Add OpenRouter AI texture generation mode to the asset pipeline

- tools/generate_textures.py: --ai mode (OPENROUTER_API_KEY, --prompt,
  --name, --model, --size; base64 PNG decode; clamp to <=1024px) with a
  default batch covering the authored texture set; seedream minimum-size
  handled by requesting 2048 and downscaling
- Generated AI asset set committed under assets/textures/ai/ (15 files,
  1024x1024 each), kept separate from the deterministic PIL set
- README/MEMORY/TODO document the mode and live verification
- (work authored in a parallel session; linted/verified here — 57/57 tests)
This commit is contained in:
zwitschi committed 2026-10-31 14:04:37 +01:00
1 parent 0abd67fbaa
commit 5b994982bd
19 files changed
+274 -23

No files matched your search

+8
View File
@@ -59,6 +59,14 @@ Session memory for durable decisions, state, and gotchas. Keep concise, remove o
- Assets pipeline: `python3 tools/generate_textures.py` produces all authored PNGs under
`assets/textures/` (≤1024px, pow2); regenerate after changing the script, then run
`godot --headless --import` before tests. 57 headless tests guard the pipeline.
- AI texture mode: `python3 tools/generate_textures.py --ai` generates textures via the
OpenRouter image API (`POST https://openrouter.ai/api/v1/images`, model
`bytedance-seed/seedream-4.5` by default, base64 PNG in `data[].b64_json`). Requires
`OPENROUTER_API_KEY` (a gitignored `.env` in the repo root holds it locally; source it
with `set -a && . ./.env && set +a`). `--prompt`/`--name`/`--model`/`--size` override the
default batch. Output is downscaled to ≤1024px. NOTE: seedream rejects sizes below
3,686,400 px (1920×1920), so the script requests `GEN_SIZE=2048` and downscales — do not
lower `GEN_SIZE` below 1920. Verified end-to-end 2026-10-31 (3 textures generated).
- Detailed graphics: textured facade/awning/skyline, humanoid NPC rigs + props + walk bob +
attention head-turn, inspection artifact cards, mode icons, escalation vignette, body-cam
overlay (REC + shift clock). Screenshot set regenerated (`docs/screenshots/`, day + night).
+1 -1
View File
@@ -136,7 +136,7 @@ The architecture maps onto Godot 4 nodes, scenes, and custom resources as descri
- **Navigation:** `NavigationRegion3D` and `NavigationAgent3D`
- **NPC behavior:** Custom lightweight finite state machines driven by `NpcProfile` resources
- **UI and audio:** Godot `Control` nodes, themes, audio buses, and `AudioStreamPlayer` nodes
- **Art assets:** authored PNGs in `assets/textures/` (≤1024px, power-of-two) produced by `tools/generate_textures.py` and imported by Godot; structural geometry stays procedural (`WorldBuilder`)
- **Art assets:** authored PNGs in `assets/textures/` (≤1024px, power-of-two) produced by `tools/generate_textures.py` and imported by Godot; structural geometry stays procedural (`WorldBuilder`). The same script can generate textures through the OpenRouter image API (`--ai`, requires `OPENROUTER_API_KEY`).
- **Testing:** GUT (Godot Unit Test) in headless mode
- **Source control:** git on a Gitea host with Gitea Actions (`.gitea/workflows/build.yml`) for headless tests and desktop release builds
- **Targets:** Windows, Linux, and macOS desktop first; web later
+10
View File
@@ -1,5 +1,15 @@
# TODO
## 🎨 AI texture generation (OpenRouter)
- [x] Extend `tools/generate_textures.py` with an OpenRouter image-API mode
(`--ai`, `--prompt`, `--name`, `--model`, `--size`), decoding base64 PNG
responses and clamping output to ≤1024px.
- [x] Verify a live `--ai` generation against a real `OPENROUTER_API_KEY`
(generated `ai/brick_facade.png`, `ai/concrete.png`, `ai/asphalt.png` at
1024×1024; fixed the seedream min-pixel requirement by requesting 2048px
and downscaling).
## 🪟 Windows installer (self-contained, all game data)
Goal: a real Windows installer that includes the entire game (executable + data),
Binary file not shown.

After

Width:  |  Height:  |  Size: 2.1 MiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 856 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 2.0 MiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 436 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 929 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 671 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 1.2 MiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 578 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 1.4 MiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 568 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 360 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 675 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 343 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 441 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 801 KiB

+255 -22
View File
@@ -3,17 +3,93 @@
All outputs are deterministic (fixed seeds), <=1024px, power-of-two where
possible. Run from the repo root: python3 tools/generate_textures.py
Optional AI mode generates textures through the OpenRouter image API
(https://openrouter.ai/docs/api/api-reference/images/generate-an-image.md).
Requires an API key in the OPENROUTER_API_KEY environment variable:
python3 tools/generate_textures.py --ai
python3 tools/generate_textures.py --ai --prompt "..." --name foo.png
"""
import argparse
import base64
import io
import json
import math
import os
import random
import stat
import sys
import urllib.error
import urllib.request
from PIL import Image, ImageDraw, ImageFilter, ImageFont
ROOT = os.path.join(os.path.dirname(os.path.abspath(__file__)), "..")
ASSETS = os.path.join(ROOT, "assets", "textures")
# OpenRouter image generation endpoint and default model (from the API docs).
OPENROUTER_IMAGES_URL = "https://openrouter.ai/api/v1/images"
DEFAULT_IMAGE_MODEL = "bytedance-seed/seedream-4.5"
MAX_DIM = 1024 # hard cap on the longest side of any generated texture
# px requested from the API; some models (e.g. seedream) need >= 1920
GEN_SIZE = 2048
# Default batch for `--ai`: one entry per game texture (see test_world.gd).
# Each entry is (output name, prompt, transparent_background).
AI_TEXTURES = [
# --- world surfaces (seamless, flat-lit) ---
("ai/brick_facade.png",
"seamless tileable red brick wall texture, weathered urban facade, "
"flat even lighting, frontal view, photorealistic game texture", False),
("ai/concrete.png",
"seamless tileable grey concrete wall texture, subtle cracks and stains, "
"flat even lighting, frontal view, photorealistic game texture", False),
("ai/asphalt.png",
"seamless tileable dark asphalt road texture, fine grain, "
"flat even lighting, frontal view, photorealistic game texture", False),
# --- storefront dressing ---
("ai/awning.png",
"red and white striped storefront awning with scalloped bottom edge, "
"frontal view, flat even lighting, game texture", False),
("ai/skyline.png",
"night city skyline silhouette, dark building shapes with a few lit "
"windows, flat graphic style, isolated on transparent background", True),
# --- screen-space overlay ---
("ai/vignette.png",
"radial vignette overlay, black edges fading to fully transparent center, "
"grayscale, soft gradient", True),
# --- HUD icons (white glyphs on transparent) ---
("ai/icons/flashlight_on.png",
"simple flat flashlight icon, white glyph on transparent background, "
"minimal UI icon", True),
("ai/icons/door_locked.png",
"simple flat locked door icon with keyhole, white outline glyph on "
"transparent background, minimal UI icon", True),
("ai/icons/wand.png",
"simple flat handheld security wand icon, white glyph on transparent "
"background, minimal UI icon", True),
("ai/icons/rec_dot.png",
"record button icon, red filled dot inside a white circle outline, on "
"transparent background, minimal UI icon", True),
# --- access artifact cards ---
("ai/cards/id_card.png",
"driver license ID card, blue plastic card with photo placeholder and "
"text lines, flat frontal view, game UI asset", False),
("ai/cards/wristband.png",
"red event access wristband with strap, flat frontal view, game UI asset",
False),
("ai/cards/ticket.png",
"green admission ticket stub with perforated edge, flat frontal view, "
"game UI asset", False),
("ai/cards/credential.png",
"dark grey staff credential pass with lanyard clip, flat frontal view, "
"game UI asset", False),
("ai/cards/guest_list.png",
"beige guest list document with ruled lines and checkboxes, flat frontal "
"view, game UI asset", False),
]
FONT_CANDIDATES = [
"/usr/share/fonts/liberation-sans-fonts/LiberationSans-Bold.ttf",
"/usr/share/fonts/dejavu/DejaVuSans-Bold.ttf",
@@ -21,12 +97,14 @@ FONT_CANDIDATES = [
"/usr/share/fonts/adwaita-sans-fonts/AdwaitaSans-Bold.ttf",
]
def font(size: int):
for path in FONT_CANDIDATES:
if os.path.exists(path):
return ImageFont.truetype(path, size)
return ImageFont.load_default()
def save(img: Image.Image, name: str) -> None:
os.makedirs(ASSETS, exist_ok=True)
path = os.path.join(ASSETS, name)
@@ -34,6 +112,7 @@ def save(img: Image.Image, name: str) -> None:
img.save(path, optimize=True)
print(f"wrote {os.path.relpath(path, ROOT)} {img.size}")
def grain(size, base, spread, seed, blur=1.0):
rnd = random.Random(seed)
img = Image.new("L", (size, size), base)
@@ -44,6 +123,7 @@ def grain(size, base, spread, seed, blur=1.0):
px[x, y] = max(0, min(255, v))
return img.filter(ImageFilter.GaussianBlur(blur))
def brick_facade():
size = 512
mortar = (118, 110, 104)
@@ -68,13 +148,14 @@ def brick_facade():
bx = rnd.randint(0, size - 1)
by = rnd.randint(0, size - 1)
d.point((bx, by), fill=(max(0, base[0] - 18 + rnd.randint(0, 30)),
max(0, base[1] - 10 + rnd.randint(0, 22)),
max(0, base[2] - 8 + rnd.randint(0, 18))))
max(0, base[1] - 10 + rnd.randint(0, 22)),
max(0, base[2] - 8 + rnd.randint(0, 18))))
# horizontal mortar lines on top of bricks (wraps vertically by drawing at 0 and size)
for y in [-mortar_t, size - mortar_t]:
d.rectangle([0, y, size, y + mortar_t], fill=mortar)
save(img, "brick_facade.png")
def seamless_concrete():
size = 512
img = grain(size, 150, 26, 7, blur=2.5).convert("RGBA")
@@ -83,7 +164,8 @@ def seamless_concrete():
for _ in range(90):
x, y = rnd.randint(0, size), rnd.randint(0, size)
r = rnd.choice([2, 3, 4])
d.ellipse([x - r, y - r, x + r, y + r], fill=(0, 0, 0, rnd.randint(8, 22)))
d.ellipse([x - r, y - r, x + r, y + r],
fill=(0, 0, 0, rnd.randint(8, 22)))
for _ in range(8): # subtle crack segments
x, y = rnd.randint(0, size), rnd.randint(0, size)
pts = [(x, y)]
@@ -94,6 +176,7 @@ def seamless_concrete():
d.line(pts, fill=(70, 70, 70, 170), width=1)
save(img, "concrete.png")
def asphalt():
size = 512
img = grain(size, 52, 14, 23, blur=1.2).convert("RGBA")
@@ -101,9 +184,11 @@ def asphalt():
rnd = random.Random(31)
for _ in range(600):
x, y = rnd.randint(0, size), rnd.randint(0, size)
d.point((x, y), fill=(max(0, 40 - rnd.randint(0, 18)),) * 3 + (rnd.randint(40, 110),))
d.point((x, y), fill=(max(0, 40 - rnd.randint(0, 18)),)
* 3 + (rnd.randint(40, 110),))
save(img, "asphalt.png")
def awning():
w, h = 1024, 256
img = Image.new("RGB", (w, h), (140, 40, 45))
@@ -111,7 +196,8 @@ def awning():
stripe = 64
for i in range(w // stripe):
if i % 2 == 1:
d.rectangle([i * stripe, 0, (i + 1) * stripe - 2, h], fill=(230, 225, 220))
d.rectangle([i * stripe, 0, (i + 1) * stripe - 2, h],
fill=(230, 225, 220))
# scalloped bottom edge
for x in range(0, w, 32):
d.arc([x, h - 24, x + 32, h + 40], 0, 180, fill=(80, 20, 24), width=6)
@@ -125,6 +211,7 @@ def awning():
img = Image.alpha_composite(img.convert("RGBA"), shade)
save(img, "awning.png")
def skyline():
w, h = 1024, 256
img = Image.new("RGBA", (w, h), (0, 0, 0, 0))
@@ -141,11 +228,13 @@ def skyline():
wx = x + rnd.randint(4, max(4, bw - 8))
wy = h - rnd.randint(4, bh - 4)
if rnd.random() < 0.5:
d.rectangle([wx, wy, wx + 6, wy + 8], fill=(255, 200, 120, 220))
d.rectangle([wx, wy, wx + 6, wy + 8],
fill=(255, 200, 120, 220))
x += bw + rnd.randint(10, 46)
d.rectangle([0, h - 6, w, h], fill=(12, 13, 22, 255))
save(img, "skyline.png")
def vignette():
size = 512
img = Image.new("L", (size, size), 0)
@@ -159,37 +248,51 @@ def vignette():
px[x, y] = int(255 * (t * t))
save(img, "vignette.png")
def icon(name, draw_fn, size=64):
img = Image.new("RGBA", (size, size), (0, 0, 0, 0))
d = ImageDraw.Draw(img)
draw_fn(d, size)
save(img, name)
def icons():
def flashlight_on(d, s):
d.rounded_rectangle([s*0.30, s*0.30, s*0.62, s*0.62], radius=4, fill=(255,255,255,255))
d.polygon([(s*0.62, s*0.30), (s*0.82, s*0.18), (s*0.82, s*0.74), (s*0.62, s*0.62)], fill=(255,214,120,255))
d.rounded_rectangle([s*0.30, s*0.30, s*0.62, s*0.62],
radius=4, fill=(255, 255, 255, 255))
d.polygon([(s*0.62, s*0.30), (s*0.82, s*0.18), (s*0.82, s *
0.74), (s*0.62, s*0.62)], fill=(255, 214, 120, 255))
def door_locked(d, s):
d.rectangle([s*0.18, s*0.18, s*0.82, s*0.82], outline=(255,255,255,255), width=5)
d.rounded_rectangle([s*0.36, s*0.34, s*0.64, s*0.66], radius=3, fill=(255,255,255,255))
d.arc([s*0.40, s*0.28, s*0.60, s*0.50], 180, 360, fill=(255,255,255,255), width=5)
d.rectangle([s*0.18, s*0.18, s*0.82, s*0.82],
outline=(255, 255, 255, 255), width=5)
d.rounded_rectangle([s*0.36, s*0.34, s*0.64, s*0.66],
radius=3, fill=(255, 255, 255, 255))
d.arc([s*0.40, s*0.28, s*0.60, s*0.50], 180,
360, fill=(255, 255, 255, 255), width=5)
def wand(d, s):
d.rounded_rectangle([s*0.30, s*0.10, s*0.44, s*0.90], radius=3, fill=(255,255,255,255))
d.ellipse([s*0.24, s*0.06, s*0.50, s*0.32], fill=(120,220,160,255))
d.rounded_rectangle([s*0.30, s*0.10, s*0.44, s*0.90],
radius=3, fill=(255, 255, 255, 255))
d.ellipse([s*0.24, s*0.06, s*0.50, s*0.32], fill=(120, 220, 160, 255))
def rec(d, s):
d.ellipse([s*0.14, s*0.14, s*0.86, s*0.86], outline=(255,255,255,255), width=5)
d.ellipse([s*0.14, s*0.14, s*0.86, s*0.86],
outline=(255, 255, 255, 255), width=5)
d.ellipse([s*0.30, s*0.30, s*0.70, s*0.70], fill=(230, 60, 60, 255))
icon("icons/flashlight_on.png", flashlight_on)
icon("icons/door_locked.png", door_locked)
icon("icons/wand.png", wand)
icon("icons/rec_dot.png", rec)
def app_icon():
"""Game app icon: dark badge, gold star, white 'S' — used by desktop/AppImage
(PNG) and the Windows installer (ICO)."""
def badge(d, s):
d.ellipse([2, 2, s-2, s-2], fill=(24, 26, 38, 255))
d.ellipse([int(s*0.10), int(s*0.10), int(s*0.90), int(s*0.90)], outline=(120, 130, 170, 255), width=max(2, s//64))
d.ellipse([int(s*0.10), int(s*0.10), int(s*0.90), int(s*0.90)],
outline=(120, 130, 170, 255), width=max(2, s//64))
# gold star
cx, cy, r = s*0.5, s*0.44, s*0.30
pts = []
@@ -197,7 +300,8 @@ def app_icon():
ang = -90 + i * 36
rad = r if i % 2 == 0 else r * 0.42
import math
pts.append((cx + rad * math.cos(math.radians(ang)), cy + rad * math.sin(math.radians(ang))))
pts.append((cx + rad * math.cos(math.radians(ang)),
cy + rad * math.sin(math.radians(ang))))
d.polygon(pts, fill=(255, 205, 80, 255))
icon("icon.png", badge, size=1024)
icon("icon.ico", badge, size=256)
@@ -205,29 +309,33 @@ def app_icon():
# multi-size ICO for Windows (from the 256 render)
src = Image.open(os.path.join(ASSETS, "icon.png"))
ico_path = os.path.join(ASSETS, "icon.ico")
src.save(ico_path, format="ICO", sizes=[(256, 256), (128, 128), (64, 64), (48, 48), (32, 32), (16, 16)])
src.save(ico_path, format="ICO", sizes=[
(256, 256), (128, 128), (64, 64), (48, 48), (32, 32), (16, 16)])
print("wrote assets/textures/icon.ico (multi-size)")
def cards():
CW, CH = 256, 128
def base(card_name, bg, label, band=(0,0,0,0)):
def base(card_name, bg, label, band=(0, 0, 0, 0)):
img = Image.new("RGBA", (CW, CH), bg + (255,))
d = ImageDraw.Draw(img)
d.rounded_rectangle([6, 6, CW-6, CH-6], radius=10, outline=(0,0,0,90), width=2)
d.rounded_rectangle([6, 6, CW-6, CH-6], radius=10,
outline=(0, 0, 0, 90), width=2)
if band and band[3] == 255:
d.rectangle([0, CH-34, CW, CH], fill=band)
f = font(26)
d.text((14, 10), label, font=f, fill=(255,255,255,235))
d.text((14, 10), label, font=f, fill=(255, 255, 255, 235))
# body text lines
f2 = font(15)
y = 52
for _ in range(4):
w = 150 - (24 if _ == 3 else 0)
d.rectangle([16, y, 16 + w, y + 8], fill=(255,255,255,110))
d.rectangle([16, y, 16 + w, y + 8], fill=(255, 255, 255, 110))
y += 24
# photo square for id-like cards
if "photo" in card_name:
d.rectangle([CW-84, 14, CW-18, 96], fill=(180,190,205,255))
d.rectangle([CW-84, 14, CW-18, 96], fill=(180, 190, 205, 255))
save(img, f"cards/{card_name}.png")
base("id_card", (46, 92, 168), "CA DRIVER LICENSE", band=(20, 36, 70, 255), )
base("wristband", (196, 62, 78), "EVENT ACCESS", band=(150, 35, 50, 255))
@@ -235,7 +343,132 @@ def cards():
base("credential", (58, 62, 74), "STAFF PASS", band=(20, 22, 28, 255))
base("guest_list", (200, 190, 168), "GUEST LIST", band=(120, 90, 60, 255))
def _api_key() -> str:
key = os.environ.get("OPENROUTER_API_KEY", "").strip()
if not key:
sys.exit("error: OPENROUTER_API_KEY is not set; export it before using --ai")
return key
def _generate_image(prompt: str, model: str, size: int,
transparent: bool = False) -> Image.Image:
"""Call the OpenRouter image API and return the decoded PIL image.
Requests a square image at `size` px and decodes the base64 PNG/JPEG/WebP
bytes returned in `data[].b64_json`. The requested size is clamped to at
least GEN_SIZE because some models (e.g. seedream) reject small outputs;
callers downscale the result to MAX_DIM afterwards. When `transparent` is
true, the API is asked for a transparent background (PNG alpha).
"""
size = max(GEN_SIZE, int(size))
body = {
"model": model,
"prompt": prompt,
"size": f"{size}x{size}",
"output_format": "png",
"n": 1,
}
if transparent:
body["background"] = "transparent"
req = urllib.request.Request(
OPENROUTER_IMAGES_URL,
data=json.dumps(body).encode("utf-8"),
headers={
"Authorization": f"Bearer {_api_key()}",
"Content-Type": "application/json",
"HTTP-Referer": "https://git.allucanget.biz/zwitschi/security-game",
"X-Title": "Step Line",
},
method="POST",
)
try:
with urllib.request.urlopen(req, timeout=300) as resp:
payload = json.loads(resp.read().decode("utf-8"))
except urllib.error.HTTPError as exc:
detail = exc.read().decode("utf-8", "replace")
sys.exit(f"error: OpenRouter returned HTTP {exc.code}: {detail}")
except urllib.error.URLError as exc:
sys.exit(f"error: could not reach OpenRouter: {exc.reason}")
data = payload.get("data") or []
if not data or not data[0].get("b64_json"):
sys.exit(f"error: OpenRouter returned no image data: {payload}")
raw = base64.b64decode(data[0]["b64_json"])
try:
img = Image.open(io.BytesIO(raw))
img.load()
except Exception as exc: # noqa: BLE001 - surface any decode failure clearly
sys.exit(f"error: could not decode returned image: {exc}")
return img
def _fit(img: Image.Image, max_dim: int = MAX_DIM) -> Image.Image:
"""Downscale to fit within max_dim on the longest side, preserving aspect."""
w, h = img.size
longest = max(w, h)
if longest <= max_dim:
return img
scale = max_dim / longest
return img.resize((max(1, round(w * scale)), max(1, round(h * scale))),
Image.Resampling.LANCZOS)
def generate_ai_textures(prompt: str, name: str, model: str, size: int,
transparent: bool = False) -> None:
"""Generate one texture via OpenRouter and save it under assets/textures/."""
img = _generate_image(prompt, model, size, transparent)
img = _fit(img)
if transparent:
img = img.convert("RGBA")
elif img.mode in ("RGBA", "LA") or (img.mode == "P" and "transparency" in img.info):
img = img.convert("RGBA")
else:
img = img.convert("RGB")
save(img, name)
def _run_ai(args: argparse.Namespace) -> None:
if args.prompt:
if not args.name:
sys.exit("error: --name is required when --prompt is given")
generate_ai_textures(args.prompt, args.name, args.model, args.size)
return
for name, prompt, transparent in AI_TEXTURES:
print(f"generating {name} ...")
generate_ai_textures(prompt, name, args.model, args.size, transparent)
def _build_arg_parser() -> argparse.ArgumentParser:
parser = argparse.ArgumentParser(
description="Generate Step Line textures (authored or via OpenRouter).")
parser.add_argument(
"--ai", action="store_true",
help="generate textures through the OpenRouter image API "
"(requires OPENROUTER_API_KEY)")
parser.add_argument(
"--prompt", default="",
help="single prompt for --ai (overrides the default batch)")
parser.add_argument(
"--name", default="",
help="output filename under assets/textures/ for a single --prompt")
parser.add_argument(
"--model", default=DEFAULT_IMAGE_MODEL,
help=f"image model slug (default: {DEFAULT_IMAGE_MODEL})")
parser.add_argument(
"--size", type=int, default=GEN_SIZE,
help=f"square size requested from the API in px (min {GEN_SIZE}); "
f"output is downscaled to <= {MAX_DIM} (default: {GEN_SIZE})")
return parser
if __name__ == "__main__":
args = _build_arg_parser().parse_args()
if args.ai:
_run_ai(args)
print("done")
sys.exit(0)
brick_facade()
seamless_concrete()
asphalt()