Video Plan Page
A script in a chat message is hard to judge. You can't feel the pacing, and nobody can see which shot goes with which line. This skill builds one HTML file that plays the script as an animatic. Each caption shows for its estimated read time. Each shot shows its key frame, or a card that describes the picture still to make. Below the animatic, the reviewer reads the plan, answers a few choices, and copies a single reply back to you.
Build nothing until that reply arrives.
When to use
- A script is written and the next step costs real time: key frames, animation, voice, edit.
- Someone else signs off (a client, a channel owner, a teammate) and reads on a phone.
- You want one link that holds the whole plan, not five messages.
Skip it for a single-shot clip or a script with no picture notes.
Steps
-
Write
script.json. Required:titleandlines. Each line has awhocode (Ccaption,Nnarrator,Sthe enemy or inner voice,Xpicture only) andtext. A|splits the caption into two rows. Addseeto any line that starts a new shot: one sentence about the picture. Addlevelandbeatfor the structure bar, andhudfor an on-screen counter. Leave outtanddur. The script times each line at 2.6 words a second (at least 2.4 s, orminif you set it). -
Write
plan.json(optional). It holds everything that is not the script: header chips, level names, speaker labels,coloursfor theknowsdots, the order of work, a Gantt chart in hours, extra cards, and thedecisions. Mark one option per decisionrec: true. Put key frames underkeys("S03": "keys/s03.png"or a start time such as"12.5"). Paths are relative toplan.json. -
Build.
python3 scripts/plan_page.py example/script.json --plan example/plan.json --out plan.htmlAdd
--no-webfontsif the page must not call outside servers. Missing images print a warning, and those shots stay "to paint". -
Check it yourself. Open it at desktop and phone width (390 px). Press play once at 4×. Click a shot row, check that the animatic jumps to it, then change one choice and read the reply.
-
Send one link. Host the file anywhere static. It has no other files. Ask for the reply, and treat "approve" as all recommended choices.
-
Build from the reply only. Apply the choices to the script before you make any shot.
Rules
- Keep private or unreleased material out of a page you host in public. Images are embedded, so anyone with the link sees them.
- Each image is shrunk to 960 px JPEG (when Pillow is installed). The page refuses to build over 15 MB.
- Fonts: the page asks Google Fonts for its display, body, mono, and handwriting faces. Every
font stack ends in system fonts, so offline (or with
--no-webfonts) it still reads well. - The output is generated. Edit
script.jsonandplan.json, then rebuild. - Hook check letters (optional, on
quest.hooks[].checks): M mirrors the viewer, H leaves a half guess, S has one surprise, N uses an exact number, B fits in one breath.
Files
- scripts/plan_page.py: the builder. The docstring lists every field.
- scripts/template.html: the page; the builder fills in the data.
- scripts/test_plan_page.py:
python3 -m pytest -q skills/video/video-plan-page/scripts - example/script.json and example/plan.json: a made-up 45-second film, Seven Days to One Loaf. Its 11 key frames and 2 thumbnails are AI-generated stills in example/keys/ (960 px JPEG), so a fresh clone builds the page with pictures.
Checklist
- Every shot line has a
seesentence - Each decision has exactly one recommended option
- Page opened at desktop and 390 px; play, scrub, tabs and copy work
- No private names, emails or account IDs in the inputs or images
- Nothing built until the reply came back
Pro: from approved plan to Shorts. When the long cut is done, the Shorts Builder Kit turns it into 9:16 Shorts with QA included. The free shorts-reframe skill covers the crop on its own.
Inspired by
Dani Acosta's storytelling-video plan page in a private channel repo (yt-daniacostaai, pull request #204): a single page that plays the script as an animatic, lists the shots, and ends with choices and a copy-ready approval reply. That version was built for one channel. It named that channel's characters in code, its example film pointed at image files on one machine, a missing image stopped the build, and the page had no doctype or viewport, so phones got a scaled-down desktop layout.
What this version adds
- A real HTML document (doctype, charset, viewport), so the phone layout works, and the "to paint" note no longer runs under the caption at 390 px.
- Hidden tabs and cards now actually hide (
[hidden]wins over the grid layout). - No character names in code: dot colours come from
plan.jsonor a neutral palette. - A missing image warns and leaves the shot "to paint". The build continues.
- System-font fallbacks for every face, plus
--no-webfonts. - UTF-8 reads and writes, a byte-accurate size limit, safer escaping of script text, and tolerance for hooks without checks and lines without a level.
- The approval reply names every picked option by its label (recommended ones marked), not by a separate reply code, so the reply always matches what was ticked.
- A neutral demo that builds from a fresh clone, and tests that cover each fix.