Director's Guide · Version 1.0
HAL
A professional AI film production pipeline built for directors. Takes a screenplay and a director's creative vision and produces ready-to-generate Seedance 2.0 prompts — with every creative decision documented, every asset referenced, and every shot approved before a single credit is spent.
"Writing and planning is free. Generation costs money and time."
HAL is a linear pipeline. Each screen unlocks the next. The director is always the director — HAL never makes creative decisions without approval.
01Overview→Fill creative context first — style, references, vision, genre. Claude reads your script with this context already loaded.
02Story→Upload script. Claude produces the Story Intelligence Document and act breakdown. Director confirms structure.
03Cast→Review extracted characters. Upload reference images for each stage. Lock generation prompts.
04World→Review extracted locations, props, vehicles. Upload reference images. Lock generation prompts.
05Shots→Write and refine the shot list. This is where you spend most of your time. Approve shots when ready.
06Prompts→Review assembled Seedance prompts. Mark ready. Copy to Dreamina or send via API.
The core principle: Get the shot list right before you generate a single frame. Every generation costs credits and 5–8 minutes. Every iteration on the written shot list costs nothing.
Fill this first — before uploading your script. Claude reads your screenplay with all this context already loaded, producing cinematically informed analysis from line one rather than a generic breakdown.
Foundation
GenreOptional but useful. Helps Claude understand the world and inform the story breakdown.
Style PresetRequired. The visual DNA of the entire project. This single text block travels verbatim into every Seedance prompt without exception. Aspect ratio lives here too. Choose carefully — it defines how your film looks and feels.
Creative Team & References
CinematographerName the DP and specify which film and which dimension. "Deakins / Blade Runner 2049 for lens compression ONLY — not colour" is more useful than just "Deakins."
Production DesignerSame principle. Name + specific film + what dimension you're referencing from it.
Film ReferencesFor each reference specify: TITLE / TAKE (what to use from it) / AVOID (what not to copy). Precision prevents Claude from pulling the wrong dimension of a reference.
Visual References
MoodboardUpload concept art, location references, world-building material. These stay persistently visible as context throughout the session.
Director's Vision Notes
DirectionWrite freely. What is this film about? How should it feel? No structure required. This is the connective tissue HAL uses across Cast, World, Shots and Prompts. Brief it like you would brief your crew.
The story intelligence layer. Written once from your complete screenplay, confirmed by you, then locked as permanent context for everything that follows.
SID — Story Intelligence Document
The SID is the compressed dramatic map of your entire film. The header shows page count, estimated runtime, aspect ratio, genre, and period — all auto-calculated. Below it, WHAT IS THIS FILM is always visible: a 5-sentence synopsis anyone on your team can read in 60 seconds to understand the film.
Four collapsible sections sit below the synopsis — click to expand, click again to collapse, multiple can be open simultaneously:
Central ConflictThe engine of the film in 1-2 sentences.
Story StructureAct breakdown, page ranges, dramatic function of each act, key structural pivots.
Plot PointsThe structural hinges of the whole film. The moments where the story turns and cannot go back. Your red cards on the story wall.
Character ArcsWhere each principal starts emotionally, key pivot moments, where they end.
Acts
Click any act in the left column to load its detail. Each act shows: SYNOPSIS (what happens), KEY SCENES (2-3 most important scenes), DRAMATIC INTENT (what this act must achieve, with FOCUS POINTS and CONTEXT as inline expandable pills), and CHARACTER STATES (@reference pills for active characters, clickable to navigate to CAST).
Three tiers of characters. Principals get full detail. Supporting get simplified detail. Background gets one composite reference image that defines the visual vocabulary Seedance draws from.
Permanent Fields — same across all stages
PhysicalAge, height, weight, ethnicity, skin tone, hair, eyes, body type, distinguishing features. Everything Seedance needs to maintain visual consistency across generations.
Character TraitsPersonality, behaviour, drives, psychology. Who they are as a person.
Story ArcWhere they start emotionally, key pivot, where they end. Their journey through the film.
ConnectionsRelationships to other characters. Who they trust, fear, love, protect. Drives shot decisions in scenes they share.
VoiceHow they speak, tone, cadence, register. Plus audio file upload slot for voice reference. Goes into Seedance prompt as voice description line.
Stage-Specific Fields — update per stage
Stage ImageThe character's complete visual look for this stage: face and wardrobe together in one reference image. Click + to upload. Click uploaded image for full-screen view.
Changes from Previous StageWhat changed: hair, injuries, makeup, facial changes, wardrobe alterations. Empty for Stage 1.
WardrobeSpecific garments for this stage. Texture, cut, colour, accessories, signature details. Be verbatim — every detail matters for generation consistency.
Generation PromptThe locked verbatim character description for Seedance. Never paraphrase. Never abbreviate. This exact text travels unchanged into every prompt for this stage. Lock it when finalised.
The filmstrip: ALL button opens full-screen comparison of all stages. Numbered thumbnails switch between stages. + adds a new stage. Most characters have 3-5 stages maximum.
Locations, props, and vehicles. Same stage system as Cast — every asset can have multiple stages for different conditions across the film.
Location Fields
Header Tags[INT/EXT] [CITY, COUNTRY] [PERIOD/YEAR] — shown as pills below the location name.
DescriptionPhysical description: materials, textures, scale, what it looks and feels like.
Dramatic FunctionWhat this location does in the story. Why scenes happen here.
AtmosphereLighting conditions, weather, time of day, mood. Highest-leverage element for Seedance generation quality.
Prop & Vehicle Fields
DescriptionPhysical details, materials, condition, scale.
Story SignificanceWhat this object means in the narrative. Informs how it gets shot.
The primary working document. This is where you spend most of your time. The strip board organises your shot list by act → scene → shot, collapsible at every level.
Scene Strip (collapsed)
Scene NumberIndustry standard scene numbering. SC-014 = Scene 14.
Location + TimeINT/EXT, location name, time of day — always visible on the collapsed strip.
Dramatic IntentOne-line description of what this scene is about and what it achieves.
Density FlagKINETIC, RELATIONAL, or HYBRID. See Density Flags section below. Hover over any pill for explanation.
Shot Card Fields (Mode 1 — Planning)
Shot IDSC14-010, SC14-020 — increments of 10. Allows insertion (SC14-015) without renumbering approved shots.
DescriptionWhat happens: action, acting, emotional beat. Write for a director, not for a camera operator.
CameraInline camera description as part of the shot context.
Camera PromptStandalone clean camera instruction ready for Seedance. Same information as Camera but isolated — drops directly into the prompt without surrounding scene description.
Continuity InExact state at the START of this shot: character positions, prop states, emotional state, lighting, environment. Must match the previous shot's Continuity Out.
Continuity OutExact state at the END of this shot. This becomes the next shot's Continuity In. Never skip this field — it's how visual consistency flows through a sequence.
ApproveDirector's confirmation this shot is ready for prompt generation. Grey until clicked. Only approved shots appear in Prompts.
Mode 1 vs Mode 2: MODE 1 PLAN shows full shot cards for writing and editing. MODE 02 APPROVED shows compact rows for overview. Switch between them with the toggle in the top bar.
The generation queue. Same strip board structure as Shots. Only approved shots appear here. This is where you review the assembled Seedance prompts before generation.
Prompt Card Fields
Shot DescriptionRead-only summary of what happens in the shot.
CameraRead-only camera instruction.
References Required@reference pills showing exactly which assets to attach in Dreamina before generating. Click any pill to navigate to that asset in Cast or World.
Seedance PromptThe full assembled CUT-divided Seedance prompt. Read-only by default. Click the Edit icon to open an editable popup. Confirm and Save to write back.
CopyCopies the full Seedance prompt to clipboard. Phase 1 workflow: copy → paste into Dreamina → generate.
Mark ReadyDirector's final sign-off. Grey until clicked. Only READY prompts are sent via the BytePlus API in Phase 2.
SID — Story Intelligence Document
The compressed dramatic map of your entire film. Written once from the full screenplay, confirmed by the director, locked. Never regenerated unless you explicitly trigger a full re-analysis. All work in HAL runs with SID always accessible.
@reference
An asset tag linking a named reference image to a Seedance prompt. @copeland_S1 points to Copeland's Stage 1 image. @warehouse_district points to that location's reference image. Update the image behind the tag at any time without touching any prompt.
Character Stage
One complete visual look for a character: face and wardrobe together in one reference image. A character can have multiple stages for different looks across the film — injuries, costume changes, different periods. Each stage has its own @reference tag.
Continuity Bridge
The Continuity Out field from one shot becomes the Continuity In for the next. This is how visual consistency flows through a sequence — character positions, prop states, emotional states, lighting — without losing track between generations.
CUT-Divided Format
The Seedance prompt structure used in Prompts. Each shot within a generation is separated by a CUT — marker. The model reads the full sequence, understands the dramatic arc, and distributes it across 15 seconds according to its own timing sense.
Generation Prompt
The locked verbatim character or location description that travels unchanged into every Seedance prompt. Never paraphrase. Never abbreviate. Copy verbatim every time. Lock it when finalised to protect from accidental edits.
Approve
Director's confirmation that a shot is ready for prompt generation. Only approved shots appear in Prompts. Grey until clicked, then white.
Mark Ready
Director's final sign-off on a Seedance prompt. Only Ready prompts are sent to Dreamina or the BytePlus API. The last gate before generation.