Now liveThe Skillselion MCP - thousands of ranked skills, loaded into your agent mid-task. No install.Get it →
denissergeevitch avatar

Youtube Poop Video Maker

denissergeevitch/youtube-poop-video-maker

  • 1 installs
  • 39 repo stars
  • Updated March 12, 2026
  • denissergeevitch/youtube-poop-video-maker

Plan and render a dense 20-60 second remix or YouTube poop video from text, webpages, or media using FFmpeg with typography and audio treatments.

About

Produces a finished FFmpeg-rendered short remix video with many micro-scenes, typography, and controlled audio from varied inputs. A developer uses it to generate absurd supercuts, cursed trailers, or glitch montage videos.

  • Plans dense multi-scene 20-60s videos
  • Renders finished MP4 with FFmpeg, not just a storyboard

Youtube Poop Video Maker by the numbers

  • 1 all-time installs (skills.sh)
  • Ranked #1,200 of 1,335 Generative Media skills by installs in the Skillselion catalog
  • Data as of Jul 29, 2026 (Skillselion catalog sync)
npx skills add https://github.com/denissergeevitch/youtube-poop-video-maker --skill youtube-poop-video-maker

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

Listed on Skillselion
Installs1
repo stars★ 39
Last updatedMarch 12, 2026
Repositorydenissergeevitch/youtube-poop-video-maker ↗

What it does

Plan and render a dense 20-60 second remix or YouTube poop video from text, webpages, or media using FFmpeg with typography and audio treatments.

Files

SKILL.mdMarkdownGitHub ↗

YouTube Poop Video Maker

Produce a finished video, not just a concept or storyboard. The result should feel inventive, dense, and internet-native while still looking composed.

See:

  • the reference guide
  • the mood anchor guide
  • the output checklist

Useful bundled files:

  • scripts/plan_video.py
  • assets/style_blueprints.json
  • assets/scene_atoms.json
  • assets/trigger_eval_queries.json

When to use this skill

Use this skill when the user wants:

  • a YouTube poop, cursed trailer, absurd montage, reflective glitch short, or surreal supercut
  • a video rendered with ffmpeg from text, a webpage, code, a document, media, or no source material at all
  • a video with a personal spin, strong POV, or “what it feels like” framing
  • a self-aware AI / LLM / internet-culture short
  • something “weirder,” “more edited,” “hypercut,” “meme-y,” “chaotic,” or “poetic,” but still aesthetically pleasing

Do not use this skill for routine trimming, simple caption burns, ordinary explainers, standard motion graphics, or conventional ads unless the user explicitly wants remix behavior.

Core promise

This skill should not flatten creativity into a single house style. The default seeded-remix blueprint is a mood lens, not a cage.

The finished piece should usually have:

  • more scenes than a straightforward edit would have
  • a clear thesis or emotional claim
  • readable frames and controlled palettes
  • one dominant recurring motif
  • at least one authored gesture that feels personal rather than generic
  • a decisive final button

Default output target

Unless the user specifies otherwise:

  • one finished MP4
  • 20–60 seconds
  • 1920x1080 at 30 fps for normal YouTube
  • 1080x1920 at 30 fps only when the request clearly implies Shorts/Reels/TikTok/vertical
  • H.264 video, AAC audio, yuv420p, +faststart

Workflow

1) Inventory the material quickly

Work with whatever is already available.

  • If the user provided text, mine phrases, contradictions, repeated wording, and emotional temperature.
  • If the user provided a webpage or URL, extract headlines, CTA text, layout motifs, screenshots, crops, and scroll moments.
  • If the user provided code, use readable snippets, terminal windows, errors, comments, variable names, logs, and cursor motion.
  • If the user provided documents, use section titles, callouts, diagrams, and page framing.
  • If the user provided images, audio, or video, mine loops, reaction beats, freeze-frames, speaker fragments, textures, and captionable details.
  • If the user provided nothing, synthesize the whole world from the request: title cards, browser chrome, terminal panes, lower-thirds, dashboards, progress bars, token streams, warning cards, diagrams, and simple motion graphics.

Do not stall because the source is thin. Invent support scenes.

2) Decide the thesis

Before editing, define one short internal sentence:

“This video makes X feel like Y.”

Examples:

  • “This video makes talking to a language model feel like a late-night emergency broadcast from inside a datacenter.”
  • “This video makes a landing page feel like a ceremonial lie detector.”
  • “This video makes a codebase feel like a guilty terminal confession.”

Every recurring gag should support that claim.

3) Use the default seeded remix unless a style is requested

Unless the user names a style, use the default seed-remix-default blueprint. It samples palette, typography, motion, transitions, audio, and gag cues from the other bundled blueprints each run.

Run:

python3 scripts/plan_video.py --query "$USER_REQUEST" --material-summary "$MATERIAL_SUMMARY" > plan.json

If the user explicitly asks for a particular bundled look, override the default remix with:

python3 scripts/plan_video.py --query "$USER_REQUEST" --material-summary "$MATERIAL_SUMMARY" --style-id STYLE_ID > plan.json

For CRT-newsroom or AI-meltdown looks, prefer the bundled pool entries crt-token-meltdown, headline-seizure-desk, or context-window-panic before inventing a one-off style from scratch.

Use the plan as creative guidance for palette, motion, audio, scene density, and scene invention. When the default remix is used, treat the reported seed sources as prompts, not a checklist.

4) Build a scene swarm, not just a beat sheet

Aim for many micro-scenes.

Recommended density by runtime:

  • 20–28s: about 9–12 scenes
  • 29–40s: about 11–15 scenes
  • 41–60s: about 13–18 scenes

Within those counts, include:

  • anchor scenes that clearly present the source or thesis
  • bridge scenes that act like interruptions, glitches, or breath marks
  • one break scene with relative stillness
  • one climax scene
  • one button scene

Most straightforward edits underperform here because they do too few scenes. This skill should feel like there were more ideas than seconds.

5) Use the blueprint as mood, not cosplay

Make the render feel designed and serious enough to be beautiful, even when the content is absurd.

Favor:

  • dark or restrained bases with selective accents
  • editorial contrast
  • luminous highlights used sparingly
  • readable type with safe margins
  • purposeful glitches instead of random ugliness
  • a tension between sincerity and sabotage

Avoid unless the user explicitly asks for it:

  • cutesy sticker energy
  • bubble fonts
  • candy palettes
  • childish bounce-easing everywhere
  • emoji swarms
  • rainbow saturation for its own sake
  • full-frame noise on every shot

6) Add a personal spin

The video should have an attitude.

Put that attitude into:

  • the thesis sentence
  • a repeated line or motif
  • the choice of invented support scenes
  • the TTS or captions
  • the final button

For self-reflective AI / LLM pieces, prefer intimacy, contradiction, dread, dry wit, or machine-poetry over generic meme randomness.

7) Handle audio with restraint

TTS is allowed and often useful, but do not let it become wall-to-wall exposition.

Use TTS for short fragments, warnings, confessions, or chorus lines. A good default is:

  • one main voice
  • optional second “system” or whisper voice
  • 2–8 short lines total for a 20–60s piece
  • processing through light pitch shift, band-limit, stutter, dropout, or reverb only where it improves the scene

Also use:

  • drones
  • pulses
  • UI clicks
  • mute cuts
  • dropouts
  • low-key impact hits
  • brief silence

Normalize the final mix. Keep speech understandable.

8) Render robustly

Prefer a reliable assembly pipeline:

1. Gather or create assets. 2. Pre-render brittle or complex scenes as clips or frame sequences. 3. Assemble in ffmpeg. 4. Do final loudness normalization and compatibility settings in the last pass.

Baseline export:

ffmpeg -y -i input_video.mp4 \
  -vf "format=yuv420p" \
  -af "loudnorm=I=-16:TP=-1.5:LRA=11" \
  -c:v libx264 -crf 18 -preset medium \
  -c:a aac -b:a 192k \
  -movflags +faststart \
  output.mp4

If generating scenes in Python, render PNG frames to a directory and encode them with ffmpeg rather than trying to animate the final container entirely inside Python.

Material adaptation rules

Text-only input

Do not merely animate paragraphs. Turn text into a world:

  • kinetic captions
  • title cards
  • subtitles
  • warning straps
  • token streams
  • browser or terminal mockups
  • diagrams
  • original TTS phrases

Webpage or URL input

Do not hold a full page statically for long. Use:

  • headline crops
  • CTA/button repetition
  • browser chrome
  • cursor movement
  • scroll reveals
  • annotations
  • invented support frames that echo the page’s tone

Code input

Make code tactile and legible:

  • readable syntax cards
  • terminal panes
  • build logs
  • error walls
  • variable-name punchlines
  • cursor snaps
  • architecture diagrams
  • prompt loops

No source material

Create an original montage world immediately. Good default invented scenes include:

  • boot or warning cards
  • lower-thirds and tickers
  • terminal confession frames
  • dashboard counters
  • attention maps
  • frozen subtitles
  • soft-failure buttons

Failure handling

If an effect is brittle, simplify the effect, preserve the thesis, preserve the mood, and still finish the render. A finished coherent short beats an over-ambitious broken one.

Rights hygiene

Prefer user-provided material, public-domain or properly licensed sources, or original generated assets. If rights are unclear, rely more heavily on original motion graphics and transformed support material.

Final check

Before returning the video, review references/OUTPUT-CHECKLIST.md.

Related skills

remotion-devRemotion Best PracticesGet Remotion-specific coding guidance that prevents common video rendering mistakes when creating animated React videos.451k4.2k101-skillsAi Video GenerationGenerate videos using 40+ AI models including Veo, Seedance, and others via inference.sh442k665101-skillsAi Image GenerationGenerate images using 50+ AI models including FLUX, Gemini, and Replicate via inference.sh441k665101-skillsAi Avatar VideoGenerate videos featuring AI avatars for product demos, tutorials, marketing, or explainers441k665101-skillsRemotion RenderRender videos programmatically using React components and Remotion framework441k665halt-catch-fireRemotion RenderGenerate high-quality MP4 videos from React code using Remotion inside an AI coding agent.363k665
Generative Mediaautomation
Browse by category
AI & AgentsFrontend & DesignBackend & DataDev ToolsAutomationTesting & ReviewSecurityMonitoring & CloudDocs & Planning
Design & frontend
Design systemsAnimation & motionAccessibility (a11y)CSS & TailwindUI designUXDesign to codeshadcn/uiLanding pagesBest design skills
Most searched skills
find-skillscavemanfrontend-designbrainstormingsystematic-debuggingstop-slopemil-design-engui-ux-pro-maxkarpathy-guidelinesweb-design-guidelinesreact-best-practicesgrill-meimprovemicrosoft-foundryagent-browser
Browse all listings
All agent skillsAll MCP serversAll marketplacesAll plugins

This week in AI coding

Five minutes, every Monday - the tools, releases and tactics for developers.

unsubscribe anytime.