01
Real traits, extracted and pinned
Face geometry, skin, hair, styling, lighting, and camera character all get read from the reference and locked to one concrete value each — no ranges, no "natural-looking" left for the model to interpret.
Upload a photo, get back one JSON prompt with every detail pinned — so the same character stays consistent across every regeneration, without the doll-faced AI look.
Click, drag and drop, or paste a reference photo here
PNG, JPG, or WEBP up to 4MB
Click, drag and drop, or paste up to 5 photos here
Same creator/style, PNG/JPG/WEBP, up to 4MB each
The first photo you add is used as the anchor for pose and lighting. The rest help lock the recurring style across all of them.
Your locked character prompt will appear here.
Image generation — coming soon. For now, paste the JSON into Flux, Nano Banana, GPT Image, Seedream, or your preferred structured-prompt-friendly AI image tool.
Loose prompts leave room for the model's own defaults. This tool removes that room entirely.
Face geometry, skin, hair, styling, lighting, and camera character all get read from the reference and locked to one concrete value each — no ranges, no "natural-looking" left for the model to interpret.
The result deliberately isn't an identity clone — it keeps the reference's energy and vibe while changing enough specifics that it reads as a different, original person.
No "4K, ultra-detailed" quality boosters, no HDR over-sharpening. Instead: a real capture pipeline, located imperfections, flat grading, and a fixed seed — so it reads like an unretouched photo, not an AI render.
Straight answers about how this tool works.
No — on purpose. It locks an original character that shares the reference's vibe, styling, and energy, but is deliberately a different, non-identifiable person. It's built for consistent original characters, not for cloning a real identity.
Every image model has default biases — bigger eyes, poreless skin, an over-sharpened HDR look. A loose text prompt leaves room for the model to fall back on those defaults. Locking every attribute to one concrete value in JSON removes that room, so regenerations stay consistent.
Any model that accepts a detailed structured prompt — Flux, Nano Banana, GPT Image, Seedream, and SDXL/Leonardo-style tools all respond well to it. Some platforms want it as plain text rather than a JSON block; you can paste it either way.
PNG, JPG, or WEBP files up to 4MB — click to browse, drag and drop, paste from your clipboard, or paste an image URL.
Use it to steer specifics the reference alone can't tell you — a different age, build, wardrobe, or setting, while keeping the same overall vibe.
Single Photo locks a character from one image. Multiple Photos (2-5 images of the same creator or style) also locks the recurring presentation pattern across all of them — the pose they default to, how they typically frame a shot, their consistent energy and styling — not just their physical look from one snapshot.
Yes — uploads with NSFW or otherwise disallowed content aren't supported, and photos with a real, identifiable face may occasionally be declined by the underlying AI for privacy reasons.