Blender Film Kit · first pass
A late-night coin laundry, built from a reference and rendered on CPU
Proof that the pipeline works end to end: reference → layout → blockout → detail → one shared procedural weathering group → practical lighting → shot list → frames → encode. Not a finished shot. A working factory with a known quality gap.
Motion test — 1 second, 24 fps
Lookdev still — 960×540
What is actually verified
- Scene builds and renders headless from Python; no GUI, no manual steps.
- One weathering node group drives edge wear, gap grime, top dust, floor dirt and drips across every material, each at its own strength.
- Print signage is designed in 2D, aged, and mapped onto UV quads — real type, never 3D text.
- Shot renderer reads a JSON spec, renders resumable frames, grades and encodes h264.
- Encode output confirmed by ffprobe: 24 frames, 1.000 s, 24/1 fps, 480×270.
The constraint that shapes everything
EEVEE cannot run in this container — no GPU, no EGL surface, it aborts on context creation. Cycles on 4 CPU cores is the only engine. A real interior costs about 3.3× more per pixel-sample than a box test, which puts production 1080p at roughly 13 minutes per frame — 5.4 hours per second of footage.
- 720p · 16 samples
- ~1.5 min/frame · 36 min per second
- 1080p · 64 samples
- ~13 min/frame · 5.4 hours per second
- Rented 4090, 1080p
- roughly 20–60× faster · a 10 s shot in 1–2 hours
Gap to the reference
The result in the original thread is many hours of layering on top of this exact structure: per-unit wear variation, props and clutter, dust, print materials everywhere, and a lighting contrast pass. The structure was the missing part and it now exists. What remains is detail budget — which is a render-time question, not a capability question.