MilkPEP x NexRender

Three layers. One contract.

The system is composed of three distinct layers that communicate exclusively through a shared data schema. The web front end collects user input. NexRender carries that input into After Effects. After Effects evaluates it at render time. No layer has any direct knowledge of the others — they share only a vocabulary of keyword strings.

Data flow

Identical contract. Unique content.

Every template in the library shares the same structural specification regardless of the recipe it contains. The specification defines reserved layer names, required composition names, the expression pattern for variety switching, and the asset naming conventions that the render pipeline depends on.

2.1 — Composition structure

Each template contains exactly two compositions whose names are reserved and must not be altered.

2.2 — The MILK_LABEL layer

A placeholder layer named MILK_LABEL exists in every template. It is blank in the source file. NexRender replaces this layer with the user-uploaded label asset at job execution time by matching the layer name string exactly.

• The layer is pre-parented to a null object positioned at the vessel label zone
• The null carries the vessel-specific transform: position, rotation, and the perspective distortion derived from the depth scan
• Displacer Pro applies the depth map to the incoming label asset via the null's Effect controls, adding surface curvature and foreshortening
• The layer blend mode and track matte are set in the source file; the replacement asset inherits them
• The layer name MILK_LABEL is case-sensitive and must match exactly as written

2.3 — Variable text layers

Four text layers in MILK_EXPORT hold the variable values. Their names are reserved. NexRender's setText action targets them by name and writes the uppercase schema keyword value before rendering begins.

2.4 — Expression pattern

All visibility control uses the same expression pattern. An expression on a precomp's Opacity property reads the source text of the relevant variable layer and evaluates it against the keyword value associated with that precomp. The expression returns 100 if the strings match, 0 if they do not.

This pattern is intentionally minimal. The text layer contains the raw uppercase keyword string with no whitespace and no formatting. Any deviation — a trailing space, a capitalisation difference — causes the expression to evaluate to 0 and the precomp to be invisible in the render.

2.5 — Asset naming constraints

Long file names cause decompression failures when NexRender unpacks the project bundle for rendering. This limit was discovered during testing and is not documented by NexRender. All source assets in the templates use the following conventions:

• Source footage files: maximum 20 characters including extension
• Depth map files: prefixed dm_, short descriptor, extension — e.g. dm_jug_hg.exr
• Precomp assets: prefixed with a two-letter vessel code — e.g. jg_white_hg (jug, white, half-gallon)
• No spaces in any asset name; underscores only
• No special characters, accented characters, or Unicode in asset paths

One keyword table. Three systems.

The Milk Dynamic Video Schema defines the canonical keyword string for every variable in the system. The front-end form labels, the backend API field names, the NexRender action targets, and the After Effects expression comparisons all reference the same document. The schema is the only interface between the three layers.

Full variable table — v2

Keyword string rules

• All keywords are uppercase (CAPS LOCK). This matches the naming convention of the layer names and composition names they interact with.
• Multi-word values use hyphens only — no spaces, no underscores. Hyphens are retained in uppercase: HALF-GALLON, LACTOSE-FREE
• The keyword string must match the After Effects expression comparison character-for-character, including hyphenation and case
• New keyword values cannot be added without a corresponding precomp layer and expression branch in all affected templates

What NexRender does and does not do.

NexRender is a headless After Effects render automation system. It accepts a job manifest — a JSON object — opens the specified AE project on a render node, makes the modifications described in the manifest, and executes the render queue. It has no awareness of the content it is rendering. It reads strings and writes them to layers.

4.1 — Job manifest structure

4.2 — Render node environment

4.3 — Pre-render vs. render-time execution

The setText actions are pre-render actions — they execute before the AE render queue is opened. This is critical: AE evaluates expressions at render time against the layer state in the project file. If text layer values are written after the render begins, expressions have already been evaluated and the variable switch has no effect. All variable injection must complete in the pre-render phase.

The label asset replacement via layerName matching also occurs in the pre-render phase. NexRender locates the layer by name, downloads the asset from the specified URL, and replaces the source footage reference before AE opens the composition.

The label sits in the scene, not on it.

The footage was shot with vessel placement and lighting engineered for a subsequent depth-scanning pass. The depth pipeline produces a per-frame displacement map that allows an incoming label asset to be composited with accurate perspective foreshortening, surface curvature, and lighting response — without any manual tracking or matchmoving.

5.1 — Depth map generation

5.2 — Safe zone

Each vessel precomp defines a label safe zone — a rectangular region on the vessel surface within which any processor's label will be legible and correctly composited regardless of label artwork dimensions. The safe zone was defined during the shoot by marking the vessel with reference tape and is encoded in the vessel null's transform values in the template.

The MILK_LABEL layer is scaled to fill the safe zone by default. Labels with non-standard aspect ratios are letterboxed within the safe zone; the letterbox background matches the vessel colour inferred from the cap colour variable.

5.3 — Lighting response

The depth map drives a secondary luminosity pass on the MILK_LABEL layer. The same practical light used on the original shoot is simulated as a directional gradient, derived from the depth map's surface normal approximation. This gives the label the appearance of being lit by the same source as the surrounding footage rather than pasted flat over it.

Where 12,460 comes from.

The 12,460 figure is the total number of distinct renders the system can produce across all recipes, vessel configurations, milk varieties, cap colours, and output formats. It is not a count of renders produced — it is the scope of the system. Each combination is a valid, distinct After Effects render with a unique precomp configuration, label placement, and expression state. All recipes produce a 9:16 output; some also produce 4:5 and 1:1 outputs.

The complexity implied by this number lives entirely in the template build. Every cap colour that needed to work required a precomp layer, a colour-matched asset, and an expression branch. Every carton size that differed from the default required a size-specific composition with its own depth-adjusted label transform and a corresponding safe zone calibration. Every milk variety required a footage variant — white milk and chocolate milk have visibly different liquid colour — with a separate depth-composited label zone.

At render time, NexRender writes four keyword strings and replaces one layer. The 12,460 combinations are resolved before the render job is ever queued.

What breaks and why.

The following failure modes were identified during alpha and beta testing. Most are silent — NexRender does not surface AE expression errors in its job logs, and a render that completes with the wrong variable state is indistinguishable from a correct render at the API level.

AE version sensitivity

Templates built in After Effects 2025 use expression engine features and effect parameter names that differ from AE 2023 and earlier. Running a template in an older AE version on a misconfigured render node will cause expression errors that are not surfaced in the NexRender log. The render may complete with all precomps defaulting to 0 opacity. The AE version on the render node is pinned and verified as part of the environment setup.