memgrafter/create-gif

Create reproducible GIFs quickly without guessing encode settings.

Why use it

• Precise clip timing (exact start/end and offset fixes)
• Better color fidelity for difficult footage (explicit BT.709 → sRGB before palette quantization)
• Predictable size/quality tuning (resolution, fps, colors, lossy level)
• Reusable presets from tested runs, not one-off trial-and-error

When to use

• You need a GIF deliverable (chat apps, docs, issue comments)
• Source is YouTube or local MP4
• You need to balance smoothness and file size intentionally
• Scene has skin tones/high contrast and default GIF conversion looks wrong

When NOT to use

• If target supports MP4/WebP (these are dramatically smaller at better quality)
• Very long clips (GIF size grows fast)
• You need perfect color preservation (GIF is limited to 256 colors)

Requirements (what must be installed)

• yt-dlp
• ffmpeg
• imagemagick (magick)
• gifsicle

Usage

Segment-first methodology (recommended)

1. Define segment in seconds first (example: start=14, end=23 or duration=9)
2. Download only the needed YouTube window
3. If timing is off, do final trim by seconds with ffmpeg (-ss, -to/-t)
4. Convert to GIF with the quality preset

Why seconds-based segmenting helps:

• Fast iteration: change start/end by 1s without rethinking timestamps
• Reproducible edits: exact same segment can be regenerated later
• Clear intent: start+duration is great when clip length is fixed
• Less waste: avoids downloading/processing full videos

1) Download time slice from YouTube

yt-dlp -f "bv*[ext=mp4]+ba[ext=m4a]/b[ext=mp4]/best" \
  --download-sections "*00:00:14-00:00:23" \
  --force-keyframes-at-cuts --merge-output-format mp4 \
  -o "yt_clip.%(ext)s" "<YOUTUBE_URL>"

1b) Alternative: start + duration (seconds)

If you know clip length, use ffmpeg on the downloaded clip:

# start at 15s, keep 8s
ffmpeg -y -ss 15 -t 8 -i yt_clip.mp4 -c copy yt_clip_segment.mp4

Or if you prefer start + end:

# start at 15s, end at 23s
ffmpeg -y -ss 15 -to 23 -i yt_clip.mp4 -c copy yt_clip_segment.mp4

2) Gold-standard preset (best compromise found)

ffmpeg -y -ss 1 -t 8 -i yt_clip.mp4 \
  -vf "fps=16,scale=520:-1:flags=lanczos,hqdn3d=1.0:1.0:3:3,colorspace=iall=bt709:ispace=bt709:itrc=bt709:iprimaries=bt709:irange=tv:space=gbr:trc=srgb:primaries=bt709:range=pc,format=rgb24,split[s0][s1];[s0]palettegen=max_colors=160:reserve_transparent=0:stats_mode=full[p];[s1][p]paletteuse=dither=sierra2_4a" \
  jack_typing_520w_16fps_srgb_160_raw.gif

magick jack_typing_520w_16fps_srgb_160_raw.gif -layers Optimize jack_typing_520w_16fps_srgb_160.gif
gifsicle -O3 --lossy=45 jack_typing_520w_16fps_srgb_160.gif -o jack_typing_520w_16fps_srgb_160_lossy45.gif

Validated ranges tested

• Width/size: 420w, 480p, 520w, 560w, 640w
• FPS: 10, 12, 16, 20
• Palette colors: 96, 112, 128, 160, 192, 224, 256
• Lossy levels: 20, 30, 35, 40, 45, 50, 80
• Dither modes: none, bayer, sierra2_4a

Output

Typical files produced:

• *_raw.gif (pre-optimization)
• *.gif (optimized)
• *_lossyXX.gif (final size/quality tradeoff)

How it works

Cut exact segment → color-space normalize → generate/apply palette → optimize frames → optional lossy compression.

Trade-off: more fidelity requires bigger GIFs; this workflow gives controlled compromises instead of random artifacts.