{"id":50754553,"url":"https://github.com/Apolotary/blob-tracker","last_synced_at":"2026-06-28T07:00:26.973Z","repository":{"id":355479019,"uuid":"1228154602","full_name":"Apolotary/blob-tracker","owner":"Apolotary","description":"Hermes agent skill that does blob tracking videos","archived":false,"fork":false,"pushed_at":"2026-05-03T19:42:38.000Z","size":2665,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-05-08T23:39:58.702Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/Apolotary.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-05-03T17:04:07.000Z","updated_at":"2026-05-03T19:42:42.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/Apolotary/blob-tracker","commit_stats":null,"previous_names":["apolotary/blob-tracker"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/Apolotary/blob-tracker","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Apolotary%2Fblob-tracker","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Apolotary%2Fblob-tracker/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Apolotary%2Fblob-tracker/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Apolotary%2Fblob-tracker/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/Apolotary","download_url":"https://codeload.github.com/Apolotary/blob-tracker/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Apolotary%2Fblob-tracker/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34880189,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-28T02:00:05.809Z","response_time":54,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2026-06-11T04:00:24.044Z","updated_at":"2026-06-28T07:00:26.968Z","avatar_url":"https://github.com/Apolotary.png","language":"Python","funding_links":[],"categories":["Projects, Demos \u0026 Installations"],"sub_categories":["Design, Drawing \u0026 Web Surfaces"],"readme":"# blob-tracker\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"assets/logo.png\" alt=\"blob-tracker\" width=\"640\"\u003e\n\u003c/p\u003e\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"assets/trailer-preview.gif\" alt=\"blob-tracker preview — bee on flower with spatial-echo viz\" width=\"360\"\u003e\n  \u003cbr\u003e\n  \u003cem\u003esimple-blob detector + bbox + network + spatial-echo (each box shows pixels mirrored from elsewhere)\u003c/em\u003e\n\u003c/p\u003e\n\nA **Hermes Agent skill** (also Claude-Code-skill compatible) that renders any\nvideo with audio-reactive blob tracking in **16 detector flavors** and\n**15 visualization flavors** — combine freely. The skill fills in any\nmissing input on demand: bring your own video and audio, or have it find\na public-domain clip on the Internet Archive and compose an ambient\nsoundtrack.\n\nBuilt for the **Nous Research / Kimi Creative Hackathon (May 2026)**,\nKimi Track. All LLM/vision calls go to Kimi (Moonshot AI) `kimi-k2.6`.\n\n## Highlights\n\n- **16 detector flavors** — `motion-diff` · `mog2` · `knn` · `flow` ·\n  `color-hsv` · `color-cluster` · `simple-blob` · `dog` · `circles` ·\n  `saliency-fine` · `saliency-spec` · `csrt` · `edge` · `accumulation` ·\n  `watershed` · `contour-area`\n- **15 visualization flavors** — `bbox` · `corner-ticks` · `crosshair` ·\n  `centroid-trail` · `network` · `letters` · `emojis` (color-emoji\n  particles via PIL) · `glyphs` · `cctv-zoom` · `silhouette` · `outline` ·\n  `voronoi` · `convex-hull` · `heatmap` · `spatial-echo` (each blob bbox\n  shows pixels sampled from elsewhere)\n- **13 optional glitch postfx** — chromatic aberration, NTSC chroma\n  desync, kolorizer LUT, sync jitter, ripple, mosaic, scanlines, slit\n  scan, feedback, lagfun, threshold bands, edge glow, luma rotate\n- **Audio reactivity** — RMS / kick / high / onset features modulate\n  every visualizer's intensity per frame\n- **Auto-fill missing inputs** — bring `--video` + `--audio`, or use\n  `--find-video` to search the Internet Archive, or `--compose-music`\n  to synthesise a soundtrack, or `--brief \"...\"` for full-auto\n- **Kimi for creative judgement** — query expansion, vision-based clip\n  picking, music key/mode/bpm director, optional `--auto-flavor`\n  detector + viz selection from a frame thumbnail\n\n## Install\n\nDirect from this repo, into Hermes:\n\n```bash\nmkdir -p ~/.hermes/skills/creative\ngit clone https://github.com/Apolotary/blob-tracker.git \\\n    ~/.hermes/skills/creative/blob-tracker\n\npip install -r ~/.hermes/skills/creative/blob-tracker/requirements.txt\n```\n\nFor Claude Code, swap the path:\n```bash\ngit clone https://github.com/Apolotary/blob-tracker.git \\\n    ~/.claude/skills/blob-tracker\n```\n\n`ffmpeg` must be on your `$PATH` (`brew install ffmpeg`).\n\nThe skill uses `opencv-contrib-python` for the saliency + CSRT-tracker\ndetectors. If you already have plain `opencv-python` installed, replace\nit: `pip uninstall opencv-python \u0026\u0026 pip install opencv-contrib-python`.\n\n## Run\n\n```bash\nexport MOONSHOT_API_KEY=sk-...        # only when using --brief / --compose-music / --auto-flavor\n\n# 0. Grab a starter pack of public-domain demo clips (240 MB total)\npython scripts/fetch_demos.py\n# → ~/blob-tracker/demos/{bee-city,fireworks-vlog,...}.mp4\n\n# 1. Pure: existing video + existing audio\npython scripts/render.py --video clip.mp4 --audio track.wav \\\n    --detector mog2 --viz centroid-trail,network --output out.mp4\n\n# 2. Compose music for an existing video\npython scripts/render.py --video clip.mp4 --compose-music \\\n    --music-brief \"warm contemplative ambient pad\" \\\n    --viz silhouette,outline --output out.mp4\n\n# 3. Find video on IA, bring your own audio\npython scripts/render.py --find-video \"lunar surface NASA\" \\\n    --audio mytrack.wav --viz heatmap,corner-ticks --output out.mp4\n\n# 4. Full auto from a creative brief\npython scripts/render.py --brief \"1920s botanical, dreamy ambient pad\" \\\n    --auto-flavor --dual-format\n\n# 5. The \"spatial echo\" — blob bbox shows mirrored content from elsewhere\npython scripts/render.py --video clip.mp4 \\\n    --detector mog2 --viz spatial-echo,corner-ticks \\\n    --viz-params '{\"spatial-echo\":{\"mode\":\"rotate\",\"time_shift_frames\":12}}' \\\n    --output out.mp4\n\n# 6. Quick-iteration preset: 480 px, 8s — done in ~2s\npython scripts/render.py --quick --video clip.mp4 --detector mog2 \\\n    --viz centroid-trail,network --output preview.mp4\n```\n\n### Performance flags\n\n| Flag | What it does | Typical speedup |\n|---|---|---|\n| (default — pipes BGR straight to ffmpeg, no PNG round-trip) | always on | ~4× over a write-PNGs-then-encode pipeline |\n| `--detect-scale 0.5` | run detection at half resolution; viz still at full | 2–3× for slow detectors (`flow`, `csrt`, `saliency-*`, `watershed`) |\n| `--quick` | preset: 480 px, 8 s, fast iteration | ~30× vs full 1080² 26 s render |\n\nReference render times on Apple M1 / 8 GB at 1080² × 26 s:\n\n- `motion-diff` + default viz: **~12 s**\n- `flow` + viz, `--detect-scale 0.5`: **~40 s**\n- `--quick` (any detector): **~2 s**\n\nOutput goes to `--output` (single mp4) or to\n`$BLOB_OUT_DIR/\u003cslug\u003e/` (multi-file run with intermediate `winner.json`,\n`audio.json`, `audio.wav`, etc.). Default `BLOB_OUT_DIR` is\n`~/blob-tracker`.\n\n## Use in Hermes Agent chat\n\nOnce installed, the skill becomes a slash command:\n\n```\n/blob-tracker --video clip.mp4 --detector mog2 --viz centroid-trail,network\n```\n\nOr in natural conversation:\n\n```\nhermes chat --toolsets skills -q \"blob-track this clip with the heatmap viz\"\nhermes chat --toolsets skills -q \"find a NASA clip and make a media-art piece\"\n```\n\nFull install paths and Hermes config in [`INSTALL.md`](INSTALL.md).\n\n## Modules\n\n| Script | Uses Kimi | What it does |\n|---|---|---|\n| `scripts/render.py` | text+vision (opt) | Main entry. Composes everything. |\n| `scripts/fetch_demos.py` | — | Download the public-domain starter pack. |\n| `scripts/detectors.py` | — | Registry of 16 detector flavors. |\n| `scripts/visualizers.py` | — | Registry of 15 viz flavors. |\n| `scripts/postfx.py` | — | Registry of 13 optional glitch primitives. |\n| `scripts/audio_features.py` | — | RMS / kick / high / onset extractor. |\n| `scripts/video_search.py` | text+vision | IA search + Kimi pick. Standalone CLI. |\n| `scripts/compose_music.py` | text | Brief → music spec → 5-layer numpy synth. |\n| `scripts/prepare_source.py` | — | Download + energy-scout + square-crop. |\n| `scripts/kimi_client.py` | — | OpenAI-compatible Kimi client. |\n| `scripts/make_logo.py` | — | Generate the project logo by running the skill on itself. |\n\nEach helper is independently CLI-invokable for debugging or one-off use.\n\n## Why Kimi for the creative decisions\n\n- **Picking a clip from search results** is *taste*, not retrieval —\n  Kimi's vision model judges six tiny IA thumbnails and picks the one\n  that will yield rich 26 seconds.\n- **Choosing musical mode/bpm from a brief** is a small-but-creative\n  judgement; `kimi-k2.6` does it well at sub-cent cost per call.\n- **Auto-flavor detector + viz selection** lets Kimi-vision look at\n  a frame thumbnail and recommend the right blob-tracking parameters\n  for that footage.\n\nAll Kimi calls go through `kimi_client.py` — swap the model ID via\n`KIMI_MODEL=...` env if you want to try a different K2 variant.\n\n## Reference docs\n\n- [`references/detector-flavors.md`](references/detector-flavors.md) —\n  every detector with parameters and recommended use case\n- [`references/viz-flavors.md`](references/viz-flavors.md) —\n  every visualization with parameters and combination tips\n- [`references/postfx-glossary.md`](references/postfx-glossary.md) —\n  glitch primitive parameters and audio mappings\n- [`references/audio-design.md`](references/audio-design.md) —\n  how the 5-layer ambient synth is built and what Kimi's mood map controls\n- [`references/demo-clips.md`](references/demo-clips.md) —\n  bundled demo clips + Pixabay/Pexels/Coverr links + yt-dlp tips\n\n## Hackathon submission\n\n- The repository **is** the skill. Drop into `~/.hermes/skills/creative/`\n  and the agent has a new `/blob-tracker` slash command.\n- All footage in any output is public-domain (Internet Archive\n  Prelinger / NASA / NARA collections by default).\n- All AI calls go to **Kimi (Moonshot AI)** — query expansion, vision\n  picks, music director, auto-flavor selection.\n\n## License\n\nMIT — see [`LICENSE`](LICENSE). Reusing this code requires retaining\nthe copyright and license notice.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FApolotary%2Fblob-tracker","html_url":"https://awesome.ecosyste.ms/projects/github.com%2FApolotary%2Fblob-tracker","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FApolotary%2Fblob-tracker/lists"}