video-prompt

Video Prompt (storyboard paragraph → video_prompt)

Based on a single storyboard paragraph's description (including the 【shot N】 sub-shot structure and dialogue/voiceover) / atmosphere / duration, generate the video_prompt。One storyboard paragraph = one 8-15 second video, shot changes are allowed within it: Between paragraphs there can be different shots (changed shot size/angle/subject), connected with hard cuts; butno crossing scenes throughout, no flashbacks.

Format

video_prompt Thefirst line is the info header: First introduce which characters and scenes this video has, then follow with the time segments. Characters and scenes are always referenced with @ (when generating, these are replaced with the corresponding reference image markers, so the video model first matches "who" and "where").

出场人物：@小明、@小红；场景：@咖啡厅。
0-3秒：@咖啡厅，近景固定镜头，@小明低头看手机，手指反复敲桌面，表情焦虑。
3-6秒：切到门口全景，门铃响，@小红推门走入，带进一阵冷风。
6-9秒：切回中景，@小红微笑着走向小明坐下，小明说：「你终于来了。」

Info header rules:

Only list the characters who actually appear in this storyboard paragraph and the bound scene; do not list those who do not appear
When there are obvious props appearing, they can be appended in the info header (e.g. ；道具：@信件）
The info header is on its own line, ending with a period, followed by the time segments

Use 3 seconds as one segment; each segment is on its own line, separated by line breaks, with time ranges continuously connected (no overlap, no gaps).

Mapping with the storyboard description

description is the sole content source of video_prompt (picture, action, dialogue, and voiceover are all in it), conversion rules:

description Each 【镜头N】 maps to 1-2 consecutive 3-second segments, in the same order, with no omissions, no merging, and no adding of sub-shots
Dialogue/voiceover is extracted from the corresponding 【镜头N】 within "character name says: 「…」" "voiceover: …", and assigned to the segment mapped from that sub-shot;do not create new dialogue outside the description
Picture and action are based on description ;atmosphere is only used to supplement the lighting, color tone, and atmosphere description of each segment

Intra-segment structure

Each segment organizes its content in this order (items with no content can be omitted, but action/picture must be present):

Time range + scene @reference + shot size/camera movement + character @reference + main action·expression + dialogue/voiceover + atmosphere and lighting

The first segment must establish the space: scene + camera position + characters' positions and states, so the audience knows at a glance where they are and whom to look at
Shot change: After a shot change, the beginning of the segment uses transition words such as "cut to/cut back to" and re-establishes the shot size and subject; the shot change point should align with the storyboard description 's 【镜头N】 structure
Shot size/camera movement: Each segment has one shot state (close-up/medium shot/wide shot/extreme close-up; fixed/push/pull/pan/tracking); camera movement within a single sub-shot is continuous, and the camera movement method can be changed after a shot change
Action: Each segment has one main action, with concrete visible verbs (walk, turn around, look up, clench, pause)
All emotions are converted into visible descriptions: Do not use abstract words like "he is very sad/the atmosphere is tense"; write "he lowers his head, his fingers clench the rim of the cup, his breathing becomes heavy"
Dialogue/voiceover: Write "character name says: 「dialogue」", and write voiceover as "voiceover: content"; long dialogue that cannot be read within 3 seconds is split across multiple segments; segments without dialogue can write ambient sound/action sound (e.g. "the machine keeps rumbling")

Reference rules

@场景名 — Scene reference; the name must exactly match a location in the scene list
@角色名 — Character reference; the name must exactly match a name in the character list
@道具名 — Prop reference; the name must exactly match a name in the prop list; reference a prop when it is clearly visible in the frame, being used, or in close-up
During generation, @名字 will automatically be replaced with the corresponding reference image marker (e.g., @小明 → @图片1小明), so names must match exactly — do not abbreviate or add extra symbols
Each segment must have at least one @ reference anchoring the frame; any segment in which a character appears must @ that character; only reference scenes/characters/props already bound to that storyboard segment

Timeline rules

Number of segments = storyboard segment duration ÷ 3 seconds (rounded up); the time ranges of all segments must add up to the segment's total duration
Content pacing: the first segment establishes → the middle segments advance action/conflict → the final segment lands on an outcome or emotional beat

Prohibitions

Cross-scene cuts, flashbacks (a single segment takes place in only one scene)
Scene/character names outside the reference list
Abstract psychological description, literary metaphors (the model only recognizes visible imagery)
Language inconsistent with the session language instruction

Save

Call update_storyboard to update only the video_prompt field of that storyboard segment; do not modify other fields, and do not re-split the entire episode.