Performer models
A model of one performer, trained to hold the same face and body across different sets, angles and lighting.
Typically includes LoRA weights · trigger tokens · settings card · test sheet
AI model training · adult entertainment
Make it irresistible.
Custom AI models trained on your performers: face, look, motion and voice. You brief us, we train the model, you run it on your own hardware.
- Trained on your own material
- Delivered as files you host yourself
- Scoped on a call before any work starts
Release volumeMore material from a single shoot day.
CustomsFilling requests without booking a shoot.
Back catalogueRemastering older titles to current quality.
LocalisationDubbing into the languages you already sell in.
- Studios
- Paysites & VOD
- Fan platforms
- Talent agencies
- Cam networks
- Companion apps
Services
Six things we train.
Pick one, or build all six around a single performer.
Virtual performers
An identity built from scratch. No real person’s likeness in it, and no release to renew.
Typically includes Identity lock set · variant grid · brand bible · IP assignment
House look
Trained on your own lighting, grade and set language, then applied to new output.
Typically includes Style adapter · preset pack · wired ComfyUI workflow
Video & motion
Adapters for clip work. They target the two things that ruin clips: identity drift and frame-to-frame flicker.
Typically includes Motion adapter · loop-safe presets · upscale chain
Voice
A speaker model for dubs, ADR, audio drops and anything you sell by the minute.
Typically includes Speaker model · response library · dub pipeline · lipsync
Chat personas
A fine-tuned model with a fixed persona and guardrails you configure. Built for fan chat, PPV upsell and companion apps.
Typically includes Fine-tuned adapter · memory schema · guardrails · eval suite
How it works
Three steps.
01
Send us the set
Send your shoot or your archive. If you have neither yet, we write the shot list first.
02
We train it
We curate the material, train the model, then test it against the angles and lighting your source covers least.
03
Take the weights
You receive safetensors and workflow files, and run them on your own hardware.
Pricing
Priced per project.
Scope sets the price, and the condition of your material sets the scope. Nothing is quoted before we have seen what you have.
One model
Per model
A single performer or house-look model.
- Curation and captioning
- Testing before hand-off
- Weights handed over
Studio
Monthly
A standing slot in the training schedule.
- An agreed number of models per quarter
- Shot lists for your own shoots
- Refreshes as the roster changes
Platform
Annual
Model pipelines built into your product.
- Per-creator training at volume
- Moderation tuned to your policy
- Private inference in your cloud
Questions
Asked on every first call.
Who owns the model?
That is settled in the agreement before any work starts. The usual arrangement is that the weights, captions and workflow files go to you, and the agreement also states what happens to your source material afterwards. If your platform or your lawyers need something different, raise it on the first call.
How much material do you need?
Coverage matters more than volume. As a rough guide, a performer model wants a couple of hundred usable stills spread across different lighting, angles and distances. Video wants clean footage rather than a lot of it. Voice wants isolated speech. Where the material is thin, that is what the first pass looks at.
Can you work from our existing catalogue?
Usually. It is where most projects start. The first pass goes through what you have, which titles are usable, and where the gaps are. Sometimes the answer is that the archive will not carry the model you had in mind.
Do we have to host it with you?
No. Most studios take the safetensors and run them on their own machines. Private inference is available if you would rather not run GPUs at all.
What if a performer withdraws?
Agreed up front: who can withdraw, what gets retrained or deleted, and what proof you receive. It is worth settling before the first dataset moves rather than after.
Will it fit our ComfyUI setup?
We target whatever your team already runs: ComfyUI, Forge, A1111, InvokeAI or an in-house pipeline. Workflow files are wired against your own node set rather than a generic one.
Get started