CreatorMotion · Prompted Result Inspection

Prepare sets with prompt text and two resolvable image keys. Supply two modeLabels that accurately identify the approaches. The bundled reading-corner/workstation examples illustrate the composition and do not establish a model benchmark.

Choose preferredSide as 0 or 1 for each set, or use the top-level fallback. The highlight follows that input without analyzing either image. Write or narrate the reason for selection so the outline does not stand in for an argument.

Replace reading-a, reading-b, workstation and mini-server through --images or use other valid keys. If adding --media, supply a continuous local take covering the helper’s 6.8-second visible window; it starts at zero and fades from 6.4 seconds even when the full comparison is longer. No alpha matte is required.

Prepare Node.js 22+, FFmpeg and the documented HyperFrames CLI 0.8.30. Run node scripts/create.mjs --out PROJECT --data CONTENT.json with the chosen --style from the extracted package. Map local image keys through --images; relative paths resolve beside that JSON and unlisted keys retain defaults. Add semantic --overrides if needed and follow the image policy when adapting materials. Include --media only for the optional speaker inset. Each additional set extends the composition by its authored slot; review the generated duration rather than assuming it is always twelve seconds.

Lint, validate and inspect the generated HyperFrames project, then render at 1920×1080 and 30 fps with the pinned CLI. Inspect actual MP4 frames at entry, every change and final hold for image fit, captions, seams and text bounds. Keep at least 1.2 seconds for final reading when adjusting timing and check all palettes with identical content after module changes. Narration, music and full-film assembly remain separate work. Inspect both images, prompt legibility, selected-side outline and synchronized set change. Check the presenter exit separately and preserve enough time to explain each choice.