Stack
Everything that touches a finished video, and what each step costs. The list holds the tools that are actually in use, not the ones we are still testing. Prices are either the public list price or our own measured per unit cost.
No affiliate links exist yet. If one ever does, the row says so.
| Stage | Tool | Cost |
|---|---|---|
| Transcript | OVH AI Endpoints, Whisper-large-v3 / turboDefault speech-to-text with word timings, `pipeline/slice1.py` | credits |
| Transcript | yt-dlpPulls existing captions for free (no model call) when the source ships them; its `license` field is the mechanical gate for CC-BY reuse | free |
| Script, on-screen text | OVH AI Endpoints, Qwen3.5-397B-A17BText that ends up on screen: hooks, narration lines, key cards | credits |
| Script, bulk text | OVH AI Endpoints, gpt-oss-120bBulk low-stakes text: title variants, caption chunking | credits |
| Script, hook picking | OVH AI Endpoints, Meta-Llama-3.3-70B-InstructCheap hook-picking JSON, wired in `slice1.py` | credits |
| Narration | OVH Riva TTS, `nvr-tts-en-us` (English-US.Male-Calm)Free English narration voice, `pipeline/tts.py` | free |
| Screen capture | Simon's own screen recordingsPrimary b-roll: real terminal runs, real diffs, real pricing pages, target 60% of runtime | free |
| Screen capture | PlaywrightScripted shots (pricing pages, terminal runs) against a URL allow-list | free |
| Screen capture | Capturo (`DiverseProjects/capturo`)Existing Playwright-record, Remotion-compose, ffmpeg-export engine; produced 64+ videos for ShipMate, Printora, AkunIndo | free |
| B-roll | Pexels APIStock footage, last-resort gap-fill | free |
| B-roll | Pixabay APISame fallback role | free |
| Stills, key cards, thumbnails | OVH SDXL Base 1.0Key cards, stills, thumbnails | free |
| Music and sound | Royalty-free bed (fixed, ducked to 0.2) or `NO_MUSIC=1`The one music bed already wired in `render_narrated.py` | free |
| Repost format: PIP, split-screen, reaction | ffmpeg `overlay`/`hstack` filtersStatic picture-in-picture, split-screen and reaction composition in one filter-graph line | free |
| Render and captions | ffmpegLocal render: ASS captions from Whisper word timings, `loudnorm` audio | free |
| Posting and storage | OVH Object StorageMedia library for everything produced | credits |
| Ops, team | TimeCampEko logs hours against the real GitHub issue, not "General Task" | paid |
| Total | 17 tools in use, no affiliate link on any of them |