YT-AI-Automation / backend /src /core /thumbnail_template_renderer.py
github-actions
Sync Docker Space
5f3e9f5
"""Server-side renderer for the JSON thumbnail-template format.
This module is the Pillow counterpart to the React/Canvas renderer in
``frontend/src/lib/thumbnailBuilder.ts``. The frontend serialises the
editable thumbnail as a flat element tree:
{
"canvasWidth": 1920,
"canvasHeight": 1080,
"canvasBackground": "#4caf50",
"elements": {
"title": { "type": "title", "posX": ..., "text": "...", ... },
...
}
}
That same payload can be POSTed to ``/render-thumbnail-template`` and the
backend will produce a PNG that matches the editor preview pixel-for-pixel
(within the limits of font availability on the host).
The renderer is intentionally tolerant: missing keys fall back to sensible
defaults so the endpoint can be used as a public template renderer too,
not just as an internal parity check for the Flask pipeline.
"""
from __future__ import annotations
import io
import ipaddress
import math
import os
import re
import socket
from dataclasses import dataclass
from typing import Iterable
from urllib.parse import urlparse
import requests
from PIL import Image, ImageDraw, ImageFont, ImageOps
_IMAGE_TIMEOUT_SECONDS = 8
_MAX_IMAGE_BYTES = 12 * 1024 * 1024
_DEVANAGARI_RE = re.compile(r"[\u0900-\u097F]")
@dataclass(frozen=True)
class _Element:
"""Normalised view of a single template element.
The dict-based JSON payload is converted to this typed view once so the
renderer can stay focused on geometry/drawing instead of dict lookups.
"""
id: str
type: str
pos_x: float
pos_y: float
width: float
height: float
z_index: int
rotation: float
opacity: float
visible: bool
text: str
font_size: float
font_weight: str
font_family: str
text_align: str
color: str
background_color: str
border_color: str
border_width: float
border_radius: float
padding_x: float
padding_y: float
image_url: str
image_offset_x: float
image_offset_y: float
image_zoom: float
image_fit_mode: str
image_overlay: float
shape_type: str
def render_template_png(payload: dict, pixel_ratio: float = 1.0) -> bytes:
"""Render the JSON thumbnail template into a PNG byte string."""
width = max(64, int(payload.get("canvasWidth") or 1920))
height = max(64, int(payload.get("canvasHeight") or 1080))
background = _color(payload.get("canvasBackground"), "#ffffff")
pixel_ratio = max(0.25, min(4.0, float(pixel_ratio or 1.0)))
out_w = int(round(width * pixel_ratio))
out_h = int(round(height * pixel_ratio))
image = Image.new("RGBA", (out_w, out_h), background or "#ffffff")
draw = ImageDraw.Draw(image, "RGBA")
raw_elements = payload.get("elements") or {}
if not isinstance(raw_elements, dict):
raise ValueError("elements must be an object keyed by id")
elements = [_normalise(eid, e) for eid, e in raw_elements.items() if isinstance(e, dict)]
elements = [e for e in elements if e.visible]
elements.sort(key=lambda e: e.z_index)
for el in elements:
_draw_element(image, draw, el, pixel_ratio)
out = io.BytesIO()
image.convert("RGB").save(out, format="PNG", optimize=True)
return out.getvalue()
# ──────────────────────────── element normalisation ──────────────────────────
def _normalise(eid: str, raw: dict) -> _Element:
return _Element(
id=str(raw.get("id") or eid),
type=str(raw.get("type") or "title"),
pos_x=float(raw.get("posX") or 0),
pos_y=float(raw.get("posY") or 0),
width=float(raw.get("width") or 0),
height=float(raw.get("height") or 0),
z_index=int(raw.get("zIndex") or 5),
rotation=float(raw.get("rotation") or 0),
opacity=float(raw.get("opacity") if raw.get("opacity") is not None else 100),
visible=raw.get("visible") is not False,
text=str(raw.get("text") or ""),
font_size=float(raw.get("fontSize") or 0),
font_weight=str(raw.get("fontWeight") or "400"),
font_family=str(raw.get("fontFamily") or ""),
text_align=str(raw.get("textAlign") or "left"),
color=str(raw.get("color") or "#000000"),
background_color=str(raw.get("backgroundColor") or "transparent"),
border_color=str(raw.get("borderColor") or "transparent"),
border_width=float(raw.get("borderWidth") or 0),
border_radius=float(raw.get("borderRadius") or 0),
padding_x=float(raw.get("paddingX") or 0),
padding_y=float(raw.get("paddingY") or 0),
image_url=str(raw.get("imageUrl") or ""),
image_offset_x=float(
raw.get("imageOffsetX") if raw.get("imageOffsetX") is not None else 50
),
image_offset_y=float(
raw.get("imageOffsetY") if raw.get("imageOffsetY") is not None else 50
),
image_zoom=float(
raw.get("imageZoom") if raw.get("imageZoom") is not None else 100
),
image_fit_mode=str(raw.get("imageFitMode") or "cover"),
image_overlay=float(raw.get("imageOverlay") or 0),
shape_type=str(raw.get("shapeType") or "rectangle"),
)
# ──────────────────────────── drawing dispatch ───────────────────────────────
def _draw_element(
image: Image.Image, draw: ImageDraw.ImageDraw, el: _Element, ratio: float
) -> None:
if el.type == "image":
_draw_image_element(image, el, ratio)
return
if el.type in ("panel", "shape"):
_draw_shape(draw, el, ratio)
return
if el.type == "badge":
_draw_badge(image, draw, el, ratio)
return
_draw_text_box(image, draw, el, ratio)
def _draw_shape(draw: ImageDraw.ImageDraw, el: _Element, ratio: float) -> None:
x0, y0, x1, y1 = _scaled_box(el, ratio)
fill = _color(el.background_color, None)
border = _color(el.border_color, None)
bw = max(0, int(round(el.border_width * ratio)))
radius = int(round(el.border_radius * ratio))
if el.shape_type == "circle":
draw.ellipse((x0, y0, x1, y1), fill=fill, outline=border, width=bw or 0)
elif el.shape_type == "pill":
r = max(1, min((x1 - x0), (y1 - y0)) // 2)
draw.rounded_rectangle((x0, y0, x1, y1), radius=r, fill=fill, outline=border, width=bw)
elif el.shape_type == "line":
if border:
mid = (y0 + y1) // 2
draw.line((x0, mid, x1, mid), fill=border, width=max(2, bw))
else:
if radius > 0:
draw.rounded_rectangle((x0, y0, x1, y1), radius=radius, fill=fill, outline=border, width=bw)
else:
draw.rectangle((x0, y0, x1, y1), fill=fill, outline=border, width=bw)
def _draw_text_box(
image: Image.Image, draw: ImageDraw.ImageDraw, el: _Element, ratio: float
) -> None:
x0, y0, x1, y1 = _scaled_box(el, ratio)
bw = max(0, int(round(el.border_width * ratio)))
radius = int(round(el.border_radius * ratio))
fill = _color(el.background_color, None)
border = _color(el.border_color, None)
if fill or border:
if radius > 0:
draw.rounded_rectangle((x0, y0, x1, y1), radius=radius, fill=fill, outline=border, width=bw)
else:
draw.rectangle((x0, y0, x1, y1), fill=fill, outline=border, width=bw)
if not el.text:
return
needs_devanagari = _DEVANAGARI_RE.search(el.text) is not None
pad_x = int(round(el.padding_x * ratio))
pad_y = int(round(el.padding_y * ratio))
inner_w = max(1, x1 - x0 - 2 * pad_x)
inner_h = max(1, y1 - y0 - 2 * pad_y)
lines = el.text.split("\n")
font_size = max(8, int(round(el.font_size * ratio)))
min_size = max(8, int(round(font_size * 0.4)))
bold = _is_bold(el.font_weight)
# Auto-shrink: same heuristic as the canvas renderer β€” drop 2 px at a
# time until the widest line + total height fits the inner box.
while font_size > min_size:
font = _font(font_size, bold=bold, prefer_devanagari=needs_devanagari)
widest = max((_measure(draw, line, font)[0] for line in lines), default=0)
line_height = font_size * 1.08
total_height = line_height * len(lines)
if widest <= inner_w and total_height <= inner_h:
break
font_size -= 2
font = _font(font_size, bold=bold, prefer_devanagari=needs_devanagari)
line_height = font_size * 1.08
total_height = line_height * len(lines)
text_color = _color(el.color, "#000000") or "#000000"
align = el.text_align if el.text_align in ("left", "center", "right") else "left"
cy = y0 + pad_y + max(0, (inner_h - total_height) / 2)
for index, line in enumerate(lines):
line_w, _ = _measure(draw, line, font)
if align == "right":
tx = x1 - pad_x - line_w
elif align == "center":
tx = x0 + pad_x + (inner_w - line_w) / 2
else:
tx = x0 + pad_x
ty = cy + index * line_height
draw.text((tx, ty), line, font=font, fill=text_color)
def _draw_image_element(image: Image.Image, el: _Element, ratio: float) -> None:
x0, y0, x1, y1 = _scaled_box(el, ratio)
box_w = max(1, x1 - x0)
box_h = max(1, y1 - y0)
radius = int(round(el.border_radius * ratio))
photo: Image.Image | None = _fetch_image(el.image_url) if el.image_url else None
if photo is None:
photo = _placeholder_image(box_w, box_h)
else:
photo = photo.convert("RGB")
zoom = max(0.1, el.image_zoom / 100.0)
# Same fit-mode logic as the canvas renderer:
# * cover (default) β€” scale so the box is fully covered, may crop
# * contain β€” scale so the whole image fits, may letterbox
# * stretch β€” ignore aspect ratio, fill the box exactly
if el.image_fit_mode == "stretch":
new_w = max(1, int(round(box_w * zoom)))
new_h = max(1, int(round(box_h * zoom)))
elif el.image_fit_mode == "contain":
scale = min(box_w / photo.width, box_h / photo.height) * zoom
new_w = max(1, int(round(photo.width * scale)))
new_h = max(1, int(round(photo.height * scale)))
else:
scale = max(box_w / photo.width, box_h / photo.height) * zoom
new_w = max(1, int(round(photo.width * scale)))
new_h = max(1, int(round(photo.height * scale)))
photo = photo.resize((new_w, new_h), Image.Resampling.LANCZOS)
offset_x = int((el.image_offset_x / 100.0) * (box_w - new_w))
offset_y = int((el.image_offset_y / 100.0) * (box_h - new_h))
canvas = Image.new("RGB", (box_w, box_h), _color(el.background_color, "#111111") or "#111111")
canvas.paste(photo, (offset_x, offset_y))
photo = canvas
# Round corners using a mask, then drop the result onto the canvas.
mask = Image.new("L", (box_w, box_h), 0)
mdraw = ImageDraw.Draw(mask)
if radius > 0:
mdraw.rounded_rectangle((0, 0, box_w, box_h), radius=radius, fill=255)
else:
mdraw.rectangle((0, 0, box_w, box_h), fill=255)
image.paste(photo, (x0, y0), mask)
if el.image_overlay > 0:
overlay = Image.new("RGBA", (box_w, box_h), (15, 23, 42, 0))
odraw = ImageDraw.Draw(overlay)
steps = 12
for i in range(steps):
top_alpha = (el.image_overlay / 100.0) * 0.2
bot_alpha = (el.image_overlay / 100.0) * 0.85
t = i / max(1, steps - 1)
alpha = int(255 * (top_alpha + (bot_alpha - top_alpha) * t))
y = int(box_h * t)
y_next = int(box_h * ((i + 1) / steps))
odraw.rectangle((0, y, box_w, y_next), fill=(15, 23, 42, alpha))
if radius > 0:
overlay_mask = Image.new("L", (box_w, box_h), 0)
ImageDraw.Draw(overlay_mask).rounded_rectangle(
(0, 0, box_w, box_h), radius=radius, fill=255
)
image.paste(overlay, (x0, y0), overlay_mask)
else:
image.paste(overlay, (x0, y0), overlay)
def _draw_badge(
image: Image.Image, draw: ImageDraw.ImageDraw, el: _Element, ratio: float
) -> None:
x0, y0, x1, y1 = _scaled_box(el, ratio)
cx = (x0 + x1) / 2
cy = (y0 + y1) / 2
w = max(1, x1 - x0)
h = max(1, y1 - y0)
radius = max(0, int(round(el.border_radius * ratio)))
fill = _color(el.background_color, "#ffee00") or "#ffee00"
if radius >= min(w, h) // 2 and radius > 0:
draw.ellipse((x0, y0, x1, y1), fill=fill)
elif radius > 0:
draw.rounded_rectangle((x0, y0, x1, y1), radius=radius, fill=fill)
else:
# Starburst β€” same 14-point shape as the canvas renderer.
outer_r = min(w, h) / 2
inner_r = outer_r * 0.78
points = []
for i in range(28):
angle = math.pi * i / 14 - math.pi / 2
r = outer_r if i % 2 == 0 else inner_r
points.append((cx + r * math.cos(angle), cy + r * math.sin(angle)))
draw.polygon(points, fill=fill)
if not el.text:
return
needs_devanagari = _DEVANAGARI_RE.search(el.text) is not None
bold = _is_bold(el.font_weight)
font_size = max(8, int(round(el.font_size * ratio)))
font = _font(font_size, bold=bold, prefer_devanagari=needs_devanagari)
text_color = _color(el.color, "#0f172a") or "#0f172a"
lines = el.text.split("\n")
line_h = font_size * 1.1
total_h = line_h * len(lines)
for index, line in enumerate(lines):
lw, _ = _measure(draw, line, font)
tx = cx - lw / 2
ty = cy - total_h / 2 + index * line_h
draw.text((tx, ty), line, font=font, fill=text_color)
# ──────────────────────────── colour / font helpers ──────────────────────────
def _scaled_box(el: _Element, ratio: float) -> tuple[int, int, int, int]:
x0 = int(round(el.pos_x * ratio))
y0 = int(round(el.pos_y * ratio))
x1 = int(round((el.pos_x + el.width) * ratio))
y1 = int(round((el.pos_y + el.height) * ratio))
return x0, y0, x1, y1
def _color(value: str | None, fallback: str | None) -> str | None:
"""Normalise a CSS-ish colour string into something Pillow understands."""
if not value:
return fallback
s = value.strip()
if not s or s.lower() == "transparent":
return None
if s.startswith("#"):
return s
if s.startswith("rgb("):
match = re.match(r"rgb\(\s*(\d+)\s*,\s*(\d+)\s*,\s*(\d+)\s*\)", s)
if match:
r, g, b = (int(v) for v in match.groups())
return f"#{r:02x}{g:02x}{b:02x}"
if s.startswith("rgba("):
match = re.match(
r"rgba\(\s*(\d+)\s*,\s*(\d+)\s*,\s*(\d+)\s*,\s*([0-9.]+)\s*\)", s
)
if match:
r, g, b = (int(v) for v in match.groups()[:3])
return f"#{r:02x}{g:02x}{b:02x}"
# Named colours β€” let Pillow validate them.
return s
def _is_bold(weight: str) -> bool:
try:
return int(weight) >= 600
except (TypeError, ValueError):
return weight.lower() in {"bold", "bolder", "black"}
def _measure(draw: ImageDraw.ImageDraw, text: str, font) -> tuple[int, int]:
if not text:
return (0, 0)
bbox = draw.textbbox((0, 0), text, font=font)
return (bbox[2] - bbox[0], bbox[3] - bbox[1])
def _font(size: int, *, bold: bool, prefer_devanagari: bool):
"""Pick the best installed font for the given script + weight."""
devanagari_candidates = (
os.environ.get("THUMBNAIL_FONT_DEVANAGARI") or "",
"/usr/share/fonts/truetype/noto/NotoSansDevanagari-Bold.ttf"
if bold else "/usr/share/fonts/truetype/noto/NotoSansDevanagari-Regular.ttf",
"/usr/share/fonts/truetype/Tiro_Devanagari_Sanskrit/TiroDevanagariSanskrit-Regular.ttf",
r"C:\Windows\Fonts\mangalb.ttf" if bold else r"C:\Windows\Fonts\mangal.ttf",
r"C:\Windows\Fonts\Nirmala.ttc",
)
latin_candidates: Iterable[str] = (
os.environ.get("THUMBNAIL_FONT_PATH") or "",
"/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf"
if bold else "/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf",
"/usr/share/fonts/truetype/liberation/LiberationSans-Bold.ttf"
if bold else "/usr/share/fonts/truetype/liberation/LiberationSans-Regular.ttf",
)
sources = (
list(devanagari_candidates) + list(latin_candidates)
if prefer_devanagari
else list(latin_candidates) + list(devanagari_candidates)
)
for path in sources:
if path and os.path.isfile(path):
try:
return ImageFont.truetype(path, size=size)
except OSError:
continue
return ImageFont.load_default()
# ──────────────────────────── image fetch helpers ────────────────────────────
def _fetch_image(url: str) -> Image.Image | None:
"""Best-effort, SSRF-safe HTTP image download."""
parsed = urlparse((url or "").strip())
if parsed.scheme not in {"http", "https"} or not parsed.netloc:
return None
if _is_private_host(parsed.hostname or ""):
return None
try:
response = requests.get(url, timeout=_IMAGE_TIMEOUT_SECONDS, stream=True)
response.raise_for_status()
content_type = response.headers.get("content-type", "")
if content_type and not content_type.lower().startswith("image/"):
return None
data = bytearray()
for chunk in response.iter_content(64 * 1024):
data.extend(chunk)
if len(data) > _MAX_IMAGE_BYTES:
return None
return Image.open(io.BytesIO(data))
except Exception:
return None
def _is_private_host(hostname: str) -> bool:
try:
addresses = socket.getaddrinfo(hostname, None, type=socket.SOCK_STREAM)
except OSError:
return True
for addr in addresses:
ip = ipaddress.ip_address(addr[4][0])
if ip.is_private or ip.is_loopback or ip.is_link_local or ip.is_multicast:
return True
return False
def _placeholder_image(width: int, height: int) -> Image.Image:
"""Plain neutral placeholder used when the side image URL is missing."""
image = Image.new("RGB", (width, height), "#cbd5e1")
draw = ImageDraw.Draw(image)
draw.rectangle((0, 0, width, height), fill="#cbd5e1")
inset = max(8, min(width, height) // 12)
draw.rounded_rectangle(
(inset, inset, width - inset, height - inset),
radius=max(4, inset // 2),
fill="#94a3b8",
)
return ImageOps.autocontrast(image)