Keeping One Face Across a Whole Leonardo Phoenix Series
Character consistency is the difference between a series and a pile of similar pictures. Phoenix gives you the levers.
A character who changes nose every frame reads as a glitch, not a style. Leonardo's Phoenix model handles consistency better if you feed it the same reference image and a locked seed range instead of describing the face from scratch each time. The model remembers more when you stop re-describing.
My setup: generate one hero portrait I like, lock it as the character reference for every scene, and vary only the action and lighting in the prompt. The reference does the heavy lifting. When I described the face fresh each time, I got three different cousins. When I referenced, I got one person in different rooms.
The Phoenix-specific move is editing rather than generating from zero. Take the hero, send it into the image-to-image editor with a new pose prompt, and keep the character weight high. Each frame inherits the face. Re-rolling from text resets too much.
Consistency is a discipline, not a setting. Reference image, locked inputs, edit-instead-of-regenerate. Do those three and a series finally looks like a series instead of a coincidence.