Toward V2 — performance milestone 18

Kite3D 0.19 alpha 2 — frame and startup work · R11

Intermediate — performance acceptance pending. No scene approval.

Capture date range (UTC): 2026-09-14T09:30:17+00:00 — 2026-09-14T09:30:36+00:00
Unmodified source PNG file modification times.
Actual progress source SHA: 6552518fe50b1cd4f818e1db5a03097b15b88f8b
Capture source: mac-alpha-dynamic-lights-05
Gallery built: 2026-09-14T10:22:40+00:00

Before is the original real Mac capture. New target is the immutable selected V2 generated reference. Latest progress is an actual Mac game capture. All five original player POVs; images copied byte-for-byte without edits. Strict camera/config/device/render and source-hash checks passed. Lower RGB MAE alone is not semantic success.

Measurement details and limitations

Provisional 650s timing: possible interference from accidentally auto-opened peer tabs was discovered after collection. No Linux performance acceptance or clean comparative speedup is claimed. 650s Linux, NVIDIA RTX 3090 Ti, 1920×1080: engine-only populated combat 24.5 FPS; final dynamic-light frame candidate 30.0 FPS, CPU median/p95 12.9/15.3 ms. A separate 60-second moving route covered 299 metres with 24 actors at 29.2 FPS. These are not Mac acceptance numbers.

Apple M3 Max: combined-build measurements varied from 31.9 to 74.3 FPS in headless moving runs. The later normal headed 120 Hz-display run reached 90.9 FPS over 60 seconds with 24 actors and 299 metres of movement; CPU median/p95 10.3/13.7 ms. High actually rendered 1512×982 pixels, not the display’s physical 3024×1964 pixels. Final-build repeat and frame-tail margin remain to verify.

Targets remain sustained 75 FPS (13.33 ms frame budget) and under 10 seconds from editor Play to responsive gameplay. Actual immediate-transition startup remains over budget: the latest unprofiled Mac combined measurement was 14.96 seconds. Startup profiling and image-upload work continue; no synthetic three-second menu wait is included in this acceptance total.

This is an intermediate performance milestone, not final acceptance. The five progress images show the combined dynamic-light source on Mac; fixed images, populated timings and startup measurements are labeled separately. Art and smoke refinement remain deferred to issue #2. Original visual targets and all 17 earlier milestones are retained.

BeforeV2 targetLatest progress

01-courtyard

Original real before · Mac: 01-courtyard
Original real before · Mac · click for original resolution
Selected V2 · generated target: 01-courtyard
Selected V2 · generated target · click for original resolution
Actual latest progress · Mac: 01-courtyard
Actual latest progress · Mac · click for original resolution

02-cargo

Original real before · Mac: 02-cargo
Original real before · Mac · click for original resolution
Selected V2 · generated target: 02-cargo
Selected V2 · generated target · click for original resolution
Actual latest progress · Mac: 02-cargo
Actual latest progress · Mac · click for original resolution

03-barracks

Original real before · Mac: 03-barracks
Original real before · Mac · click for original resolution
Selected V2 · generated target: 03-barracks
Selected V2 · generated target · click for original resolution
Actual latest progress · Mac: 03-barracks
Actual latest progress · Mac · click for original resolution

04-service

Original real before · Mac: 04-service
Original real before · Mac · click for original resolution
Selected V2 · generated target: 04-service
Selected V2 · generated target · click for original resolution
Actual latest progress · Mac: 04-service
Actual latest progress · Mac · click for original resolution

05-rooftop

Original real before · Mac: 05-rooftop
Original real before · Mac · click for original resolution
Selected V2 · generated target: 05-rooftop
Selected V2 · generated target · click for original resolution
Actual latest progress · Mac: 05-rooftop
Actual latest progress · Mac · click for original resolution

Image hashes and source provenance. Matched unchanged Linux baseline is retained as supporting evidence:

The original prompt

User-message excerpts verified against the original session record. UTC timestamps and line references are retained; the private transcript and attachment paths are not included. Original spelling and casing are preserved.

Original scene-refactor prompt

· session line 1257

we need to seriously revamp and improve the scene. take multiple screenshots of the scene from different angles from the perspective the player would see right now. then write a prompt describing the ambitious final outcome we want from the scene (treat the screenshot almost like a depth map that steers the image generator to follow structure, that's how little detail and quality it has right now). do this for at least 5 different player POVs around the map. each POV shoudl show major parts of the map, enough to grasp the "gestalt" of the map. then show pairs of 5 screenshots -> 5 image gneerations (before/after) in a html file. we'll use the generated images as target refernece from now on when refactoring the scene. the plaeyr POVs should show what the play actually sees right now.

Repeatable capture and pixel comparison

· session line 1265

and the script used to take the SC should be saved, so we can compare before/after after actually doing the refactor, so we can use the pixel-diff agains the reference images as error to hill climb

Film-style correction

· session line 1495 · text portion; attachment omitted

no these image styles are bad. specifically mention james cameron and the graphics style of the original terminator movie, "https://sites.pitt.edu/~goscilo/Sci-Fi/FilmStills/TheTerminator.html". try agian. it should give vibe of this image: [Image #1]

V2 selection

· session line 1753

lets just go with v2. what's your plan for converging on v2?

Remote Astra high worktrees

· session line 1787

how will you parallelize against multiple astra high agents? i prefer you spawn agents in the 650s machine for all work (use the cli on that machine) and only integrate here when their work is up to standards. each agent should work in a worktree branch and should work on orthogonal part of the entire task. tldr ur plan

Major milestone comparison columns

· session line 2549

can you update the html periodically at major milestons to show the updated scene? add a new column (before | new target | latest progress)
Initial asset request
help me find better assets for all the scene assets in the terminator scene. in a new html page, per asset row show in the first col the original asset and in 3 cols 3 diff alternative assets you found on the internet

Public source of this copy · Back to the five views