animated-infographic

Animated Infographic

When to Use

• Create GIFs and animated diagrams from JSON / CSV / benchmark results.
• Show card movement, rankings, comparison bars, and state transitions through animation.
• Create articles / social media assets where numerical accuracy and the distinction between presentation time and measured time matter.

Out of scope: infographics made only of static PNGs, analysis of existing videos, editable structural diagrams.

Inputs

• Source artifact: JSON / CSV / text. If there are measured values, use these as the SSOT.
• Audience language and target medium: Japanese blog, X, slides, etc.
• Timing mode:
  • illustrative: Motion is presentation. It does not represent processing speed.
  • measured: Build frame timing from measured timestamps.
  • mixed: Numbers are measured; card and bar movement is presentation.
• Output size. Default is for blog/X1200x675, and create a static fallback if needed.

Workflow

1. Freeze the evidence

   • Fix the source artifact, acquisition date and time, target count, and measurement boundary.
   • latencyは1 request、serial total、fresh-process end to endetc., decide what was measured.
   • When processing external media, record the original URL, acquisition date, source-relative path, and SHA-256 in the manifest, and verify the hash before generation.

2. Design the story

   • Limit 1 GIF to 1 claim. Move long explanations back to the body text.
   • Fix the reading order, such as input → decision → next process, before → after, comparison → conclusion.
   • If a narrow decision gate is the main subject, make it the protagonist of the visual hierarchy using one of the following: accent color, thick outline, center placement, or callout. Place input and output on both sides, and write the value determined by the gate inside the box.
   • Do not assert official terminology and the author's own layers/classifications with the same appearance. For original models, within the assetこの記事での整理indicate the scope, such as, and avoid arrows or nesting that make them look like product inclusion or maturity.

3. Localize before rendering

   • Match explanatory UI, status, and captions to the reader's language. Do not leave generic English UI in Japanese articles.
   • Preserve official notation for proper nouns such as product names, API fields, and model IDs.
   • Avoid literal translation style. Example: 段階で採点rather than定義した基準のどの段階に近いか。

4. Make timing explicit inside the asset

   • illustrative: カード移動は演出（実測速度ではありません）display.
   • measured: Display the measurement boundary and units, and build frame duration from measured values.
   • mixed: 表示値は実測 / 棒・カードの動きは演出write both.

5. Generate reproducibly

   • Keep the generator script, GIF, static fallback, and, if necessary, source snapshot.
   • Do not manually enter numbers; read them from the source artifact.
   • Limit third-party videos to the minimum frames necessary for the claim (e.g., before / operation / after), and remove unrelated browser chrome. Within the asset, display the source and information to prevent misreading, such as dummy data, unsent, and author-added notes, in a distinguishable way.

6. Validate mechanically

   python scripts/helper.py output.gif `
     --expected-width 1200 --expected-height 675 `
     --timing-mode illustrative `
     --contact-sheet output-contact.jpg `
     --report output-validation.json
   

   • --timing-mode measured also requires --measured-ms <total>; --duration-tolerance defaults to 0.1. Pillow is required.

7. Inspect visually

   • Check the start, middle, and end frames at original size and target embed width (about 360px for mobile).
   • Fix text overlap, clipping, order, whitespace, unreadable annotations, and remaining English UI, then regenerate.

Timing Rules

• The total playback time of a GIF does not necessarily match API latency. Do not automatically equate them.
• For the sum of multiple requests, indicate whether they are parallel or serial.
• If the measurement boundaries of the comparison targets differ,model speedrather thanobserved end-to-end timetreat as.
• Also verify the estimated cost, accuracy, and token count against the source artifact, and add a note so they are not misread from the asset alone.

Done Criteria

• The GIF is animated and meets the expected dimensions and minimum frame count.
• The timing mode and measurement boundaries are visible within the asset.
• The values in the source artifact match the displayed values.
• If derived from external media, the source URL / retrieval date / SHA-256 match the manifest.
• The explanatory UI is unified in the reader's language.
• The subject's decision gate is visible first, and it is readable what it receives and what it returns.
• Official definitions and the author's own organization can be distinguished from the asset alone.
• The start, middle, and end frames and loop playback have been confirmed.
• The text is readable at the target embed width.
• The locations of the generator, GIF, static fallback, and validation report have been reported.