Consistent AI characters.
Delivered as a system.
Photorealistic characters that stay the same face, body and presence across hundreds of images. You get the bible, the reference views, the dataset and the content bank, and your team can keep producing without me.
Cloud pipeline · no GPU · first images in 48 h · FR EN ES
Olivia Rossinny. The scene changes. She doesn't.
Twelve separate productions: day and night, studio and street, single and two-character scenes, a product shoot, and two visuals from the same evening in the same dress. Hair, freckles, body shape and gaze do not move.
Also in continuous production: Cataleya Ozuni, second fictional character, six scenes shown.
Nine expressions. Same bones.
Shot on a neutral background in monochrome so the structure can be judged without colour: brow line, freckle map, lip shape, hairline and jaw stay put from a wink to a neutral stare. This is the layer most character work skips, and the one that breaks first.
Scene first. Identity second.
The character never depends on the scene, and the scene never depends on the character. That separation is what keeps the identity stable at volume.
Lock the bible
Face, body, skin, hair and distinctive traits written down once, versioned, never overwritten.
Generate the 19 views
Neutral background, same light, same technical outfit. Then a decontaminated version of each, engine style removed.
Build the scene
A generic base validates composition, light and outfit with a temporary person. No identity yet.
Apply and iterate
The identity is applied from the reference views and checked against the bible until approved. Every step is archived with its prompt.
What you receive.
Everything is yours: files, prompts, parameters, and the workspace. Readable without any tool, reusable by any team or training partner.
- Character bibleLocked DNA sheet, 9 fields, dated versions.
- 19 reference viewsFace, head, expression sheet, hands, feet, torso, full body from 8 angles. Original and decontaminated.
- LoRA-ready datasetStable naming, per-image captions, trigger word, reproducible prompts. Train in-house or through my partner.
- Consistency sheetExact parameters, pose and light recipes, known drift cases and their fixes.
- Avatar Anchor Deck, configuredSingle-file HTML workspace: DNA versions, views, outfits, every visual with its renders and prompts. FR, EN, ES.
- AI disclosure kitWording and placement for EU AI Act art. 50 and the main platforms' policies.
Packages.
Fixed prices, fixed delivery windows. Every package starts from the same identity core; the difference is how many finished images and how much of the production system you take with you.
Identity Core
- Character bible
- 19 reference views, original + decontaminated
- LoRA-ready dataset with captions
- Consistency sheet + AI disclosure kit
- Anchor Deck configured
Identity + 25
- Everything in Identity Core
- 25 finished images
- Platform formats: 4:5, 9:16, 1:1
Identity + 50
- Everything in Identity Core
- 50 finished images
- Platform formats: 4:5, 9:16, 1:1
Identity + 100 + System
- Everything in Identity Core
- 100 finished images
- Custom image production pipeline for the character
- Recorded handover session
48-hour test, 120 €. Four images of your character or brief. Deducted from any package ordered afterwards.
Book a testScope.
Clear limits make a clean project. These are set before we start, not negotiated during.
What I do
- Fictional adult characters, built trait by trait
- Master avatar packs (19 views) and reference datasets
- Lifestyle and editorial visuals, single or multi-character
- Production tooling and handover so your team continues without me
- Digital twins of real people, with a signed consent agreement
- Clients on adult (18+) platforms: identity, references and lifestyle assets
What I don't
- Explicit or NSFW images and videos
- In-house LoRA training or ComfyUI work (dataset delivered, training via partner)
- Likeness of a real person without their written consent
- Long-form video, talking heads or lip-sync (short motion clips only)
- Characters presented as minors, in any style
FAQ
The questions that come up in the first call.
Do you use LoRA or ComfyUI?
Not in-house. The pipeline is cloud and zero-shot, which removes the GPU and cuts delivery to days. If you want to own a LoRA file, I deliver the dataset and run the training through a partner, staying your single point of contact.
Who owns the files?
You do: bible, views, dataset, prompts, parameters, finished images, and the configured Anchor Deck. Full commercial licence, worldwide, no time limit, on full payment.
Does the character look like someone?
No. Each character is fictional and built trait by trait from a visual configurator, not from a person's photos. A digital twin of a real person is only produced with that person's signed consent.
How many images before it drifts?
Drift is handled at the source: locked bible, 19 views, decontaminated versions, and a check on every render against the bible. Known drift cases and their fixes are written into the consistency sheet you receive.
What about video?
Short image-to-video clips of 5 to 7 seconds are available as an option. No long-form video, talking heads or lip-sync for now.
Do you disclose that it's AI?
Every delivery includes a disclosure kit: wording, placement, EU AI Act article 50 requirements and the main platforms' policies. Publishing compliance stays your responsibility; the kit makes it a five-minute job.
Send a brief. Get four images in 48 hours.
Tell me whether the character exists or needs creating, where it will be used, the volume you need and the deadline. I reply within one business day.
Write to Alexis