Written for a model considering applying
What is a LoRA, actually?
No jargon left unexplained, no hand-waving. This page walks through what the file is, what it isn't, how yours would get made, and exactly how your consent controls it — every time, forever.
Here on a fan's business instead? Jump to the fan version.
The instruction card, not the camera
A LoRA is a small add-on file — usually a few hundred megabytes — that teaches an already-existing, general-purpose image generator (the "base model") one specific thing. In Loralicious's case: her.
Picture the base model as a professional camera that already knows how to photograph anyone, anywhere, in any light. A LoRA is a small instruction card clipped onto that camera. It doesn't rebuild the camera or replace the film — it just tells it: "when someone asks for her, render her face and figure this way." Take the card off, and the camera still works exactly as before. It simply doesn't know her anymore.
The base model
The big, general-purpose generator. Huge, capable of drawing almost anything — but it has never seen her until her LoRA is attached to it.
Her LoRA
The small card. Tiny compared to the base model, and useless on its own — it only means anything clipped onto the bigger system, and only within the limits she set.
What it is not
This is the section that answers the nerves directly, before the technical stuff.
Not a folder of her photos
It's a trained file, not an image library. Nobody browses through her source pictures — the file describes a style and a likeness, and that's all it holds.
Not her pictures, resold
The photos she contributes are never sold, packaged, or handed to a buyer. They exist to train the model and then to sit in the record — not to circulate.
Not a person
It doesn't think, feel, remember, or decide anything. It's a mathematical description of appearance and style — closer to a very specific filter than to a mind.
Not something a buyer keeps
Every generation happens on Loralicious's own systems, checked against her consent record on the spot. Nobody downloads the LoRA and runs it themselves.
How hers gets made
Four steps, start to finish. Nothing here is automatic until she's signed off on it.
Step 1
She supplies photos
Uploaded privately, after she's signed her release. Nothing gets trained on before that signature exists.
Step 2
Dataset preparation
Her photos get organized, cropped, and tagged so a training run can use them. Only her photos go into this set — nothing scraped, nothing borrowed.
Step 3
Training
Her dataset and the base model run through a training process together. Training happens on her images and nothing else — that's the whole input.
Step 4
One small file
The output is her LoRA — a few hundred megabytes, and inert without the consent gate sitting in front of it on Loralicious's own systems.
What it can and cannot do
Honestly, both directions.
It can
- Generate new images, video, voice lines, or chat replies that look and sound like her — inside whatever tone, style, and nudity level she's actually consented to.
- Be narrowed, paused, or withdrawn by her at any time, taking effect on the next request.
- Be retrained or refined over time, on more of her own images, so the likeness stays current.
It cannot
- Produce anything outside what she's consented to — if she granted partial nudity, full nudity simply isn't an option the system has.
- Be taken off Loralicious's systems and run somewhere else. There's no export, no download, no copy that leaves.
- Know her personally. It has no memories, opinions, or awareness — it renders style and likeness, and that's the entire ceiling.
Her consent gates every use of it
Not a checkbox at signup — a set of dials she sets herself, checked against every single generation before it's allowed to run. Here's each one, in the exact wording the system uses.
Face
Whether her actual face may show up at all. Body-only means it never does.
- Full face
- Body only
Coverage
How much of the body may be uncovered.
- Clothed
- Partial nudity
- Full nudity
Tone
How suggestive the content is, independent of nudity.
- Wholesome
- Playful
- Naughty
- Raunchy
Styles
The aesthetic treatment — photoreal carries the most weight, so it's opted into, not assumed.
- Photoreal
- Editorial
- Pin-up
- Anime
- Cartoon 3D
Where it's used
Images, video, voice, and messaging are each their own decision — one doesn't imply another.
- Images
- Video
- Voice
- Messaging
Any combination she wants, set separately per surface. A generation request is checked against her current record on the spot — if it asks for more than she granted, it doesn't run. That check happens every time, not once at signup.
She can narrow any of these, or withdraw consent entirely, at any time from her dashboard. It takes effect on the next request — not on tonight's batch job.
What she earns
Every generation made with her LoRA pays out against her record. She takes 50% — the largest single share, and it doesn't shrink as her consent broadens. Broader consent changes her payout level, never her percentage.
Payment processing fees come off the top first, before anything is divided — so her 50% is 50% of what's left after fees, not 50% of some larger number that gets whittled down later. The split is enforced in code from a single definition, so it can't drift between the contract, the dashboard, and the payout.
A plain-English glossary
Six words, one sentence each.
- LoRA
- The small add-on file described above — trained on her, and only meaningful attached to a base model.
- Base model
- The large, general-purpose generator that already knows how to render people and scenes in general — just not her specifically, until her LoRA is attached.
- Dataset
- The set of her source photos, prepared for training. Nothing else goes into it.
- Training
- The automated process that studies her dataset and produces the LoRA file. It runs on Loralicious's own systems, once per version of her model.
- Generation
- One output — an image, a video clip, a voice line, a chat reply — made on demand with her LoRA and a base model, gated by her current consent every time.
- Prompt
- The instruction that describes what a generation should show. Every prompt is checked against her consent before anything renders.
For fans
What you're paying for is a generation — one image, video clip, voice line, or chat reply, made on demand, inside the limits she set herself. You're not buying a photo of a real, non-consenting person, and you're not buying ownership of anything. Nobody downloads her LoRA and takes it home; every request runs through the same gate hers does.
She agreed to this specific kind of content, in writing, before it existed — and she gets paid every time it's generated: 50% of net revenue, the largest single share of what you spend. She can also change her mind about future generations at any time, which is the whole reason the gate checks on every request instead of once at signup.
The questions you'd actually ask
The awkward ones, answered directly.
Can someone download me?
No. Nobody receives the LoRA file itself, and it doesn't work outside Loralicious's own systems. What a person gets is a generation — one image, video clip, voice line, or chat reply, made on request — never a copy of the model, and never the source photos.
Can it make things I said no to?
No. Every generation is checked against your current consent record before it's allowed to run. If you didn't grant a category, the system doesn't have it to give — that's a hard gate in the pipeline, not a request someone can talk their way past.
What if I change my mind?
Narrow or withdraw consent any time from your dashboard. It takes effect on the next request, not on some batch job overnight. What's already been generated is a separate, harder question the platform hasn't finished answering yet — worth asking about directly before you sign anything.
Do you work with men?
Not yet. Loralicious is launching with women only — not as a statement about who deserves this, but because getting consent, verification, payouts and takedowns right is easier to prove in one lane before widening it. Everything underneath is already built gender-neutral: the consent record has no gender field, and nothing in the system assumes a woman. Opening to men is a question of when the first lane is working, not of rebuilding anything. If you're a man reading this and interested, the honest answer is: not today, and we'd rather say that than take your photos and stall.
Can people tell it's AI?
Sometimes. A single still of a real person from a well-trained model can be genuinely hard to call; hands, small text, reflections and anything moving for more than a few seconds are still where it gives itself away. That line moves every few months and anyone who tells you otherwise is selling something. What isn't left to taste is the label. Saying "this was generated" is a separate obligation from her consent — a signed release doesn't cover it, it's now the law in Europe and in California, and the companies whose rails we run on ask for it too — so plan on generated content being marked as generated wherever it's published. That's not a choice being weighed; it's a rule being followed. Nothing generates anything yet, so there's nothing labelled today either.
What's the watermark for, then?
It's a receipt, not a label. The plan is that every output carries a hidden marker tying it back to the model it came from, the request that produced it, and the account that asked — so if something surfaces where it shouldn't, we can say where it came from, and the signed consent record sitting next to it proves she agreed to it being made in the first place. Be clear about what that is and isn't: it's forensic, not preventive. It doesn't stop anyone saving a file, and a screenshot or a hard re-encode can destroy the mark. Finding a leak in the first place is a different job again — matching images out in the world, not reading anything hidden in them. None of this runs today: nothing generates anything yet, so nothing is watermarked yet.