BarryD-777/screen-throw-freeze-video

Big Screen Throw-Catch Static-Frame Pacing Video

Create a 18.77-second video using a fixed timeline. Each iteration replaces background, person, product image, and product copy while preserving the action transitions, static-frame starting point, text reveals, and page-turn rhythm. The public package includes instructions, a clock, and a reference BGM, but does not contain a video generation model, rendering engine, or the original sample footage.

Collect Four Inputs

Start by filling in from the conversation and provided materials. Do not ask again for what is already known; for missing items, combine them into a single brief question.

Input	Details to Determine
Background	Location, time of day, and atmosphere; arrange screens or billboards so the product can be thrown from the screen into reality.
Person	Specify role, yourself, a live model, or cosplayer; distinguish game CG from real cosplay costumes; do not default to an old character.
Product Image	A clear packaging image for this iteration; verify brand, product name, capacity, labels, and cap; do not substitute an old product.
Product Copy	User provides ingredient / not-detected text or specifies requirements; if no text is given, the assistant defaults to searching based on the accurate SKU; do not ask the user to write the copy for the assistant first, nor ask if searching is allowed.

When the user says "you choose," assistant may select on their behalf. Assistant search results are also valid copy input. If authorized to write based on complete information, proceed directly without a separate copy approval step. When input is missing, continue advancing parts that do not depend on it; do not submit generation with an incorrect person or product.

Copy Only Contains Product Content

• Default: left column is the product's own ingredients, right column is the "not detected" ingredients for this product; retain corresponding category titles, numbering, and original packaging text.
• Do not add "proposed copy," "copy sample," "non-test conclusion," debugging text, explanatory bars, or generic efficacy slogans; do not pad rows with specifications or packaging features. Opening text is added only when explicitly requested by the user.
• Verify against the current SKU per the Copy Research reference. An ingredient not listed on the label does not mean "not detected"; "not added," "does not contain," and "not detected" are not interchangeable. Assistant-verified "not detected" entries must have corresponding testing evidence.
• When evidence is missing, do not fabricate 150 items; do not recycle old video lists. Record the missing items in the production notes, ask the user to supplement the report, provide a clear text draft, or accept a version with fewer items, while simultaneously advancing completable visuals. Text transcribed at the user's request is recorded as user-provided, not as an independently verified assistant result.

Fixed Timeline

First read the Timeline and Text Method. This version is fixed at 1280x720, 30fps, 563 frames, ending at 18.766667 seconds.

• Frames 0–96 complete a continuous throw-and-catch; frames 97–100 are taken over by a rotation blur; from frame 101 the left title and first item enter.
• From frame 101 onward, the background remains the same static frame until the end; background position and scale are fixed; no street loops, no push/pull. Frame 103 dark areas stabilize; frame 104 the product enters with slight floating motion.
• Left 8 items start at frames 101, 106, 111, 118, 124, 130, 135, 141; right 7 pages start at frames 158, 213, 270, 329, 388, 446, 504.
• Use the normalized per-frame crop curve from assets/reveal-clock.json; do not BPM-average or replace with slow fades. For copies of different lengths, multiply new glyph width by progress, preserving the start and reveal range.
• Default to using the included assets/reference-bgm.m4a, playing from 0 seconds at original speed; do not pan or stretch. Audio is not under MIT license; see Music Material License. If the user specifies other music, recheck the drum hits; do not claim the fixed clock matches any arbitrary song.

Run scripts/make_cues.py --copy <copy.json> --output <new_directory/cues.json> to generate the text reveal plan, which automatically locates the included BGM. Use --audio <local_music> to override the default track. If the included track is missing and no alternative is specified, flag the need for music replacement. The tool does not generate music, verify copy, or render videos.

Copy is limited to at most 8 items on the left and 150 items on the right (6 pages of 22 rows each, last page 18 rows). When valid copy is insufficient, do not repeat or fabricate; after the user accepts a version with fewer items, maintain full length and original clock; empty slots remain invisible. When capacity is exceeded, ask the user to prioritize; do not silently truncate.

Visuals and Production

1. Check the current environment's existing video generation and rendering capabilities. If a Hypit workflow is already present, read that environment's Hypit skills; alternatively, use existing tools that support frame-by-frame compositing to implement the timeline. The public package does not contain the original private project or rendering components; do not claim a packaged one-click render engine.
2. When starting and ending images are needed, follow the current project's image generation preferences. If a template-assisted image generation workflow is installed, first search templates, then combine with new material. Use the tools already selected in this session; do not automatically trigger paid calls just because this skill is loaded.
3. Prepare the start and end images per Throw-Catch and Compositing: one person inside the screen, one product, one foreground catching hand; one release, one contact, one stable catch. The person does not disappear and does not grab the bottle a second time.
4. Continue using currently valid service and budget authorizations. If already covered for this task, do not reconfirm; one-time authorizations already used in the past do not apply to new videos. When new authorization is needed, first complete the start/end images, specific request, and budget to avoid automatic paid retries.
5. Use the new product's original packaging for a clear display; crop according to this iteration's outline; do not reuse old bottle masks, change capacity, or let AI write packaging text. Keep spacing between the central product and left/right copy using purple-white gradient bold font, purple 3D edge, and thin black outline.
6. For new text drafts, re-layout using real Chinese fonts. Render each page as a preloaded transparent text image with per-row joint cropping, or use reliable native text rendering; avoid encoding errors from numerous repeated image elements.
7. Save materials, copy drafts, clock, render files, and evidence in the new production directory; for same-video revisions, preserve valid materials and previous exports; do not overwrite other projects.

Video Service Credentials

Each user configures their own generation account; this skill does not share the author's account or quota. Generation tools only read API keys from the current user's local credential management or runtime environment; do not write keys into the skill, code, prompts, examples, logs, production records, or public repositories; do not ask the user to paste keys into chat. When configuration is missing, guide the user through the selected tool's local configuration method; do not substitute the author's account.

scripts/make_cues.py only processes local clock and copy; it does not require keys or call video APIs.

Verification and Delivery

• Review the encoded video: throw, flight, contact, and stable catch are continuous; frames 97–104 have no black flashes, hard cuts, double bottles, or misalignment.
• For frames 104–562, check background areas without product, text, or glow; person and vehicle positions are fixed; do not mistake loop or camera drift for encoding errors.
• Verify all text start points, reveals, page turns, and retention times; no missing rows, early text exposure, extra annotations, garbled characters, or occlusions.
• Confirm frame 563 and audio starts at 0. When the audio track does not need modification, prefer lossless encapsulation of the original AAC; if performing only data checks, do not claim to have listened to the music.
• Save material sources, copy evidence, generation authorizations, and inspection results; background previews are hidden on startup; do not modify unrelated services.
• Deliver an MP4 and a usable editing file/preview, along with replacement instructions; if not rendered, truthfully state the status.