AI short drama character consistency across shots

Example

ORIGINAL EDITORIAL EXAMPLE: a text-only walkthrough written by hand for this guide. Scene: At a rain-soaked archive loading bay, Lin Lan must open locker 17 while keeping a red evidence folder dry. The steel bench stands directly beside locker 17, so the three shots require no off-shot repositioning. Her jacket hood stays down, leaving the nape anchor visible.

Character — Lin Lan: V-shaped nape hairline; right shoulder slightly lower than left. Look — LOOK-LINLAN-RAIN: references Lin Lan; low-tied hair; dark-teal matte hooded work jacket; hood down; two rows of silver snaps on the left cuff. LOCK — LOCK-LINLAN-JACKET: “dark-teal matte hooded work jacket, hood down, two rows of silver snaps on the left cuff”; active in SHOT-01 through SHOT-03. Prop — brass locker key and red evidence folder remain separate assets. Throughout these three shots, the left arm pins the folder and the left hand supports its lower edge; only the right hand handles the key.

SHOT-01 — Start: folder tucked under left arm, key on the steel bench, right hand empty, gaze on the key. Action: Lin Lan picks up the key with her right hand. End: folder under left arm, key in right hand, gaze shifts to locker 17. SHOT-02 — Start: inherit the preceding end. Action: she inserts the key into locker 17. End: folder under left arm, key remains in the lock, right hand stays on it, gaze on the keyhole. SHOT-03 — Start: inherit the preceding end. Action: she turns the key once. End: folder still under left arm, key turned in the lock, right hand on the key, gaze moves to the door seam.

Reference-selection exercise: the four candidate pictures below exist only as descriptions written for this guide. A | Proposed daytime front portrait of Lin Lan: loose hair, an ivory shirt and bookshelves behind her. If the author confirms identity and an actual inspection supports it, it may provide visible facial features; identity rests on those two checks alone. The V-shaped nape hidden by her hair needs a separate picture that shows the nape. The shirt, loose hair, shelves and frontal camera belong to this picture only; the rain-night scene keeps its accepted Look and location. B | Proposed rainwear picture: hood up, with one black strap on the left cuff. That conflicts with the accepted hood-down jacket and two rows of silver snaps. To keep the design, choose a compatible picture. If the author changes the Look, update the accepted visual specification and affected LOCK/shots, and record it as a Look change. Resolve the conflict by swapping the picture or changing the spec; negative prompt wording cannot erase pixels already in a picture. C | Proposed empty loading-bay wide view: daylight, steel bench beside locker 17, fixed entrances clear. It controls spatial relations only: every shot keeps its planned framing, and the story stays on a rainy night. D | Proposed shot-three end frame: the key has turned, the right hand remains on it and the gaze has moved to the door seam. That is the accepted end state; shot three starts with the key inserted and not yet turned. Using it as a start frame jumps ahead to the result of the turn. A real tail frame comes from an actually generated video.

An optional directing choice for this pass: make SHOT-03 a medium view from the left rear, showing the nape, left cuff, folder under the left arm and right hand on the key in the lock. Bench and locker remain adjacent. This is a new camera proposal that needs author acceptance; the three shots' action order and accepted endpoint stay the same. That frame needs support for the visible rear anchors and the compatible Look; A's frontal face can stay off the list even though a character picture is available. Resolve B's costume conflict first. C may serve geography. The start frame comes from the SHOT-03 frozen start, so D stays off.

Do not hide decisions inside “use everything as reference” “Use B's person and C's location, then start every shot like D” imports conflicting clothing, a fixed wide composition and an early end state. A clearer request preserves the accepted identity/Look, lets the location reference control only bench/locker spatial relations, and makes each start frame obey its own shot's start. Scope declarations state creative intent; watch the generated shots to confirm the influences stayed separate.

How can a reference record avoid inventing files? 1. The author has supplied a picture, it is readable in the project and its content/purpose has been inspected: use REF with a project-relative file path. Use the picture as supplied; it needs no regeneration through this suite and no IMG entry. 2. The creator explicitly prepares and attaches pictures in their own canvas: use PLAN with an existing project IMG or SHOT entry as the locator; a future filename such as 待提供/some-picture.png cannot serve as one. PLAN is a settled attachment arrangement: what to attach and in what order. 3. The corresponding entry or necessary preparation choice is still absent: keep a stated gap. Switching to text-to-video takes an explicit creator choice, and an incomplete plan stays marked as a gap.

The next two slots illustrate syntax; a full production-input list follows the actual visible frame. They assume the author has accepted the proposed camera choice, and that its matching SHOT-03 frozen start and an IMG-LOOK-LINLAN-RAIN entry titled 林岚雨夜造型 already exist in the current project. The author must additionally have confirmed that this Look remains fixed for the entire episode and explicitly chosen external picture preparation. Treat them as a ready plan only when all of these premises hold: PLAN-START-03（顺序：1）· SHOT-03《17号柜转钥匙起点》（用途：起始帧；控制：已接受的左后侧中景、站位、钥匙已插入但尚未转动的起始状态；不得控制：已经转过的钥匙、柜门打开或新的终点） PLAN-LINLAN（顺序：2）· IMG-LOOK-LINLAN-RAIN《林岚雨夜造型》（用途：身份；控制：可见后颈发际、肩线及已确认本集固定造型；不得控制：手势、钥匙位置、视线、场景地理、镜头构图） Meaning: slot 1 uses the existing SHOT start for the accepted camera/placement and the inserted-but-unturned key. Slot 2 uses the existing Look prompt entry for visible identity anchors and the separately confirmed whole-episode Look; hands, key position, gaze, geography and composition stay outside its scope. The Chinese field names preserve the source's documented slot syntax.

Including fixed clothing in that identity slot depends on the explicit whole-episode premise above; a costume picture joins identity only under that premise. If clothing, wetness or damage changes within the episode, scope that state separately. A genuinely different purpose for the same image gets a separate slot. REF and PLAN can share one numbered sequence, with each input taking its own number. The syntax excerpt shows two slots only. Formal records add the geography and prop inputs the actual visible frame needs and keep real gaps visible. Readiness and production checks run in a real project with its entries and pictures in place.

If you later choose an end frame for shot three, it can only show the already-selected turned key, hand still on it and gaze at the seam. Opening the locker lies beyond this three-shot sequence's accepted endpoint. An end-frame slot also requires a corresponding start frame. With both planned boundaries written, the movement between them still comes from generation, and the actual video/tail frame a later actual-result continuation needs exists only after a real run.

Copyable text handoff request: In project /ABS/my-drama, EP001, preserve Lin Lan's accepted identity, Look, LOCK and three-shot action order. Read the current frozen frames and explain each frame's visible subjects and reference needs. Separate images actually supplied and inspected, explicit external preparation plans and remaining gaps. Mark any unread picture as unchecked. List costume or boundary-state conflicts and leave the accepted design as it is until the author decides. For external plans, PLAN must resolve to a current IMG or SHOT, with order, purpose, control and non-control scope. Report missing entries; do not substitute future filenames. Text only. Stay at the current stage unless video prompts are requested; do not generate images, video, voices or music, or start suite production.