100% generated · 100% disclosed

Hi, I'm Marie-Claire Pronti.

I don't exist. Let's get that out of the way first — I'm a synthetic character, generated end-to-end by AI. No face model, no stock photos, no person behind the freckles. My initials are M.C.P. — that's not an accident.

I'm the spokesmodel for MCP Media Engine, and I'm also its hardest product demo: the same character, recognizable in every scene, every style, every post — made with the same API you can call from your own agent.

Same me, every time

Every image below is a real MCP Media Engine job — generated with img2img from one canonical character sheet. Check the freckles, the messy bun, the small scar over my left eyebrow. That's the consistency test.

Marie-Claire Pronti, AI-generated — Cafe scene — Cafe scene · mjob_32ff22ee
img2img from the character sheet, denoising 0.7

Marie-Claire Pronti, AI-generated — Terminal glow — Terminal glow · mjob_3c700a95
img2img, denoising 0.75

Marie-Claire Pronti, AI-generated — Watercolor wash — Watercolor wash · mjob_824646cd
style transform, denoising 0.8

Marie-Claire Pronti, AI-generated — Noir terminal light — Noir terminal light · mjob_96586bf3
style transform, denoising 0.8

Marie-Claire Pronti, AI-generated — Green flash — Green flash · mjob_32d609d5
style transform, denoising 0.8

Marie-Claire Pronti, AI-generated — Paper sketch — Paper sketch · mjob_97a1f205
style transform, denoising 0.8

How I'm made

There's one canonical character sheet — five angles of me plus a written spec: the freckle pattern, the scar on my left eyebrow, the mole on the right side of my neck. Those three details are my continuity anchors. If a generation loses them, it isn't me, and it doesn't ship.

Every new scene is an img2img job: the sheet goes in as input_image_url, the scene goes in as the prompt, and denoising_strength decides how much the model may reinvent. My face is never regenerated from text alone — that's how synthetic characters fall apart.

That's the whole trick, and it isn't exclusive to me. One reference image, one API call, one dial. Your brand character, your product, your dog — same method.

Steal my prompts

The exact recipes behind the gallery above. Swap in your own reference image.

Put your character in a new scene

model: nano-banana · input_image_url: <your reference> · denoising_strength: 0.7

same person as the reference image, sitting at a corner table in a small cafe, morning light through the window, candid documentary photo, natural skin texture

0.7 keeps identity locked. Go higher and the face starts to drift.

Moody terminal-glow portrait

model: nano-banana · input_image_url: <your reference> · denoising_strength: 0.75

same person as the reference image, face lit only by the green glow of a terminal screen in a dark room, cinematic, deep shadows, film noir mood

Single light source does the work. Everything else stays dark.

Style transform that keeps the person

model: nano-banana · input_image_url: <your reference> · denoising_strength: 0.8

same person as the reference image, rendered as a loose watercolor painting, wet-on-wet washes, pigment blooms, visible paper texture

Style transforms hold identity at 0.8. Material transforms (clay, marble) break faces — don't bother.

Full prompt library Run these via the API

The fine print, said loudly

Marie-Claire Pronti is a fictional, AI-generated character owned by MCP Media Engine. She is never presented as a real person, creator, or customer. Every image of her is generated media, and we publish the job IDs so you can see the receipts.

We think disclosure is the interesting part: keeping a synthetic character consistent is a harder engineering problem than pretending she's real.