Lazy Frames logo

Lazy Frames

Organization
cosmicstack-labs
lazy-frames

Turn any input (a website URL, a text brief, existing footage, a music track) into a finished, pixel-perfect MP4 — fully local, byte-stable across renders. The agent researches the input and dumps a typed SPEC; the Lazy Frames CLI turns that spec into video. Use for website promos, cinematic clips, motion graphics, explainer videos, social ads, and any programmatic video generation that must be deterministic and work offline. Install: npm install lazy-frames | Raw skill URL: https://lazy-frames.cosmicstack.ai/skill.md

Overview

Publishercosmicstack-labs
Repositorylazy-frames
Skill namelazy-frames
Stars
169
Forks
19
Bundled files
5
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.

  • 5 bundled files

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

  • Open source

    Published by cosmicstack-labs on GitHub. Read the source before you install it.

Installation

Install the Lazy Frames 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/cosmicstack-labs/lazy-frames/tree/main/skill
  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/cosmicstack-labs/lazy-frames.git /tmp/lazy-frames
mkdir -p .claude/skills
cp -r /tmp/lazy-frames/skill .claude/skills/lazy-frames
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Lazy Frames 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 Lazy Frames 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 Lazy Frames 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.

Lazy Frames

Lazy Frames renders video from a typed JSON spec — a composition is a spec.json file whose scenes declare timing, content, and style; the CLI turns it into a deterministic, pixel-perfect MP4. The core renderer needs no cloud or keys. Reviewed plugins may use explicitly declared external services; generated media is cached locally before deterministic rendering.

Current CLI release: lazy-frames@0.6.3.

  • Captured website media is center-cropped by default in browser-frame, ui-callout, and both flat and depth parallax scenes. Override params.position only when the subject requires a different anchor.
  • capture, check, gen, preview, and render report live phases on stderr. Render reports frame percentages. With --json, stdout remains clean machine-readable JSON.

1. Start from project state

StateAction
Existing project with spec.jsonRead the spec, make the edit or run the requested operation. Skip routing.
Fresh request with a website URLRoute to website-promo workflow (§ 3).
Fresh request with a text briefRoute to cinematic workflow (§ 3).
Fresh request with existing footageRoute to edit workflow (§ 3).
UnclearAsk one question: "Is this about a specific website, a topic/brief, or existing video footage?"

2. Prerequisites

bash
node --version   # >= 20
ffmpeg -version  # required
python3 --version   # Python 3; on Windows `python` is fine
ls "/Applications/Google Chrome.app"  # macOS; Windows uses Chrome in Program Files, or set CHROME_PATH

Install from npm:

bash
npm install lazy-frames
npx lazy doctor

Or build from source:

bash
npm install && npm run build

Verify the environment:

bash
node packages/cli/dist/index.js doctor

Doctor reports: Node platform/memory, Python version + tier, and provider availability (procedural imagery, depth, music, TTS). Local narration uses tts.say (macOS say or Windows SAPI); external providers are installed as scoped plugins. If image.mlx-photoreal is unavailable (expected on <16 GB machines), the procedural image generator is used instead — still produces cinematic stills, just not photoreal.

Plugins and external providers

ElevenLabs is included by default but still requires project-scoped approval before network use. When the user requests any provider or storytelling capability, search the registry, inspect its status and permissions, and ask before approving or installing it:

bash
node packages/cli/dist/index.js plugin search tts
node packages/cli/dist/index.js plugin info elevenlabs
node packages/cli/dist/index.js plugin install elevenlabs -p projects/cine

Marketplace scaffolds may be installed but cannot execute until a reviewed adapter ships. Never install arbitrary package URLs or store API keys in project files. ElevenLabs reads ELEVENLABS_API_KEY from the environment. Registry: https://lazy-frames.cosmicstack.ai/plugins/index.json

3. Workflows

website-promo

Input: a URL. Output: a promo/showcase MP4 built from the site's own captured visuals.

bash
# 1. Capture the site (screenshots @2x, palette, copy, fonts, logo)
node packages/cli/dist/index.js capture https://example.com projects/acme

# 2. Review the generated starter spec + captured assets
cat projects/acme/spec.json
ls projects/acme/assets/sites/example.com/

# 3. Edit the spec if needed (adjust scenes, copy, timing, transitions)

# 4. Validate + run gates
node packages/cli/dist/index.js snapshot projects/acme --update
node packages/cli/dist/index.js check projects/acme

# 5. Render
node packages/cli/dist/index.js render projects/acme

The capture command writes:

  • assets/sites/<domain>/hero.png + full.png (screenshots at 2x DPR)
  • assets/sites/<domain>/ledger.json (palette, fonts, copy blocks, metadata)
  • A starter spec.json with 5 scenes built from the captured content
  • Screenshot scenes explicitly use { "x": "center", "y": "center" } unless the spec overrides the crop anchor

cinematic

Input: a text brief or topic. Output: a 30–60 s cinematic clip with generated stills, depth-parallax, narration, music, SFX.

bash
# 1. Generate procedural stills + matching depth maps
node packages/cli/dist/index.js gen image -p projects/cine --seed 21 --style ridge --palette "#070B14,#F59E4C,#F4F7FB" --name ridge-01
node packages/cli/dist/index.js gen image -p projects/cine --seed 42 --style dune --palette "#070B14,#7FB7D9,#F4F7FB" --name dune-01

# 2. Write spec.json referencing the generated assets (see references/spec-format.md)

# 3. Draft scene-linked narration, then generate optional audio assets
node packages/cli/dist/index.js script projects/cine --apply
node packages/cli/dist/index.js gen music -p projects/cine --mood calm --bpm 90 --bars 12 --seed 21
node packages/cli/dist/index.js gen tts -p projects/cine --text "Every frame computed locally." --name n1

# 4. Validate + render
node packages/cli/dist/index.js snapshot projects/cine --update
node packages/cli/dist/index.js check projects/cine
node packages/cli/dist/index.js render projects/cine

edit

Input: existing footage (MP4) + a brief. Output: the footage composited into a video with overlays, grading, transitions.

Write a spec with video-layer scenes referencing the footage file, plus typography/atmosphere scenes for titles and transitions. The engine owns video seeking (deterministic per-machine frame extraction). See references/scene-types.md for video-layer params.

4. Agent contract

The agent's job across all workflows:

  1. Research the input (read the site, understand the brief, inspect the footage).
  2. Produce or refine spec.json — a valid typed spec (see references/spec-format.md).
  3. Generate the narration script when requested (lazy script <project> --apply), then review its scene-linked beats.
  4. Generate assets if needed (lazy gen image, lazy gen music, lazy gen tts).
  5. Run lazy check and fix every error. Warnings are advisory.
  6. Run lazy snapshot --update to establish the regression baseline.
  7. Run lazy check again — snapshot + seek-determinism gates must pass.
  8. Run lazy preview and hand the URL to the user. Ask whether to revise or render.
  9. Render only after approval. lazy render — never before the user says go.
  10. Verify the output: confirm the file exists, is non-empty, has the expected duration.
  11. Report the output path + sha256.

Non-negotiable rules

  • Never render before the user approves. Always pause at preview.
  • Never skip lazy check. It catches schema errors, missing assets, and nondeterminism.
  • Always run lazy snapshot --update before the first lazy check in a new project.
  • Never edit .lazy/ — it's generated. Edit spec.json and re-render.
  • Never install an unreviewed plugin or persist provider credentials. Use the registry and environment variables.
  • Reference assets by project-relative paths in spec (e.g., assets/gen/ridge-01.png).
  • Scenes are sequential with optional overlap. Overlap + fade/dissolve = crossfade.
  • All timing is in milliseconds. Start times are absolute from the composition start.
  • Palette convention: palette[0] = bg, palette[1] = accent, palette[2] = fg.
  • Width/height must be even (h264 yuv420p requirement).
  • Fonts are bundled: Inter (body) and Space Grotesk (display). No external fonts.
  • Progress belongs on stderr. Commands using --json reserve stdout for their final JSON result.

5. Key concepts

Determinism

Same spec + same machine = byte-identical MP4 (verified by sha256). Three Chrome flags enforce this: --jitless (V8 interpreter only), --use-angle=swiftshader (software rendering), and integer-pixel transform rounding in the runtime. CSS sub-pixel text rendering is cross-session nondeterministic — the runtime snaps all transforms to integer pixels.

Gates

lazy check runs two blocking gates:

  • Snapshot regression: renders keyframes (scene midpoints + transition midpoints), hashes their pixels, and compares against snapshots/baseline.json. Drift = error with scene + timestamp.
  • Seek determinism: renders 5 sampled timestamps twice in one browser session, compares hashes.

Audio

Audio is declarative in the spec (narration, music, sfx). The renderer generates and caches canonical WAV audio, then mixes via ffmpeg:

  • Narration: local say or an installed TTS plugin, placed at startMs or anchored to sceneId + offsetMs
  • Music: procedural synth (chord pads + percussion) seeded by seed, looped to video duration
  • SFX: synthesized whoosh/hit/rise/boom, placed at atMs with gain

LUT / grade

Two layers:

  • Stage grade (style.grade): CSS filter preset on the composition root (fast, subtle)
  • Output LUT (outputs[0].lut): 3D LUT .cube file applied via ffmpeg lut3d during encode (cinematic color grade). Bundled LUTs: teal-orange, noir-film, faded-vintage.

6. References

FileURLRead it to…
spec-format.mdhttps://lazy-frames.cosmicstack.ai/references/spec-format.mdauthor a valid spec.json — full schema, all fields, defaults
scene-types.mdhttps://lazy-frames.cosmicstack.ai/references/scene-types.mdpick a scene type and configure its params
workflows.mdhttps://lazy-frames.cosmicstack.ai/references/workflows.mdstep-by-step for website-promo, cinematic, and edit workflows
cli.mdhttps://lazy-frames.cosmicstack.ai/references/cli.mdevery CLI command with flags and examples
gates.mdhttps://lazy-frames.cosmicstack.ai/references/gates.mdsnapshot regression + seek determinism gate usage

These references are also available in the installed package at skill/references/ and in the repo at skill/references/.

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 Lazy Frames AI skill do?

Turn any input (a website URL, a text brief, existing footage, a music track) into a finished, pixel-perfect MP4 — fully local, byte-stable across renders. The agent researches the input and dumps a typed SPEC; the Lazy Frames CLI turns that spec into video. Use for website promos, cinematic clips, motion graphics, explainer videos, social ads, and any programmatic video generation that must be deterministic and work offline. Install: npm install lazy-frames | Raw skill URL: https://lazy-frames.cosmicstack.ai/skill.md

Why use Lazy Frames on TypingMind?

Because you install it once and use it with any model. Lazy Frames 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 Lazy Frames in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/cosmicstack-labs/lazy-frames/tree/main/skill. 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 Lazy Frames?

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 Lazy Frames?

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

Is the Lazy Frames AI skill free?

It is published on GitHub by cosmicstack-labs. Check the repository for licensing terms. 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 👇