AI influencer
One face that survives a hundred posts. Generate the canonical still once, build a board of many angles from it, and attach that board to everything afterwards. The board is the character — the posts are disposable.
The rule above all. ONE HOP FROM THE CANONICAL STILL. Attach the canonical image, not a look you generated earlier. A hop from a hop drifts, and six looks later it is a different person with the same name.
0 of 4 steps ready
01. Generate the canonical still — once
The whole persona is downstream of this one image. It is generated ONCE and never regenerated: a second canonical is a second person. Everything after this attaches it.
The prompt
Still needs face, build, wardrobe, room
Hyperrealistic portrait photograph. She wears Photographed in Natural realistic skin with visible texture, soft directional light, hyperreal detail, shot wide with room to crop, vertical 9:16. No text. — Never describe hair as identity. A haircut description rebuilt one face into a different man. Identity is bone structure, asymmetry and skin — hair is wardrobe. — Never ask for 'older' or 'younger'. The model slims the face, and on a full face slimming reads as ageing. If a face came back wrong, ask to RESTORE VOLUME. — Generate wide and crop. It defaults wide, and cropping only ever goes tighter — you cannot recover frame you did not generate. — The profile photo IS the canonical still, unaltered, same wardrobe, same room. If the profile picture is not visibly the same person as the videos, the whole account falls over.
What this guide will not do
These come from the same sources as the method above, and we decided against them. They are written here rather than in a file somewhere, because a refusal nobody can see gets adopted again by the next person who watches the same video.
Ask ATLAS
It already knows these rules — the character lock, the room lock, one unbroken take, vertical 9:16. Ask it to write a field and it writes it ready to paste, and tells you what is missing first.