gif_to_360: retry with sharp scaling when too large; add new GIF

Smooth upscaling blends pixels and dithers per-frame palettes, which
breaks GIF delta compression (7.2 MB source -> 24.9 MB). Nearest-
neighbour scaling keeps it at 7.2 MB, under the 11.5 MB flash limit.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
mangavalkandClaude Opus 5.5 committed 2026-10-24 17:29:18 +02:00
1 parent 15b4ef5cef
commit b2f6a52902
3 files changed
+9 -2

No files matched your search

+8 -2
View File
@@ -2,6 +2,7 @@
Each frame is scaled so the GIF covers 360x360, then center-cropped. Frame durations and looping are kept.
Pillow stores only the changed part of each frame, which keeps files small.
If the result is too large, the GIF is redone with sharp (nearest-neighbour) scaling, which is often much smaller.
Usage:
pip install pillow
@@ -19,7 +20,7 @@ SIZE = 360
MAX_BYTES = 11_500_000
def convert(src, dst):
def convert(src, dst, resample=Image.LANCZOS):
im = Image.open(src)
w, h = im.size
scale = max(SIZE / w, SIZE / h)
@@ -31,7 +32,7 @@ def convert(src, dst):
durations.append(frame.info.get("duration", im.info.get("duration", 100)) or 100)
rgba = frame.convert("RGBA") # fully composited frame
if (nw, nh) != (w, h):
rgba = rgba.resize((nw, nh), Image.LANCZOS)
rgba = rgba.resize((nw, nh), resample)
rgba = rgba.crop((left, top, left + SIZE, top + SIZE))
# the device shows the GIF on black: flatten transparency onto black
bg = Image.new("RGBA", rgba.size, (0, 0, 0, 255))
@@ -41,6 +42,11 @@ def convert(src, dst):
frames[0].save(dst, save_all=True, append_images=frames[1:], duration=durations,
loop=im.info.get("loop", 0), optimize=True, disposal=1)
size = os.path.getsize(dst)
if size > MAX_BYTES and resample != Image.NEAREST and (nw, nh) != (w, h):
# smooth scaling blends pixels, so frames change everywhere and the per-frame palettes get
# dithered: the GIF's delta compression stops working. Sharp scaling keeps it small.
print(f"{os.path.basename(src)}: {size / 1e6:.2f} MB with smooth scaling, retrying with sharp scaling")
return convert(src, dst, Image.NEAREST)
fps = len(durations) * 1000 / max(1, sum(durations))
note = "" if size <= MAX_BYTES else " ** TOO LARGE for the device (max 11.5 MB) **"
print(f"{os.path.basename(src)}: {w}x{h}, {len(frames)} frames, {fps:.1f} fps "