Oct 22, 2026year 2026relevance 63619 min read

Claude Opus 5.5 Video · 8 GitHub Repositories to Bookmark

The useful Claude video ecosystem is larger than a folder of viral prompts. These eight repositories cover four distinct jobs: discovering references, installing reusable production systems, studying complete source projects and learning from evidence-linked archives. This guide verifies every shortened link, updates the repository counts as of October 22, 2026, explains the technical route behind each project and flags the licensing boundaries that matter before you copy code, assets, songs or prompts.

Video credit: mexicat.Source video: i'm upping my p(doom).

Quick Summary

All eight shortened links resolve to active GitHub repositories. None was archived when checked on October 22, 2026.
The original star estimates are already stale. The repositories ranged from about 685 to 9,181 stars at the evidence cutoff, so treat stars as a dated popularity signal rather than a quality score.
The largest discovery library, awesome-opus5-5-videos, currently reports 703 prompts and 418 live remakes, not merely a small gallery of viral clips.
Lemo Opuscar now lists 59 reusable film styles, not 51. It can be installed as a Claude Code plugin and can also help create and preserve a custom style.
The two Pdoom repositories are independent implementations of the same song concept. One emphasizes deterministic Three.js rendering and true 4K60 export. The other uses p5.js and p5.brush with a storyboard and animation guide generated during the project.
Licensing differs. Several repositories use MIT, one uses CC BY 4.0 for original research, and two do not currently expose a root repository license through GitHub. Source availability does not automatically grant reuse rights.

--what to try first

Start with the repository that matches your job. A prompt gallery, a reusable skill, a complete film and a research archive solve different problems.
Use awesome-opus5-5-videos for broad visual discovery and creator-linked prompt research.
Use Lemo Opuscar when you want a guided, reusable production system with many named styles and a Claude Code plugin route.
Use mexicat/pdoom-video to study deterministic Three.js rendering, lyric timing, post-processing and true 4K60 output.
Compare JohnHeibel/PDoomVideo with mexicat/pdoom-video to see how the same song can produce very different architectures, scene systems and visual languages.
Use mg-styles-15 when you want a compact style atlas where every ten-second film includes the prompt, source, notes and a repeatable render harness.
Use ClaudeAnimationBase for hand-painted character animation, expression systems and a storyboard-first p5.js workflow.
Use awesome-claude-5-5-videos when provenance matters. It separates model labels, links to original posts and documents the limits of its visual sampling.
Use opus-video-prompts when Chinese prompts and multilingual examples are useful, but verify usage rights because the repository currently has no root license listed by GitHub.
Clone into a test folder, inspect scripts and dependencies, confirm asset licenses, render a short proof and keep human approval before publishing anything derived from these projects.

The verified snapshot: what changed from the shared list

The eight t.co links resolve to the repositories documented below. GitHub's repository API reported 9,181 stars for awesome-opus5-5-videos, 5,503 for Lemo Opuscar, 3,256 for mexicat/pdoom-video, 2,483 for JohnHeibel/PDoomVideo, 1,822 for mg-styles-15, 1,563 for ClaudeAnimationBase, 718 for awesome-claude-5-5-videos and 685 for opus-video-prompts on October 22, 2026.

Those numbers are a snapshot, not a ranking of production quality. Stars move constantly and can reflect timing, audience size, documentation or novelty. The more useful comparison is what each repository contains, whether its source can be reproduced and whether its license covers the material you want to reuse.

Two content counts also changed. Lemo Opuscar's README now lists 59 styles. The Chinese opus-video-prompts collection lists 18 complete prompts, 54 public cases and nine technical routes. The article uses the live repository claims instead of preserving older social-post numbers.

Choose by job, not by star count

Discovery and prompt research: yihui-dev/awesome-opus5-5-videos and joeseesun/opus-video-prompts. Reusable production systems: lemomo-ai/lemo-opuscar and JohnHeibel/ClaudeAnimationBase. Complete films to reverse engineer: mexicat/pdoom-video and JohnHeibel/PDoomVideo. Compact style studies: Vincentwei1021/mg-styles-15. Source-linked research and model comparison: athemeroy/awesome-claude-5-5-videos.

A curated prompt can teach you how a creator described the goal, but it may omit hidden context, revisions, installed tools and assets. A complete source project exposes more of the production system. A reusable skill goes further by packaging the method, but it also deserves a deeper security and dependency review before installation.

For a first experiment, pick one small example, replace the story and brand material with content you own, preserve its output specification and render only five to ten seconds. That is enough to validate the environment before spending tokens and rendering time on a complete film.

1. awesome-opus5-5-videos · The broad discovery library

Repository: yihui-dev/awesome-opus5-5-videos. GitHub snapshot: 9,181 stars, 921 forks and an MIT license on October 22, 2026. The README reports 703 prompts, with 418 examples paired with live remakes on Skillry and 100 highlighted remakes covering distinct prompts.

Its value is breadth plus attribution. Entries link to the creator's original post and identify routes such as Canvas, SVG, GSAP, Three.js, WebGL, GLSL, particles and Web Audio. Categories include motion graphics, explainers, 3D scenes and interactive work.

Best use: reference research and prompt anatomy. Do not assume every example is a complete reproducible project. Some entries preserve a public prompt or original post rather than the entire source, asset set and revision history. Follow the creator link before presenting a technique or result as your own evidence.

2. Lemo Opuscar · The reusable film-style system

Repository: lemomo-ai/lemo-opuscar. GitHub snapshot: 5,503 stars and 605 forks. The current library contains 59 ready-to-use styles and a workflow for adapting a library style or creating and saving a new one. The recommended Claude Code route installs the lemo-opuscar plugin, while other coding agents can use the supplied skill folder.

This is more than a lookbook. The agent reads shared direction, story and technical guides plus a style-specific file. It can stop for a storyboard, render a style frame for approval and keep approved custom styles in a separate user folder so library updates do not overwrite them. The README estimates roughly 30 to 60 minutes of agent work for a film and lists Node 20+, FFmpeg and Python 3.11+ or uv as prerequisites.

License boundary: code and documentation are MIT. Demo films and stills are CC BY 4.0 with LemoLab credit, but third-party assets retain their own terms and three named voice tracks are excluded from the demo-film license. Read each demo's credits instead of treating one repository badge as permission for every file.

3. mexicat/pdoom-video · Deterministic Three.js at 4K60

Repository: mexicat/pdoom-video. GitHub snapshot: 3,256 stars and 357 forks under MIT. This is a generative, code-rendered music video with word-synchronized karaoke typography. Its TypeScript and Three.js renderer treats every frame as a deterministic function of song time, so the browser preview and offline export share the same scene logic.

The project documents the treatment, engine, scene modules, audio analysis and render pipeline. Bun and Vite run the app, headless Chrome produces frames, and FFmpeg encodes the output. The default path renders 1920 by 1080 at 60 fps. The documented scale-two path produces a real 3840 by 2160 render rather than enlarging a 1080p result.

The README reports demanding local costs for the 4K path, including roughly 57 minutes on an M5 Pro for the full song and high memory use. Code is MIT, but the repository explicitly excludes the song and lyrics from that license. Use it to study engineering and visual timing, not as a blanket license to redistribute the included music.

4. JohnHeibel/PDoomVideo · A contrasting p5.js music-video architecture

Repository: JohnHeibel/PDoomVideo. GitHub snapshot: 2,483 stars and 246 forks. This is a separate music video for the same song concept, built with p5.js and p5.brush. The source is divided into nine chapters, shared character and timeline code, a browser studio, a frame renderer, a storyboard and an animation guide.

The README says the project went through two Claude Code generations. The second version used an Opus-written animation guide to brief parallel subagents, then rendered resumable frames through headless Chrome before FFmpeg assembled the MP4. That makes it a useful comparison with mexicat's Three.js and GPU-heavy approach.

GitHub did not report a root license for this repository at the evidence cutoff. The README also traces the song to an older YouTube upload rather than documenting a reusable music license. Study the implementation, but obtain explicit permission or replace protected material before redistributing or commercializing a derivative.

5. mg-styles-15 · Fifteen complete motion-design studies

Repository: Vincentwei1021/mg-styles-15. GitHub snapshot: 1,822 stars and 251 forks under MIT. It contains fifteen ten-second films, each with style notes, a full prompt and complete source. The styles range from line boil, isometric 2.5D and flat vector to 3D, cyberpunk HUD, collage, glassmorphism, Bauhaus, synthwave, pixel art and liquid motion.

Most films render a web page frame by frame through headless Chrome, then use FFmpeg to join picture and audio. The 3D example uses Blender for the picture and a web layer for typography and grain. A shared render harness, audio tools, prompt template and review rubric make the collection more reusable than a gallery of isolated clips.

The project says each original film received two review and revision rounds. That matters because a copied prompt is a starting condition, not a promise that the first output will match the showcase. Replace the creative seed and brand claims, keep the relevant signature features, then run your own visual and factual review.

6. ClaudeAnimationBase · Hand-painted characters and 31 expressions

Repository: JohnHeibel/ClaudeAnimationBase. GitHub snapshot: 1,563 stars and 152 forks under MIT. The starter kit combines p5.js and p5.brush with a reusable Clawd character, 31 acted emotions, multiple views, props, dances, timeline helpers, a browser studio and a headless renderer.

The shortest documented workflow is to ask the coding agent to read ANIMATION_GUIDE.md and make a short story. The guide pushes the model to storyboard before coding, build shot by shot, create contact sheets, inspect its own work and output an MP4. Node.js, Chrome and FFmpeg are required.

Best use: learning character-state systems and hand-painted 2D animation without building the full engine from zero. The README warns that watercolor fills can render slowly without a dedicated GPU. On weaker hardware, ask for a lighter fill treatment before the agent commits to hundreds of frames.

7. awesome-claude-5-5-videos · The evidence-oriented archive

Repository: athemeroy/awesome-claude-5-5-videos. GitHub snapshot: 718 stars and 54 forks. Its October 11 collection summary records 424 distinct source posts, including 301 Opus, 54 Sonnet, 21 Haiku and 68 creator-reported Fable entries. Groups can overlap, so those model rows should not be added as a total.

This repository is useful when provenance matters more than immediate copying. It separates model labels, links original posts, samples selected outputs as nine-frame previews and explicitly says frame sampling cannot verify full motion or model calls. It also documents production paths, prompt lengths, inputs, revision evidence and cost examples without pretending unrelated tasks form a clean model benchmark.

Original writing and charts are CC BY 4.0. Linked videos, posts and stills remain with their creators. Cite the original source when using a case, and cite the archive separately when relying on its analysis or categorization.

8. opus-video-prompts · Chinese prompts and multilingual examples

Repository: joeseesun/opus-video-prompts. GitHub snapshot: 685 stars and 60 forks. The README currently lists 18 complete prompts, 54 public cases and nine technical routes. It covers Canvas, p5.js, Three.js, Remotion, Manim and Blender scripting, plus workflows that use headless browsers, FFmpeg, generated audio or external video models.

The collection is especially useful for Chinese-language briefs and for comparing short one-line requests with long production specifications. Every included prompt is presented with an original source link when available, and the examples span motion graphics, explainers, product films, lyric videos and interactive 3D work.

GitHub did not report a root license at the evidence cutoff. Public prompts and public source links are evidence of publication, not automatic permission to relicense the collection or its examples. Copy the learning pattern, verify the original creator and replace brand, music, footage and factual claims with material you are authorized to use.

What the two Pdoom projects teach when compared

mexicat/pdoom-video behaves like a deterministic graphics engine. Its strength is exact timing, GPU effects, typographic treatment, structured scene modules and documented 1080p60 or true 4K60 export. JohnHeibel/PDoomVideo behaves more like a hand-painted character-animation studio, using p5.brush, nine narrative chapters, a storyboard and a reusable animation guide.

Neither approach is universally better. The right question is whether the piece depends on precise lyric typography and sophisticated real-time graphics, or on expressive characters, painted textures and story-led transitions. Compare file structure, rendering strategy and revision tools before comparing only the finished MP4.

Both projects also reveal the difference between code rights and media rights. A permissive code license does not necessarily cover a song, lyrics, voice recording, fonts or third-party visuals. Build a source and license manifest before adapting either project for a client.

A safe way to evaluate any of the eight repositories

1. Read the README, license and credits before cloning. 2. Check recent commits, issues and release notes. 3. Inspect package scripts and installation commands. 4. Clone into a disposable test directory. 5. Install only the documented dependencies. 6. Render the smallest included example. 7. Replace the story with content you own. 8. Review a contact sheet and the complete video. 9. Record the repository commit and every external asset license. 10. Publish only after human approval.

Do not paste private client material into a public example or allow an unfamiliar installer to access unrelated folders. A Claude Code plugin or skill can be convenient, but it can also invoke tools and scripts. Read its files and understand the requested permissions before installation.

Keep the first deliverable deliberately small. A five-second test with one title, one transition and one audio cue is enough to confirm fonts, browser rendering, FFmpeg, color, frame rate and output paths. Scale only after that test is reproducible.

Which repository should you open first?

For inspiration: start with awesome-opus5-5-videos. For a guided style workflow: start with Lemo Opuscar. For engineering depth: open mexicat/pdoom-video. For painterly character animation: use ClaudeAnimationBase. For a compact prompt-to-source curriculum: use mg-styles-15. For research and provenance: use awesome-claude-5-5-videos. For Chinese prompt examples: use opus-video-prompts.

The strongest learning sequence is discovery, source study, controlled remake and original production. Pick a reference, inspect the prompt and source, reproduce a tiny test, then replace its story, design system and assets. Bookmarking becomes useful only when it turns into a repeatable, rights-cleared workflow.

Evidence cutoff: October 22, 2026. Star counts, prompt totals, style counts, installation instructions and repository licenses can change. Recheck the live repository immediately before installing, citing or redistributing its contents.

Repository evaluation prompts · Inspect before you install

These are original prompts for evaluating and adapting the repositories safely. They are not copied from the projects. Replace every bracketed field and keep installation, external downloads and publication behind human approval.

Prompt pack credit: JQ VISIONS · Original repository evaluation prompts.

1. Read-only repository audit

Understand the project, permissions and risks before running any code.

Inspect [repository URL] without installing or executing it. Summarize its purpose, architecture, entry points, build and render commands, required runtimes, external downloads, package scripts, network access, output folders and current license files. Separate repository-owned code from songs, fonts, footage, voices and other third-party assets. List anything that is unclear or lacks an explicit reuse license. End with a safe test plan that uses no private data.

2. Minimal reproducibility test

Prove the environment with the smallest possible output before attempting a complete film.

Using the already approved local clone of [repository], identify the smallest documented example. Before running it, show the exact commands, expected downloads, estimated disk use, ports, files created and render duration. After approval, render no more than [5] seconds at [preview resolution] and [fps]. Produce a contact sheet and report missing dependencies, visual defects, console errors and output paths. Do not modify global configuration or publish anything.

3. Rights-cleared adaptation plan

Replace the original creative material while preserving only reusable technical ideas.

Create an adaptation plan for [project] using only these approved materials: [owned or licensed assets]. Preserve the technical lessons, not the original brand, story, characters, lyrics, music, copy or distinctive artwork. List which source files can be reused under the repository license, which assets must be replaced and what attribution remains required. Propose a new design system, storyboard and five-second proof. Stop before editing files.

4. Compare two implementations

Use for the Pdoom repositories or any pair of projects solving a similar problem.

Compare [repository A] and [repository B] as engineering systems. Evaluate timeline model, rendering route, scene architecture, typography, audio synchronization, preview workflow, resumability, hardware demands, licensing, maintainability and ease of revision. Do not rank them from screenshots alone. Recommend one for [specific project] and explain the tradeoffs and acceptance test.

5. Final video quality gate

Review the output before it leaves the local workspace.

Review the complete render and a contact sheet for [video]. Check spelling, factual claims, safe margins, contrast, frame pacing, transition continuity, lyric or voice timing, clipping, missing frames, audio peaks, codec, resolution, frame rate and duration. Verify every asset against the rights manifest. Return pass, fix or reject for visual quality, technical delivery, factual accuracy and usage rights. Do not upload or publish the video.

FAQ

Which repository is best for beginners?

Lemo Opuscar offers the most guided route, while mg-styles-15 gives the clearest compact prompt-to-source examples. Beginners should still read installation scripts, use a disposable folder and render a short proof before a full film.

Which repository has the most prompts?

awesome-opus5-5-videos reports 703 prompts at the October 22, 2026 evidence cutoff. Counts change, and some entries are prompt records or source posts rather than complete standalone projects.

Does Lemo Opuscar contain 51 styles?

Not in the current README. It lists 59 ready-to-use styles as of October 22, 2026, plus a workflow for adapting or creating a personal style.

Can pdoom-video really render 4K at 60 fps?

The mexicat repository documents a true 3840 by 2160, 60 fps render path using its scale-two option. That describes output format, not real-time speed. The README reports a lengthy, GPU-bound local render with substantial memory and storage demands.

Are the two Pdoom repositories duplicates?

No. They use the same song concept but different implementations. mexicat uses TypeScript and Three.js with a deterministic graphics engine. John Heibel's version uses p5.js and p5.brush with hand-painted characters, chapters, a storyboard and an animation guide.

Can I use everything commercially because it is on GitHub?

No. Check the license for every repository and every included asset. Code, demo films, music, lyrics, fonts, footage and voices may have different terms. Two repositories in this list did not expose a root license through GitHub when checked.

Do these projects use an AI video generator?

Many examples draw frames through Canvas, SVG, p5.js, Three.js, WebGL, Blender or Remotion and assemble them with FFmpeg. Some curated collections also include workflows that direct external video models. Check each case rather than assuming one production route.

Why are the star counts different from the shared list?

GitHub stars change continuously. This article uses a dated API snapshot from October 22, 2026 and treats it as a popularity signal, not a permanent repository property or measure of quality.