Code Brand Film logo

Code Brand Film

Community
nutllwhy
code-brand-film

Produce a short brand film or product promo (10-60s, typically 15s) entirely as code: one static HTML page with a paused GSAP timeline that a renderer seeks frame by frame, plus a music score synthesized in Python. Use when the user wants a 品牌宣传片, 广告片, 宣传动画, 产品短片, 发布会开场视频, brand film, promo video, product teaser, logo animation, or a generated ad spot — especially when they ask for 高级感 / 艺术感 / 电影感 / 别像模板, or when an existing video "looks like AI made it" or "too simple" and needs to be redone properly. Also use when a film already exists and needs a score, a soundtrack, background music, or 配乐, or when a code-generated video needs ghosting, contrast, font, or transition defects fixed.

Overview

Publishernutllwhy
Repositorycode-brand-film
Skill namecode-brand-film
Stars
142
Forks
19
Bundled files
36
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 36 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by nutllwhy on GitHub. Read the source before you install it.

Installation

Install the Code Brand Film AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/nutllwhy/code-brand-film/tree/main
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/nutllwhy/code-brand-film.git \
  .claude/skills/code-brand-film
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Code Brand Film in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Code Brand Film on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Code Brand Film is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Code-generated brand films

Three things decide whether a film ships. Do not skip any of them.

  1. Read all ten files in the reading table below, plus the example's src/index.tpl.html. Reading only creative.md / pipeline.md / scoring.md reproduces the rejected films described here.
  2. Search for the brand's real logo and imagery and record them in assets.md, then write shot-list.md and copy.md — all before the timeline.
  3. Run all three gates and paste their output in your delivery: scripts/film_audit.py film/index.html before rendering, node scripts/layout_audit.mjs film/index.html before rendering, scripts/frame_audit.py film/renders/silent.mp4 after. All three print 0 failed, or the film is not delivered. Lint passing means the page is valid, not that the film is good.

A film is one static HTML page inside film/, driven by a paused GSAP timeline that the renderer (HyperFrames) seeks frame by frame. Every asset is a local file. The score is a separate Python program that writes a WAV, muxed onto the silent picture at the end.

The failure this skill exists to prevent

The usual failure is not a broken film, it's an under-built one: it lints clean, has no bad frames, and still looks like a template. A real example built from an earlier version of this skill had 18 tweens, one entrance mechanic (fade + slide), only crossfades between acts, a logo redrawn as a circle, system fonts, a tagline nobody had written, and no product on screen. The two accepted films it was compared against had ~105 tweens, 9–12 mechanics, 2–4 non-crossfade transitions, and showed the product working in most acts.

So this skill sets floors you can measure and gives you code to copy. There are three gates: scripts/film_audit.py on the HTML (floors), scripts/layout_audit.mjs on the HTML in a real browser (placement), and scripts/frame_audit.py on the rendered MP4 (pixels). All must print 0 failed. A later film passed the first and last and was still rejected for placement: headline under the input bar, logo on the first character, a notification card parked on the corner mark. The layout gate exists for that.

The floors are minimums, not targets. A later film cleared the code gate at exactly 70 tweens and 8 mechanics, padded with breathing headlines, and was still an abstract ink-drop animation covering 3% of the frame with no product in it. The tween count now excludes filler, and the pixel gate exists because pixels cannot be padded. If you find yourself adding tweens to pass a number, stop: the film is missing content, not motion. Aim at the reference example, not the floor.

A third film came from an agent that read only creative.md, pipeline.md and scoring.md, never ran either gate, and "fixed the single worst frame" on a picture that was mostly empty. Metaphor: a stone thrown into a lake, ripples growing six glowing icons on a dark purple void. It linted with 0 errors and had a loudness-matched score. It also had 36 tweens, 3 mechanics, no logo, no shot list, 1.3% of the frame covered and a 4.4s still stretch: 13 failed checks. Everything it needed to avoid that was in the files it didn't open.

A fourth film followed every step, passed all three gates (109 tweens, 10 mechanics) and still lost clearly to a film of the same product that would have failed the old floors. The brand had shipped its own product animations; the losing film used three of them, held two still photos for six seconds under focus brackets and capture rings, put grain over the brand's white sweep, and changed acts with a different trick every time. The winner cut seven official clips on a stage sampled to their exact ground, drew nothing on the product, and used one transition taken from the product's hinge on every act change. When the brand ships motion, read references/footage-led.md — the audit switches to its floors automatically.

Read these — in this order

StepFileWhy
before anythingreferences/creative.mdmetaphor, five-act timing, motion rules, palettes, the AI-look traps
before placing anythingreferences/layout.mdzone map per act, hard placement rules (text never covered, logo never on text, title safe, no half-parked cards)
before anythingreferences/assets.mdwhere to search for the real logo and imagery, the five-point logo check, the assets.md manifest
before anythingreferences/quality-bar.mdthe numbers, the mechanic list, the transition list, product truth, real assets
if the brand has product videoreferences/footage-led.mdfootage-led films: harvest the motion, stage = studio ground, no texture, one signature transition from the product's shape, draw nothing on the product, localised copy
before codingreferences/shot-list.mdthe table you fill in first; worked example included
while codingreferences/recipes.mdcopyable GSAP for 27 mechanics and transitions (24–27: fold on a seam, seam open, anchor match cut, pose burst)
while codingexamples/office-agent/a complete 400-line film that clears every floor — read its timeline
scaffold / renderreferences/pipeline.mdlayout, template+build, lint warnings, macOS pitfalls, defect checklist
musicreferences/scoring.mdbeat grid from picture, energy plan, mix graph, verify without ears

Do read the example's src/index.tpl.html end to end before writing your own. Seeing how much one 15s film contains is worth more than any rule here.

Workflow

  1. Collect sources. Brief or product page: the mandatory selling line, 3–5 concrete product behaviours, required copy and CTA, red lines. Then go searching (references/assets.md): user-supplied files → press / brand kit → official site (header SVG, og:image, screenshots) → app store listing → official social posts. Bring back the current logo and ≥ 2 official images, at least one showing the real product. Run the five-point logo check (right product, current version, right variant, untouched, resolution) and write it all into assets.md with source URLs. Never redraw a logo, never take one from a logo-aggregator site. Start with scripts/harvest_page_images.py <product-page> --video — it pulls every image, SVG and product clip the page uses and makes numbered contact sheets to pick from. Harvest the country page in the audience's language too; copy comes from there. Clips found? The film is footage-led (references/footage-led.md): footage carries acts 2–4, stills only as a burst, stage colour = the clips' sampled ground. Physical product? It appears as its official photo every time it is on screen; acts 2–4 show its output through its own UI, and each feature line gets a proof shot (creative.md example C, recipes.md 20–23).
  2. Find the metaphor — one sentence with a physical verb, ideally taken from the brand's own asset or the selling line's own words (creative.md).
  3. Write shot-list.md and copy.md in the film project (next to film/; the audit fails without them). copy.md lists every on-screen line with its source; see examples/office-agent/copy.md. Shot list from the template. Every act: ≥ 5 timed events, a named camera move, a named exit transition, three depth layers, a product behaviour (except hook/lockup), and a source for each line of copy. Run the self-check at its bottom. If the user is around, show it.
  4. Scaffold. film/hyperframes.json (copy assets/hyperframes.json), film/package.json pinning hyperframes@0.8.78, src/index.tpl.html, src/build.py (copy scripts/build_template.py). Start from the example's texture layer (grain, falloff) — except in a footage-led film, where texture over the brand's studio ground is a fail. Do not add a page counter (02 / 05), a progress bar or corner labels — they make the film look like slides.
  5. Fonts. python3 src/build.py, then scripts/fetch_cjk_fonts.sh src/chars.txt film/assets/fonts for Chinese. Latin fonts: download woff2 files into film/assets/fonts/. Every @font-face uses url() to a local file; local() is not allowed.
  6. Author the timeline act by act from the shot list, pulling mechanics from recipes.md. Run npx --yes hyperframes@0.8.78 lint after each act.
  7. Gate. python3 scripts/film_audit.py film/index.html → must print 0 failed. Fix every FAIL by adding real content, not by gaming the regex. Then node scripts/layout_audit.mjs film/index.html --shots .gates/layout → must print 0 failed. Open the screenshots, move things per references/layout.md, re-run.
  8. Render silent, then python3 scripts/frame_audit.py film/renders/silent.mp4 → must print 0 failed. Make a contact sheet and put it next to examples/office-agent/reference-sheet.jpg; answer the three comparison questions in quality-bar.md. If yours is clearly thinner, go back to step 6 — add product on screen, not decoration. Then fix the single worst frame.
  9. Score it — only now, after all three gates pass. score_lib prints a NOT READY banner if .gates/ is missing any of the three pass stamps. (scripts/score_template.py → audio/score.py), mix SFX, loudnorm to −14 LUFS, mux with -c:v copy.
  10. Deliver one MEDIA:/file line per output. State what you verified (both audits' numbers, lint, loudness, the reference comparison) and what only the user can judge (taste, how the music sounds). Flag any invented copy and any low-res logo.

Commands

bash
python3 src/build.py
python3 <skill>/scripts/film_audit.py film/index.html           # code gate: must be "0 failed"
node <skill>/scripts/layout_audit.mjs film/index.html --shots .gates/layout   # placement gate: "0 failed"
cd film
npx --yes hyperframes@0.8.78 lint                                # target 0 errors; read every warning
PRODUCER_BROWSER_GPU_MODE=hardware \
  npx --yes hyperframes@0.8.78 render . -q high --fps 30 -o ./renders/silent.mp4
python3 <skill>/scripts/frame_audit.py renders/silent.mp4        # pixel gate: must be "0 failed"
<skill>/scripts/review_frames.sh renders/silent.mp4 snapshots 2 "11.0,13.5"

Hard rules (each one came from a rejected or broken film)

  • Floors are floors, not targets. ≥ 90 non-filler tweens, ≥ 8 mechanics, ≥ 2 non-crossfade transitions, ≥ 8 eases, ≥ 8 camera moves, ≥ 5 events in every act (15s film).
  • Show the product working in at least two acts — its interface, its output, a device running it, real-looking data. An abstract metaphor (ink, particles, light, a seed growing) with no product in it could be an ad for anything and is rejected even if every number passes.
  • Fill the frame. A line of text and one small shape on a flat ground is an empty frame. Every act needs background, midground and foreground layers.
  • Copy comes from the source. Every on-screen line traces to the brief or the product page; anything invented is flagged to the user.
  • Real logo, real imagery, real fonts. The brand's current logo file for this exact product, checked against the live site and app store; ≥ 2 official images used as material (screens, masks, key visuals); every file sourced in assets.md; embedded woff2 subsets. A logo "inspired by" the brand, or the parent app's favicon standing in for a sub-product, is a wrong logo.
  • When the brand ships motion, cut it; don't decorate it. Footage on a stage of its own sampled ground colour, no grain over it, no focus brackets / rings / cursors drawn on the product, one signature transition from the product's own shape on ≥ 3 act changes, every product shot anchored to the same screen line, copy from the brand's page in the audience's language.
  • The logo lives in the lockup, nowhere else. No corner mark through the film (the layout gate fails it). The last frame is one group on the product's axis; never add your own title next to an official frame that already has type in it, never let the product cross the type, never contain a non-16:9 still so it floats in bars (references/layout.md, "The last frame").
  • No template look. No beige/orange-pink lifestyle palettes, no centring everything, no equal-length acts doing the same thing. The accent colour is a highlight, not a ground.
  • One pull-back per film. It reads as "this is ending".
  • Motion lives on .cam layers; initial hidden state lives in CSS; never fade an element that also has its own opacity tween. (Ghosting.)
  • No Math.random() in the timeline — frames are seeked out of order.
  • Ship it with music, landing the final hit on a downbeat. A brand film with no score is an incomplete delivery.
  • Verify honestly. You cannot hear audio or watch motion in real time; report the checks you ran, not a feeling.
  • Client work gets reviewed before publishing. Say so in the delivery.

Bundled resources

scripts/film_audit.py       code gate, before render (floors, shot list, copy and asset sources)
scripts/layout_audit.mjs    placement gate, headless Chrome (hidden / colliding / clipped text, safe area, half-parked cards, corner logos)
scripts/frame_audit.py      pixel gate, after render (coverage, detail, motion)
scripts/harvest_page_images.py  every image (and with --video every product clip) on an official page → index.tsv + contact sheets
scripts/fetch_cjk_fonts.sh  Noto Serif/Sans SC subsets for exactly the characters used
scripts/build_template.py   template → film/index.html + CJK character list
scripts/review_frames.sh    render → contact sheet + stills + duration check
scripts/score_lib.py        instrument library + reverb + master (import, don't edit)
scripts/score_template.py   five-act arrangement to copy and fill in
assets/hyperframes.json     project config
examples/office-agent/      complete reference film (placeholder brand), with its
                            shot-list.md, copy.md and reference-sheet.jpg

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Code Brand Film AI skill do?

Produce a short brand film or product promo (10-60s, typically 15s) entirely as code: one static HTML page with a paused GSAP timeline that a renderer seeks frame by frame, plus a music score synthesized in Python. Use when the user wants a 品牌宣传片, 广告片, 宣传动画, 产品短片, 发布会开场视频, brand film, promo video, product teaser, logo animation, or a generated ad spot — especially when they ask for 高级感 / 艺术感 / 电影感 / 别像模板, or when an existing video "looks like AI made it" or "too simple" and needs to be redone properly. Also use when a film already exists and needs a score, a soundtrack, background music, or...

Why use Code Brand Film on TypingMind?

Because you install it once and use it with any model. Code Brand Film is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Code Brand Film in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/nutllwhy/code-brand-film/tree/main. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Code Brand Film?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Code Brand Film?

As many as you like. As long as a model supports skills, you can use Code Brand Film with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Code Brand Film AI skill free?

Yes. It is published on GitHub by nutllwhy under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇