Upload 10–15 photos
Selfies, casual shots, different angles. That's all the "dataset" your model needs.
A few selfies in — your own private AI model out. Swaplse trains a LoRA on managed GPUs and generates the same person in any scene you can type. Your endpoint, your face, your call.
SAME FACE · NEW SCENE · EVERY FRAME BELOW GENERATED ON SWAPLSE
From selfies to endpoint
No ML knowledge needed. Three steps, about twenty minutes, and you own a private model that puts you in any scene — forever.
Selfies, casual shots, different angles. That's all the "dataset" your model needs.
A private LoRA trains on a dedicated L40S. Nothing is shared, nothing joins anyone else's model.
Your endpoint goes live. Prompt it from the web app or the API — same face, any scene, in seconds.
FREE GPU MINUTES FOR THE FIRST WAVE · NO CARD REQUIRED
Before you ask
The ones everybody asks before uploading their face — answered with the receipts, live.
Nowhere — and that's the point. The moment training ends, your model is encrypted with a key derived from your account, cut into shards and scattered across separate storage nodes. No single server holds your face. Our engineers see ciphertext — mathematical noise. When you hit delete, the key is destroyed and every shard becomes permanent garbage. Watch it work:
Only you. Full stop. Every model ships inside Orbit — a sealed perimeter around your photos, your weights, your outputs. Nobody else can open it, generate with it, or even know it exists.
ONE CLICK · GONE FOR REAL · NO "30-DAY RETENTION"
No. And Orbit doesn't say it politely. Users, scraper bots, "communities" hunting for private models — it doesn't warn them, doesn't rate-limit them, doesn't ask twice. It throws them out. Every attempt, forever, automatically. Meanwhile you fly in and out like you own the place — because you do. Everyone else gets one answer, on repeat, until the heat death of the universe: ✕ denied.
No. Upload photos, wait ~20 minutes, type scenes in plain words. The API exists for developers — the web app needs zero technical background.
GPU time only, metered: training minutes when you create a model, generation seconds when you render frames. Idle endpoints scale to zero and cost $0. No monthly seat fees.
Private preview
We're onboarding in small batches while in preview. Leave your email — first wave gets free GPU minutes to train and play.