Drama production: the complete guide

Episodes and versions

• Episode sidebar: shows each episode's character count, scene count, and overall status, for example “169 characters · 4 scenes · episode complete”.
• New episode (新建集数): continue writing the next episode, reusing the project's characters and visual style.
• Version history (版本记录): save a version before major script or storyboard changes so you can compare and roll back.
• Episode settings (本集设置): apply the same aspect ratio, resolution, model, and narration voice.

What AI structuring does

It cuts continuous prose into shootable scenes and extracts the location, time, and a one-line summary for each, which becomes the basis for splitting shots later. You can still edit scene titles and structure by hand afterward; AI does not overwrite your edits.

• One action per sentence: avoid packing three events into one sentence so shots are easier to split.
• Give dialogue its own paragraph and use quotation marks to separate speakers, so voiceover and subtitles land more accurately.
• Keep scenes concentrated: shots in the same scene can reuse reference images, which lowers cost.

• Source text (原文依据): which line of the script this shot corresponds to; confirm nothing was missed or duplicated.
• Shot facts (镜头事实): characters, location, time, and key actions; mistakes here send the visuals off course.
• Shot boundaries (镜头边界): where the shot starts and where it ends.
• Dialogue and narration: dialogue must match the source text, while narration adds information the visuals cannot convey.
• Estimated duration (预估时长): estimated from dialogue length, so the whole episode's duration is roughly known here.

Split Shot (拆分镜头)

Splits one shot into two, each with its own scene, dialogue, and visual prompt. Use it when one line contains two actions, when the shot size needs to change, or when the visuals before and after differ greatly. Shorter shots are also easier to generate successfully.

Merge Next Shot (合并下一镜)

Combines the current shot with the next one, keeping the continuous action and dialogue. Use it when both shots sit in the same scene and the same action line and each is too short; merging keeps the pacing smoother and saves one generation.

A few actions on the panel

• Status filter (状态筛选): all statuses / used in this episode / missing reference image / incomplete profile / referenced / not referenced.
• Sorting (排序): default order, incomplete profile first, most referenced, by name.
• Fill Missing Images (一键补图): generates candidate images one by one for characters, scenes, and props that lack a base image; clues are excluded.
• Export project assets (项目资产导出): downloads the full set of base images as an archive for filing or handover.

As long as a character is checked in a shot, Reference Image mode uses their base image as the character reference, so it is the same face across shots. When AI Voiceover (AI 配音) is selected on the audio tab, lines are generated in that character's bound voice.

• Shot facts: what happens in this shot; the shared basis for both visuals and dialogue.
• Speaker (说话人): every line must be bound to a character, and voiceover and lip sync follow the speaker.
• Correcting the speaker (纠正说话人): fix it here when the script does not make clear who is talking; after changing dialogue, the voiceover must be regenerated.
• Off-screen voice (画外音·旁白): content that must be heard but not seen; it uses the narration voice and does not take up a character.
• View Source Text (查看原文依据): check against the original script line at any time so edits do not drift from the source.

• Visual prompt (画面提示词): subject, scene, shot size, composition, and lighting; what this frame looks like.
• Motion prompt (动态提示词): the action process, pacing, and start and end states; how the shot moves.
• Camera movement (镜头运动): camera position movement and speed, such as push in, pull out, pan, truck, and follow.
• Duration (时长): set by dialogue length and pacing; too short truncates the line.

• One pass: visuals, camera movement, and continuity are adjusted together, with no need to edit three fields separately.
• Example wording: “Change it to a rainy night, push in slowly, and make the shopkeeper more impatient.”
• Visuals only: with dialogue left unchecked, voiceover and subtitles stay untouched, saving one voiceover run.
• Changing dialogue: checking it clears this shot's existing voiceover, and the voiceover must be regenerated.
• Preview before applying: changes are written only after you confirm the diff, which avoids breaking the shot.

Continuity answers “does this shot line up with the previous one”: which side a character stands on, which way they look, what action they start from, and what action they end on. With a single shot you can leave it blank; for multi-shot storytelling, fill in at least blocking, eyeline, and the axis.

• Characters present (出场角色): check only the people who actually appear; extra checks pull in unrelated characters.
• Scene (所在场景): determines space and lighting; when the scene changes, change the reference with it.
• Props shown (出现道具): items that are held or displayed; checking them keeps packaging and details stable.
• Story clues (剧情线索): promises, keepsakes, symbols, and other narrative information that must echo across shots.
• Generate reference image (生成参考图): composites the checked items into a reference image and flags when References Need Update.
• Structure control (结构控制): enable it when you need to reuse a specific image's composition or shot structure.

• Pre-generation check (生成前检查): verifies that characters and reference images, scenes and props, and voices are ready, so you do not discover missing assets halfway through.
• Unified model (统一模型): image and video models are set for the whole episode, and individual shots can override them.
• Task map (任务图): shows completed, in-progress, and failed counts.
• Per-shot actions (单镜操作): generate video, regenerate, generate voiceover, switch to AI voiceover.
• Batch voiceover (批量配音): generate voiceover for several shots at once, then review them together.
• Visual Review / Preview and Merge (视觉复盘 / 预览与合并): compare the whole episode side by side and move into merging.

When to use “hand off to the Agent”

• The shot needs multi-asset compositing, for example two specific reference images plus a particular scene.
• It needs repeated conversational trial and error that the per-shot parameters on the Drama page cannot express.
• You have a special creative idea or complex action and want a few versions in the Agent first, then bring one back.

Clicking it opens the Creation Agent with this shot's context; when you are satisfied, return to Drama and continue production. Routine shots do not need this detour; generating straight from the shot card is faster.

Two merge options

• Re-run Quick Merge (重新快速合并): concatenates shots in order with no editing package or subtitle styling, which suits internal pacing checks.
• Re-run Full Compose (重新完整合成): applies timeline edits, transitions, and subtitle styles to produce a deliverable final cut.

Timeline details

• Video track (画面轨): select a clip to adjust its start time, keep-in point, and keep-out point, trimming excess head and tail.
• Intro transition (入场转场): each segment can have a transition type (such as pixel dissolve or push in) and duration; the first shot has no transition.
• Voiceover track (配音轨): shows only voiceover segments replaced in post; they can be muted or locked.
• Subtitle track (字幕轨): edit subtitle text and timing so subtitles do not cover faces.
• Stats (统计): 4/4 final cut, 2/2 voiceover, 3 subtitles, 32 seconds total; confirm at a glance that everything is in place.

How it differs from ordinary chat

• Reads project context (读取项目上下文): knows this episode's script, characters, scenes, props, clues, and existing storyboards.
• @ references assets (@ 引用资产): for example “@Ahe says to @Lu Xing, using @Ahe-Chases-Lu-Xing Invitation · Ahe's voice”.
• Produces storyboards directly (直接产出分镜): regenerates storyboard images or talking-head clips from your revision and writes them back into the project.
• Collapsible (可收起): stays out of the way and can be reopened any time to keep chatting.

How to use it

Example prompt: “In shot 2, Ahe needs to have his back to the camera, then speak to Lu Xing. Please revise the prompt and regenerate the storyboard.” The Agent returns a new storyboard and explains what it changed; keep adding requests until you are satisfied, then return to the Storyboard or Shot Generation stage.

The key to this film is shot boundaries: shot 2 keeps “catching up” and “the invitation” in the same follow shot, avoiding a cut into two fragments; shot 3 uses a static close-up to emphasize the line “the best seat”; shot 4 holds a 12-second empty shot to close, letting the emotion land.