zwitschi
bd1f8a9c2e
Tools/visual_loop.py:
- --verify gates every capture against its expectation in Tools/visual_targets.json
(missing, blank, tone-less or wrong-size shots fail), so 'each screenshot shows
what was expected' is checked rather than assumed.
- --compare measures luminance, RMS contrast, edge density, palette distance and a
3x3 luma grid against the reference photograph, with a 0-1 similarity score.
- --iteration N writes Art/reference/iterations/N/{metrics.json,feedback.md}; the
feedback names the skill to consult (blender-modeler, archviz, materials, lookdev,
camera-cinematography), so the next modelling pass starts from a measurement.
- Iterations are gitignored until the owner confirms a shot; only confirmed shots
are promoted into Screenshots/.
Screenshots/ holds demo-map captures only now: render_screenshots.sh renders just
DemoMap, collects the shots and runs the verification gate (--verify); the review
staging map is no longer rendered for review and its PNGs are removed. The README
is rewritten around the four demo shots with per-shot expected content and a
verified column, and no longer claims the review staging set.
Tests: 67 pass (13 new) covering metrics on synthetic images, comparison scoring,
the feedback rules and the target spec (including that every target is a demo shot
and that iterations are gitignored).
Screenshots — demo-map visual review
Rendered captures of the demo map only — what a player sees when the demo runs. Generated by Tools/render_screenshots.sh; the PNGs are committed so a reviewer sees exactly what the build rendered.
The review staging map (/Game/Test/ReviewMap) is not part of this set: it exists to hold the camera list and the hero detail rig, and it is not what ships.
How to regenerate
Tools/render_screenshots.sh --verify
Stages the demo shot cameras, runs the game on /Game/Hollywood/DemoMap, copies the PNGs into Screenshots/demo/ and then runs Tools/visual_loop.py --verify to check that each shot still shows the content its expectation describes. About a minute.
The shot list
Each shot has an expectation in Tools/visual_targets.json — the content the capture must show — and a verification state produced by --verify: it fails if the frame is blank, the wrong size, or has no tonal range.
| # | File | View | Expected content | Verified |
|---|---|---|---|---|
| 1 | demo/review-demo-spawn-view.png |
north sidewalk | The whole south block: the neoclassical hero façade between its neighbours, both sidewalks and the road in the foreground | ☑ 2026-11-09 |
| 2 | demo/review-demo-street-level.png |
eye height, north side | Street level: the hero colonnade with the adjacent frontage buildings either side, shopfronts at pavement level | ☑ 2026-11-09 |
| 3 | demo/review-demo-crossing.png |
along the crosswalk | The designated mid-block crossing (s=130 m, UE X −14956): zebra bars on the asphalt, south frontage and colonnade behind | ☑ 2026-11-09 |
| 4 | demo/review-demo-north-buildings.png |
across the Boulevard | The two north-side buildings (TCL Chinese Theatre, Madame Tussaud's) from the south side, full height in frame | ☑ 2026-11-09 |
Approval is per shot, in the table above. Add a row when a shot is added; remove the file in the same change when a shot is retired.
The iteration loop (uncommitted)
Tools/visual_loop.py measures each capture against its reference photograph and turns the deltas into modelling actions for the Blender skill packs:
python3 Tools/visual_loop.py --compare --iteration 3 # writes Art/reference/iterations/3/
That directory is gitignored: iterations stay uncommitted until a shot is confirmed, and only confirmed shots are promoted into Screenshots/demo/. The feedback file names the affected skill (blender-modeler, archviz, materials, lookdev, camera-cinematography) so the next modelling pass has a starting point rather than an opinion.
Note the comparison is only meaningful when the shot and the reference frame the same subject — a wide block view measured against a tight façade crop will always report low edge similarity.
Scene conventions
- Demo map (
/Game/Hollywood/DemoMap): fixed daytime — sun pitch −62 / yaw 120 (intensity 4, warm), sky light 2.6, sky atmosphere, height fog, fixed exposure. - Cameras: elevation shots sit at their target height (pitch 0, so verticals stay parallel) and use a 42–55° lens to fill the frame from across the street.
- Materials: every surface carries a look-developed material from
Tools/ue_build_demo_map.py— noise-blended base colour, masonry coursing, and a generated texture set (Tools/generate_facade_textures.py) sampled world-aligned. Explicit roughness and specular per surface type.
Known cosmetic issues
- Hero façade renders pale: the stone material is a graph and a texture set, not a scanned surface (the
0.1.4façade pass is in progress). - Banners sit on the column line rather than flush to the wall (they hang on the columns in the photographs, so this is acceptable).
- Between kerb and building line the walkable surface is the terrain mesh, not a modelled pavement.
- The El Capitan Theatre has no marquee or blade sign yet — the feature that makes it read as itself.