flatrender/docs/aep-template-convention.md

FlatRender — AEP Template Convention & Rule Engine

Single source of truth for how a FlatRender After Effects template is authored, scanned, and rendered (bound). Both the scanner (scan.jsx + Go quick-scan) and the render binder (the JSX generator) must obey this document. Conventions are versioned (convention_version per project) so the rules can evolve without breaking the existing library.

---

0. Project ↔ AEP relationship

• One project = one .aep file (in V2 stored at templates/{project_id}/template.aep or bundle.zip). A real project bundle is usually a footage folder + the .aep, zipped, and extracted on the render node so relative footage resolves.
• The project's human name never appears in AE. Comp/layer names are universal conventions, not per-project. The project is identified purely by which file it owns.
• AE version: the farm always runs the latest After Effects (a global setting, not per-project) → Master Properties + Media Replacement are always available.

---

1. Naming conventions

Composition names (universal, identical in every project)

Comp	Role
frfinal	Final render comp for Fix and MusicVisualizer (pre-built).
flatrender	Final render comp for Flexible and Mockup — assembled at render time by the binder.
frshare	Holds the shared colours (one text layer per colour; layer name = element key, sourceText = colour value).
all	Holds the shared-layer definitions.

Layer-name prefixes

Prefix	Meaning
frl_<key>	Editable visible layer (text / media / audio) — what the user fills in.
frd_<key>	Data / direction layer — hidden values, checkbox/dropdown/fill, colour data, RTL companion.
frc_<key>	Container comp used for duplication (scenes, repeat boxes).
frs_<key>	Shared-ref layer inside an frc_ container.
frd_<key>d	Direction companion of an frl_ layer (0=LTR, 1=RTL), produced by FRDMaker(key)+"d".

Key normalization (legacy Helper)

• FRLMaker(key) → if it already contains frl/frd, keep; else frl_<key>.
• FRDMaker(key) → strip frl_, ensure frd → frd_<key>.

🔑 The uniqueness invariant

No two layers ever share a name. Every editable field maps to exactly one independent value.

• This is why duplication renames (see §4): a shallow duplicate would collide names → values mirror; renaming keeps names unique → values independent.
• Consequence for the scanner: just enumerate every unique frl_/frd_ layer and emit one content-element per name — no repeat-detection needed.

---

2. Project modes (5)

Mode	Render comp	Timeline source	Duration	Binder generator	Output
FIX	frfinal	fixed (template)	= project_duration_sec	CreateV2FixJSX	video
FLEXIBLE	flatrender (assembled)	user story (Normal scenes, duplicated)	Σ scene + first overlap, clamped to project.max	CreateFlexibleJSX	video
MockUp	flatrender (assembled)	one frame per scene	Duration = sceneCount (FrameRate=1)	CreateMockupJSX	images (JPG/scene)
MusicVisualizer	frfinal	fixed (template)	min(audioLen, project.max)	CreateV2FixJSX	video
VoiceOver	(like Fix/MusicViz, timed to a VO track — to confirm)	fixed	= VO length, capped at project.max	tbd	video

Differentiators: does the user control the timeline? (Fix=no, Flexible=yes) · content-swap vs design-placement (Fix/Flexible vs Mockup) · audio-driven? (MusicViz/VoiceOver).

---

2.1 Output spec — aspects · quality · frame rate

Aspects (video) — only these three:

Aspect	Dimensions
16:9	1920 × 1080
1:1	1080 × 1080
9:16	1080 × 1920

Render quality tiers (output resolution, chosen per render): 360p · 480p · 720p · 1080p · 4K.

• This is a render-time output tier (downscales the comp), distinct from the project's native design resolution.
• The project's Resolution acts as the ceiling; the user picks a tier ≤ ceiling, also bounded by their plan.

Free vs paid:

	Free	Paid
Max quality	360p	up to project ceiling (4K)
Watermark	yes	no
Frame rate	21 fps (default, per-project configurable)	project fps

Frame rate:

• Free renders use 21 fps by default — stored per project (free_fps), configurable in project settings.
• Maximum frame rate = 60 (any mode/tier).

3. Scene types & roles

scene_type	User-pickable?	Role in assembly
Normal	✅ yes	Story scenes — duplicated into the timeline.
Config	❌ no	Global shared layers/config — applied across the whole comp (not on the timeline).
DesignStart	❌ no	BottomDesign — full-duration background, bottom of the layer stack.
DesignEnd	❌ no	TopDesign — full-duration overlay (logo/frame), top of the layer stack.

Design scenes are authored once in the template and injected at render time (single instance, full duration — no _d{N}).

---

4. flatrender assembly (Flexible / Mockup)

Layer stack, bottom → top (z-order = depth):

(global)               Config            → shared layers/colours applied across
bottom (behind)        DesignStart       → background, full duration                [auto · 1×]
middle (time-line)     Normal story      → sequenced instances, deep-dup _d{N}      [user-picked]
top (in front)         DesignEnd         → overlay, full duration                   [auto · 1×]

Story sequencing (Normal scenes), in Sort order:

• Each instance = a deep-duplicated scene comp renamed <sceneKey>_d{N} (N = 1-based duplication index). Inner frl_/frd_ names stay identical (each clone is its own namespace).
• Instance layer length = SceneLength + OverlapAtEnd; placed at the running time offset; consecutive scenes cross-overlap by OverlapAtEnd.
• Per-instance inputs are an ordered list of input-sets, one per instance (SceneStory(Name=sceneKey) → Duplications[] → Inputs[]).

---

5. Duration model (universal across modes)

Per scene: default_duration_sec, min_duration_sec, max_duration_sec, overlap_at_end_sec, can_handle_duration (may the user change it?). Per project: project_duration_sec (default/target), min_duration_sec, max_duration_sec (hard output cap).

Flexible:

each scene:  min ≤ SceneLength ≤ max
total     =  Σ SceneLength (+ first scene overlap)
clamp     =  min(total, project.max_duration_sec)        ; preview hard-capped at 180s

MusicVisualizer / audio modes:

audioLen ≤ max → duration = audioLen          (no min floor; short audio = short video)
audioLen > max → user trims (≤ max, decrease-only)   OR   untrimmed → max + fade-out

• Trimmer window is capped at project.max (can shrink, never exceed).
• Cap is enforced in both frontend (UX) and backend (authority — never trust the client).

Fade-out (audio modes): default 1.5 s, admin-configurable, studio-toggleable.

projects.audio_fade_out_sec        default 1.5            (admin default duration)
scenes.audio_fade_out_sec          nullable override      (per-item)
projects.audio_fade_out_enabled    default on             (admin default state)
projects.is_audio_fade_changeable  default true           (may the studio user toggle?)
studio: fade_out_enabled           user choice            (only if changeable)

→ fadeEnabled  = is_audio_fade_changeable ? userChoice : project.audio_fade_out_enabled
→ fadeDuration = item.audio_fade_out_sec ?? project.audio_fade_out_sec ?? 1.5

(These four projects/scenes columns + the studio flag are NOT yet in the schema — added with the audio/binder work.)

---

6. Responsive fit-box (per element)

Each frl_ element is its own precomp (a box/region) with a fixed area; content auto-fits inside it.

• Text: single-line or multi-line (TextArea); auto-scaled down if it overflows (is_text_box, max_size); wrapped for multi-line.
• Alignment in box: justify (LEFT/CENTER/RIGHT/FULL_JUSTIFY) + position_in_container anchors 0–8 (0 = centre, corners/edges).
• Media: scale-to-fit the box, keep aspect, centred.

This is why deep-dup is heavy: every element is a nested comp, so a scene is comps-inside-comps; cloning duplicates the whole tree (exactly what v2 avoids).

---

7. Input types (scanner classification targets)

Input type	content_element_type	AE detection
image input	Media	footage/placeholder layer
image / video input	Media (video_support)	footage layer
text / multi-text input	Text / TextArea	text layer (single vs multi-line box)
text input	Text	text layer
number input	Number	text layer (numeric)
option input (list)	DropDown	data layer + mapped_list
yes / no input	CheckBox / Toggle	data layer (0/1)

The detection walks layers top → bottom, classifies each editable layer into one of these, and emits scene content-elements in that order.

---

8. Expression-driven data flow ⚠️

Visible layers do not hold values — they run AE expressions that read from a central source:

binder writes ─►  frshare colour layers  +  frd_ data layers   ─►  expressions  ─►  visible frl_ layers
                  (one write)                                       (propagate automatically)

• The binder writes only the data sources (a colour once in frshare, a value in a frd_ layer). Expressions fan it out — it never touches each visible layer.
• The scanner reads the data sources (frshare layers + frd_ data layers) as the source of truth for colours/values — not the expressions.
• Colours live in frshare: layer name = element key, sourceText = colour value (hex / r,g,b).

---

9. Convention versioning + the rule engine

convention_version is an admin-set field on the project (chosen in the add/edit panel, default = latest). The rule engine maps a version → the rules the scanner + binder follow. Both versions are supported (hybrid) so nothing breaks.

Aspect	v1 — legacy (current)	v2 — Master Properties (proposed)
Per-instance independence	deep-dup + rename _d{N}	1 source comp + N layer instances, each with Master/Essential-Property overrides
Scan target	layer-name prefixes (frl_/frd_)	comp.masterProperties / Essential Graphics (typed + named)
Media per instance	duplicated footage	Media Replacement essential property
Expressions	AE auto-repoints intra-clone expressions	global colours stay expression-from-shared; per-instance values flow from the instance's master property into the internal expression/fit
Cost	N full comp trees (heavy)	tiny project, faster open/render

Shared rules (both versions): §1 naming · §3 scene-type roles · §4 z-order assembly · §5 duration/overlap/fade · §6 fit-box · §7 input types · §8 expression data flow.

v2 golden rule: global → expression-from-shared; per-instance → (v1: dup+rename · v2: master-prop feeding the expression/fit).

⚠️ v2 is "validate-first"

Expressions are the #1 thing that complicate Master Properties (expression-driven prop vs master-prop override; expression scope across the master-property boundary). v1 stays the safe default. Before re-plumbing to v2, spike on one real template to confirm promotion + media replacement + expression scope behave. (Defer to the AE author here.)

---

10. What this drives in the V2 codebase

• projects.convention_version (+ admin panel selector) — to add.
• Scanner (scan.jsx / Go quick-scan) — branch on version: v1 reads names, v2 reads Essential Graphics. Emits the canonical ScanResult the importer consumes.
• Render binder — not yet built. Hybrid generator: v1 = deep-dup + rename + bind data sources; v2 = layer instances + master-prop overrides. Per-mode (CreateV2FixJSX / CreateFlexibleJSX / CreateMockupJSX equivalents). This is the port of the legacy JSXGenerator.cs.
• Audio/duration layer — projects/scenes fade + duration columns, trimmer-cap-to-max, MusicViz resolution, fade-out. (One migration, one rebuild.)
• Dynamic scene generation — the end goal: given business + logic, select/assemble/author scenes against these rules.

---

11. Open items to confirm / validate

• VoiceOver mode mechanics (separate mode vs flavor of Fix/Flexible).
• Exact flatrender timeline offset formula (cumulative Duration − overlap).
• v2 Master-Properties spike on a real template (expressions + media replacement + scope).
• frfinal vs flatrender — confirm flatrender is always the assembled comp (built by the binder) for Flexible/Mockup.