# Drama production: the complete guide

> Updated: 2026-10-30 · Applies to KiiKii
> HTML version: https://kiikii.cc/en/doc/drama

## Quick answer

- Five stages: Script → Content Review → Storyboard → Shot Generation → Merge Video.
- Project assets manage characters, scenes, props, and clues, plus each character's base image and voice.
- Storyboard editing has four tabs (Content, Frame, Reference, Audio); a single shot can be handed to the Project Agent, and the final cut can use Quick Merge or Full Compose.

## 1. What the Drama feature does
Drama organizes content by episode: a script is split into scenes and shots, confirmed and generated shot by shot, then composed on a timeline into a complete episode. Every step has its own state, so you can stop at any stage and continue later.

## 2. New project and visual style
Create a project with a name, story synopsis, visual style, and output size. The style is one of 10 built-in presets (Photorealistic, Japanese Anime, Guofeng Xianxia, Chibi 3D, Cel-shaded 3D, Nostalgic Memory, Horror Suspense, Cyberpunk, Clay Toy, 1980s) or a custom description; the two are mutually exclusive and one is required. Size is 9:16 for short-video platforms or 16:9 for promo films, and the style feeds into every shot's prompt.

## 3. Five stages and episode management
The progress bar shows each stage's status and lets you switch between them: Script (structured), Content Review (confirmed), Storyboard (configured), Shot Generation (completed), and Merge Video (rendered). A project can hold multiple episodes with independent state; the sidebar shows character and scene counts, version history allows rollback, and episode settings apply the same aspect ratio, resolution, model, and narration voice.

## 4. Script and AI structuring
Import a document or paste a novel or outline, run one-click formatting to normalize paragraphs and dialogue, then let AI structuring extract scene number, location, one-line summary, character count, and scene count. Edits save automatically and AI does not overwrite manual changes; write one action per sentence, give dialogue its own paragraph, and keep scenes concentrated so shots can reuse reference images.

## 5. Review: splitting and merging shots
Content Review does not rewrite content; it splits the script into shots and verifies each shot's source text, shot facts, boundaries, dialogue and narration, and estimated duration. Split Shot turns one shot into two when a line contains two actions, the shot size needs to change, or the visuals differ greatly; Merge Next Shot combines adjacent shots that share a scene and action line and are each too short.

## 6. Project assets: characters / scenes / props / clues
Project assets are the film's asset library: characters (profile, appearance, base image, reference candidates, voice), scenes (location, time, environment, base image), props (items held or shown, including packaging), and clues (plot-critical objects, promises, symbols). The panel supports status filters, sorting, one-click Fill Missing Images for characters, scenes, and props, and exporting all base images as an archive.

## 7. Character profiles, base images, and voice
A character profile has four parts: appearance settings, base image (used by default when generating shots), reference candidates (alternative images promoted to primary), and voice settings (a cloned or system voice, or the narration voice). Profile status shows confirmed or pending, and pending characters are flagged before generation. As long as a character is checked in a shot, Reference Image mode uses their base image; fill in characters first, then scenes and props.

## 8. Shots · Content: speakers and narration
The Content tab defines what a shot says and who says it. Every line binds to exactly one speaker, and voiceover and lip sync follow that speaker; off-screen narration uses the narration voice and does not take up a character. You can correct the speaker, add, edit, or delete lines, and check View Source Text at any time.

## 9. Shots · Frame: prompts and three generation modes
The Frame tab holds the visual prompt (subject, scene, shot size, composition, lighting), the motion prompt (action process, pacing, start and end states), camera movement, and duration. **Reference Image** locks appearance and space with asset images for the best cross-shot consistency; **Storyboard-driven** generates a first frame (optionally a last) then drives the video for the most control; **Direct Generation** works from the text prompt alone and is fastest and cheapest for empty and atmosphere shots.

## 10. AI revision and continuity control
Click AI Revision and describe the change in natural language; visuals, camera movement, and continuity are adjusted together, with a before/after diff shown before applying. Leaving dialogue unchecked keeps voiceover untouched, while checking it clears that shot's voiceover so it must be regenerated. Continuity controls cover shot size, camera position and angle, composition, blocking, eyeline and screen direction, the axis rule, action start and end states, and notes on adjacent shots.

## 11. Shots · Reference: people, scenes, and structure
The Reference tab decides which assets a shot uses: characters present, scene, props shown, and story clues, plus optional structure control when you need to reuse a specific image's composition. Generate Reference Image composites the checked items; if selections change a lot, References Need Update appears, so regenerate the reference before generating video.

## 12. Shots · Audio: native audio, voiceover, and mute
Each shot independently chooses its audio: Native video audio keeps the dialogue, ambience, and music generated by the video model; AI Voiceover mutes the source and uses the character's bound voice; Mute produces a silent shot for unified post-production music. Shots with native audio do not use the voiceover track, and only shots whose voiceover was replaced in post appear as voiceover segments during merging.

## 13. Shot generation and handing off to the Agent
Each shot is submitted, billed, and retried independently. Run Pre-generation Check to verify characters, reference images, scenes, props, and voices; set a unified image and video model for the episode with per-shot overrides. The task map shows completed, in-progress, and failed counts, and batch voiceover handles several shots at once. Use Hand Off to Creation Agent for multi-asset compositing or repeated conversational trial and error; routine shots are faster generated from the shot card.

## 14. Merging video and the timeline
Once every shot is complete, use Re-run Quick Merge to concatenate shots in order for an internal pacing check, or Re-run Full Compose to apply timeline edits, transitions, and subtitle styles for delivery. The three-track timeline covers video (start, keep-in, and keep-out points), voiceover (only segments replaced in post), and subtitles (text and timing); each segment can carry an intro transition except the first shot.

## 15. Project Agent assistant
Open Project Agent inside the project to chat with an assistant that reads this episode's script, characters, scenes, props, clues, and storyboards, references assets with @character / @scene / @audio, and can directly revise the script or storyboard and regenerate a shot. For example, ask it to revise shot 2 so a character has his back to the camera and then speaks; bulk generation still happens on the Shot Generation page, and changed dialogue needs a new voiceover.

## 16. Case study: Ahe and Lu Xing
The campus drama "The Best Seat" starts from a 169-character, 4-scene script and composes into a 32-second final cut: four shots (wide establishing, two-person follow, static close-up, closing empty shot), 2 characters, 1 scene, 2 props, and 2 clues, with 2/2 voiceover segments and about 71.7 credits used. Shot boundaries carry the film: shot 2 keeps "catching up" and "the invitation" in one follow shot, shot 3 uses a close-up on "the best seat", and shot 4 holds a 12-second empty shot to let the emotion land.

## FAQ
### What is the difference between Drama, Digital Human, and the Creation Agent?
Digital Human is a single-take talking-head tool, and the Creation Agent focuses on conversational asset production. Drama is a project-based series pipeline: script first, then shot-by-shot review, storyboard splitting, per-shot generation, and a timeline merge into a complete episode.
### How do I choose between built-in styles and a custom style?
Built-in styles are 10 presets that keep the whole film consistent. Choose a custom style when you have a clear art direction and want to write the prompt yourself. The two are mutually exclusive and one is required.
### What is the difference between base images and reference candidates in project assets?
A base image is the primary reference for that character, scene, or prop and is used by default when generating shots. Reference candidates are alternative images you can compare and switch into the primary slot.
### When should I use Split Shot versus Merge Next Shot?
Split when a single shot contains two distinct actions, or two lines that need different shot sizes. Merge when two adjacent shots share a scene and action line and are both too short, so the final cut does not fragment.
### How do I choose among the three video generation modes?
For consistent characters and scenes across shots, choose Reference Image. When composition must be exact and you need to lock the first or last frame, choose Storyboard-driven. For empty, atmosphere, and single-shot content, choose Direct Generation.
### Does AI revision change the dialogue?
By default it only changes visuals, camera movement, and continuity. Dialogue changes only when Allow dialogue/narration changes is checked, which clears that shot's voiceover so it must be regenerated. A before/after diff is shown first.
### When should I hand a shot off to the Creation Agent?
Use Hand Off to Creation Agent when you need multi-asset compositing, repeated conversational trial and error, or special creative treatment. Polish it there and return to Drama once satisfied.
### What if shot generation fails?
Each shot is independent, so a failure only affects that shot and can be retried on its own; completed shots are not charged again. Run Pre-generation Check first to verify assets and voices.
### How do I choose between Quick Merge and Full Compose?
Quick Merge concatenates shots in order: fast, with no editing package or subtitle styling. Full Compose applies timeline edits, transitions, and subtitle styles. Use Quick Merge for pacing checks and Full Compose for delivery.
### Is the Project Agent the same as the Creation Agent?
They are different entry points. The Creation Agent is a standalone studio, while the Project Agent lives inside a Drama project and can read that project's assets and current storyboard, so revisions stay in the project.