Skip to main content

video-editing

Video editing and delivery on codeonym's Arch workstation with Kdenlive 26.08 (MLT/melt), ffmpeg 9 with AMD VAAPI hardware encoding (H.264/HEVC on the Vega iGPU), and Blender's Video Sequencer as a backup. Use when the user wants to edit a video (tutorial, demo, vlog, promo, reel), cut/trim/concatenate clips, add titles/subtitles/music, fix or normalize audio, export for YouTube/LinkedIn/Instagram/TikTok/web, make GIFs, pull thumbnails, or render a Kdenlive project from the terminal.

Quellinformationen

Repository
codeonym-org/codeonym-arch-design
Letzte Quellaktivität
25. September 2026 um 00:47
Erkannte Sprache von SKILL.md
Englisch
Sterne
0
Forks
0

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

Datei-Explorer
9 Dateien

SKILL.md wird angezeigt

SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
video-editing
description
Video editing and delivery on codeonym's Arch workstation with Kdenlive 26.08 (MLT/melt), ffmpeg 9 with AMD VAAPI hardware encoding (H.264/HEVC on the Vega iGPU), and Blender's Video Sequencer as a backup. Use when the user wants to edit a video (tutorial, demo, vlog, promo, reel), cut/trim/concatenate clips, add titles/subtitles/music, fix or normalize audio, export for YouTube/LinkedIn/Instagram/TikTok/web, make GIFs, pull thumbnails, or render a Kdenlive project from the terminal.
# Video editing (Kdenlive · ffmpeg · melt) ## Toolkit on this machine | Tool | Use it for | Launch / CLI | |---|---|---| | **Kdenlive 26.08** | Multi-track editing: cuts, titles, transitions, effects, color, audio mixing, subtitles | `kdenlive` · headless `scripts/render-project.sh` | | **melt 7.40** (MLT) | Kdenlive's render engine, scriptable | `melt` | | **ffmpeg 9 (`/usr/bin/ffmpeg`)** | Encode, convert, trim, loudness, GIFs, VAAPI hardware encoding | `/usr/bin/ffmpeg` | | **Blender VSE** | Backup editor; also for 3D/motion-graphic inserts (`3d-modeling` skill) | Blender → *Video Editing* workspace | | **OBS** | Recording source footage | `screen-recording` skill | | **GIMP / Inkscape** | Thumbnails, lower-thirds, title cards as PNG with alpha | `graphic-design` skill | **Machine quirks** - Plain `ffmpeg` in your shell is **Homebrew's (no VAAPI)**. Scripts here force `/usr/bin` first; in manual commands type `/usr/bin/ffmpeg`. - The Vega iGPU hardware-encodes **H.264 and HEVC only** (no AV1 encode, despite `av1_vaapi` being listed). - Hardware encoding is fast but makes bigger files than x264 at equal quality — use `--sw` in `encode.sh` for the smallest final uploads. ## Scripts | Script | What it does | |---|---| | `check.sh` | Kdenlive, MLT, frei0r effects, ffmpeg builds + VAAPI encoders, shadowing warnings | | `encode.sh <in> <preset> [out] [--sw]` | Presets: `web`, `youtube`, `vertical` (1080×1920 blurred fill), `square`, `archive` (HEVC), `preview` | | `loudnorm.sh <in> [--target -14] [--denoise]` | Two-pass EBU R128 loudness normalization (video copied untouched), optional noise reduction | | `gif.sh <in> [--start S] [--dur D] [--width W] [--fps F]` | Palette-optimized GIF + a far smaller WebM for READMEs/docs | | `thumbs.sh <in> [count]` | Evenly spaced frame grabs + storyboard sheet for picking a thumbnail | | `render-project.sh <project.kdenlive> [out.mp4]` | Renders a Kdenlive project with melt, no GUI | ## Guides - `references/kdenlive-workflow.md` — project setup, proxy clips, the edit (J/K/L, razor, ripple), titles, color, audio, subtitles, render. - `references/delivery-specs.md` — per-platform specs, loudness targets, and the ffmpeg one-liners worth knowing. ## Rules of thumb 1. Project profile = your footage (usually 1080p 30 or 60 fps). Enable **proxy clips** for 4K or long screen recordings on this iGPU. 2. Story first: rough cut → fine cut → B-roll/zooms → titles → color → audio → export. 3. Audio matters more than video: normalize to **−14 LUFS** (YouTube/social) or **−16 LUFS** (voice/podcast). 4. Keep masters (`archive` preset or Kdenlive lossless), deliver platform versions from them. 5. Burn captions for social (most people watch muted); upload `.srt` separately for YouTube.
Auf GitHub ansehen