cinemasetfreePortraits and the reference chain
Describe a character, generate portraits, pick the one that's right. That chosen portrait is then used as the visual reference when generating their wardrobe and the scenes they're in.
Describe, then generate
Images come from text. A character with no description generates something generic. The description is the work; the picture is the easy part.
Describe
Describe asks the text model to write a visual description from what the project already knows: the script, the scene headings, the character’s own fields. Edit what comes back; it is a draft, not an answer.
Generate
Generate renders a portrait, combining that description with the active production design’s style prompt so everything in the project looks like it belongs to one film.
Iterate
Every generation is kept, not overwritten. Open the portrait modal to page through versions, rate them 1–10, pick the one that’s right, or upload your own image instead.
Why the order matters
Portraits flow downhill. A character’s chosen portrait is passed as a reference when generating that character’s wardrobe images, and a location’s establishing shot is passed as a reference for the scenes set there. Generate the cast first and everything downstream inherits their faces.
Do it out of order and nothing breaks. You get less consistency, and you will want to regenerate later.
Bulk describe and bulk generate run the whole selection. Bulk generate spends credits per image, so the estimate shown before you confirm is worth reading.