batch
Batch pipelines
Folder-in, folder-out processing with audit logs.
Proof
Privacy review
Prove images never hit third-party endpoints.