Connect the bundle
Choose Codex or Claude and configure your channel, tools and branding.
AI VIDEO EDITING SKILL
Editing for YouTube is a bundle of instructions and tools for editing YouTube videos with Codex or Claude on Windows. It helps analyze footage and prepare speech, inserts and animation; you approve the plan and sample.
See before deciding
First: a 55-second showcase of version 4.1.0, with Russian on-screen text. The author’s project used 7.6 hours of source footage to produce a 19:36 video with 81 graphics layers. These are results from one project, not a promise for every video. The other visual examples were made with 4.0.0.
Skills by Paul Wise · Spec Skills. Demo videos supplied by the author.
The skill keeps your lasting instructions in your workspace. Its rules preserve the faces of the video’s creators; you review the final edit and rights to music and materials.
Choose Codex or Claude and configure your channel, tools and branding.
The AI reviews footage and proposes an opening, structure and inserts.
Review 60–90 seconds, then approve the full video.
For reviews, screencasts, news and talking-head videos. You approve the plan and review the footage. Some local models need a compatible GPU.
The essentials are visible. Expand the complete capability list.
Whisper word timestamps with two-model checks; synchronize a separate microphone with the camera and check drift.
Remotion chapter cards, news cards, name/role lower thirds, counters, word-timed lists, charts, quotes, transitions and HUD panels.
YouTube H.264/BT.709 export with faststart; checks for black/frozen frames, audio gaps and levels.
Inventory files, duration, audio tracks and creator footage. Propose openings, structure, chapters, inserts, graphics and effects. Agree on the plan before rendering, then review a 60–90-second sample.
Whisper word timestamps with two-model checks; synchronize a separate microphone with the camera and check drift. Find repetitions, false starts, slips and failed takes, prepare listening clips, shorten pauses naturally without speeding speech and check that unique ideas survive the edit.
Basic filtering and light noise reduction, smooth audio joins, loudness normalization and peak checks. Duck music under speech and use your music/SFX folders. Optional Demucs separates voice/background; DeepFilterNet handles stronger noise.
Face tracking, smooth zooms and framing changes. Phrase-based visual plans, own-footage search and Pexels/Pixabay material with source/license records. Check shake, repetition and scene drift; blur unrelated logos and private data.
Remotion chapter cards, news cards, name/role lower thirds, counters, word-timed lists, charts, quotes, transitions and HUD panels. Speech-synced kinetic titles. Shine, glow, electricity, glitch, typewriter, decode, scanner, drawn lines and particles; image pans, zooms, detail highlights, before/after comparisons, grids and parallax.
Optional Qwen3‑VL‑4B describes sampled frames and indexes B-roll for content search; ask questions about images. Optional tools make object masks and depth maps, support parallax and titles behind the creator, and generate FLUX illustrations through ComfyUI. Surface-tracked text needs a suitable tool and license; commercial CoTracker use requires separate permission.
A large canvas with smooth camera moves between ideas, a route and an overview. Paper style: sketch-to-color, drawn diagrams, charts and labels. Technology style: blueprint grid, laser engraving, layered reveals, LED text and pulses. Elements follow spoken words; approve the storyboard and sample chapter first.
Rules for screencasts, gadget reviews, news and talking heads. On request, Shorts from strong passages with vertical 9:16 framing. Optional Chatterbox supports creator-voice dubbing, phrase timing and back-transcription checks.
Configure the channel, formats, colors, font and signature; review sample graphics and a test edit. Lasting revisions go into a personal rules file. Preserve creators’ faces. Change the primary chat language independently of the video language. Includes Codex/Claude packages, Windows installer, documentation and license table.
YouTube H.264/BT.709 export with faststart; checks for black/frozen frames, audio gaps and levels. Transcript-derived chapters, three titles and three real-footage thumbnails. Timecoded revisions and rebuilding changed parts. After publication, compare retention with the edit and formulate a testable hypothesis.
14 additional Remotion compositions: word-timed kinetic titles, a glowing counter, shine/glitch callouts, A/B matchups, three alternatives, round scoreboards, stamps, measurement and verdict tables, dialogue bubbles and end screens. Android/ADB diagrams also ship with demonstration content to adapt to your project. Audio: new breath.py attenuates quiet breaths and clicks between words in a separate WAV, protecting speech boundaries. An optional energy guard is available. Very short recordings and phase-opposed stereo are handled correctly. For intrusive noise, the skill first proposes an approved DeepFilterNet before/after sample. Fewer unnecessary speech cuts: repetition checks account for lists, alternatives and differing before/after numbers. A nonadjacent repeated single word does not automatically trigger removal. Long words and uncertain cuts become listening candidates. Screencasts and cameras: gentle pushes normally up to 110%, focus on the readable window and overlays clear of important text. Silent screen recordings require event-based alignment and drift checks. Zoom windows correctly follow video offsets relative to the microphone. Animation stays word-timed after pause removal by mapping source timestamps to the edit. Parallel jobs use isolated temporary names; push caches include FPS, and detected render failures prevent an output being marked complete. Local insert search retains optional Qwen3-VL-4B frame descriptions, a B-roll text index, readable tables, query search and people/no-people filters. Searching a completed index does not load the model.
Optional AI components require approval for installation and use. Capabilities depend on the host application’s tools, your computer and licenses.
A video-editing skill for Codex on your computer, with instructions for working with files and tools in the Codex environment.
The same editing rules, with separate instructions for computer access and tools in the Claude environment.
39.99 € · Euro payments with Stripe are not available yet. Submit a request to be notified; no payment will be taken.
COMMUNITY / SPEC SKILLS
Reviews submitted on Spec Skills are moderated. No account is needed; an on-site rating alone does not verify a purchase.
Digiseller reviews are published automatically after purchase verification and synced hourly. Positive feedback is shown as 5 stars and negative feedback as 1 star. Reviews on the two platforms count separately.
5.0 / 5 · 1
Digiseller buyer ★★★★★
Digiseller · verified purchase
Супер, работает круто, я раньше столько времени тратил на монтаж видео и делал хуже, а теперь все просто и быстро, спасибо Павлу и Клоду.
Added to Spec Skills · Original text
PROMPTS / 3
Start by adapting the skill to your channel.
Establish your style, tools and materials for a first sample edit.
Fill in the fields in < >. The AI first reviews your footage and proposes a plan; the sample and full edit follow your approval.
For a configured skill: analysis and suggestions → your approval → a 60–90 second sample.
39.99 €. Euro payments with Stripe are not available yet. Submit a request to be notified; no payment will be taken.
It is a pair of AI skills and a toolkit. You need Codex or Claude with skill support and appropriate access to your computer.
Yes. First-time setup covers your channel, branding, software and materials, and proposes sample graphics and a test edit.
This page is assigned the English archive; the Russian page uses Russian. Change and save the primary chat language independently of the spoken-video language.
After successful installation and initial setup, the skill invites you once to leave an honest rating and comment. No account is needed; reviews appear only after moderation. Declining does not affect the skill.
Optional Qwen3‑VL‑4B analyzes sampled frames on your PC and creates a B-roll text index for finding inserts. Models and tools must first be downloaded using internet access. Local components do not make the whole workflow offline: data processing by the main AI assistant depends on its application and settings.
A GPU is not required for basic editing. Some optional frame-analysis, illustration and dubbing models need a compatible NVIDIA GPU and enough VRAM. The installer checks the environment; components are chosen with the user.
The bundle price excludes AI-application subscriptions and any third-party tool licenses. Remotion has specific commercial-team terms; CoTracker has a non-commercial license. Review the license table and terms before use.
Editing for YouTube · v4.1.0 · 39.99 €
The two latest versions are shown first. Expand earlier changes below. Downloads use only the current version.
Windows 10/11 64-bit, at least 8 GB RAM (16 GB recommended), around 10 GB of free space plus project files. All optional AI components may require around 70 GB. A GPU is optional for basic editing; some components require NVIDIA and sufficient VRAM.
The Windows installer places tools in your workspace. In its normal mode, if Python or Node.js is missing, it offers installation with your consent. Internet access is needed to obtain tools and models.
Remotion is free for individuals and teams of up to three, including commercial videos. Larger commercial teams generally need a paid license; exceptions and special rules for joint projects apply. Check the current Remotion terms and other tools’ licenses. CoTracker is non-commercial; commercial tracking requires another tool or separate permission.
Check Remotion terms ↗ ↗