[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-122":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":8,"htmlUrl":8,"language":9,"languages":8,"totalLinesOfCode":8,"stars":10,"forks":11,"watchers":12,"openIssues":13,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":14,"stars7d":13,"stars30d":15,"stars90d":14,"forks30d":14,"starsTrendScore":14,"compositeScore":16,"rankGlobal":8,"rankLanguage":8,"license":8,"archived":17,"fork":17,"defaultBranch":18,"hasWiki":17,"hasPages":17,"topics":19,"createdAt":8,"pushedAt":8,"updatedAt":20,"readmeContent":21,"aiSummary":22,"trendingCount":14,"starSnapshotCount":14,"syncStatus":23,"lastSyncTime":24,"discoverSource":25},122,"tribeV2_ViralAnalyser","amirmushichge\u002FtribeV2_ViralAnalyser","amirmushichge",null,"Python",180,44,154,1,0,12,4.96,false,"main",[],"2026-06-12 02:00:08","# TRIBE Review MVP\n\nPrivate local web app for analyzing short video ads with Meta TRIBE v2 and presenting the result as an editing-friendly review.\n\nThe app runs the official TRIBE v2 inference path, visualizes the predicted brain-response curve and heatmap, and adds a practical recommendation layer for comparing cuts and finding weak moments in the timeline.\n\n## Credits and sources\n\nApplication shell, interface, workflow layer, and practical editing wrapper:\n\n- [AI Pulse](https:\u002F\u002Fx.com\u002Fyouraipulse)\n- [Amir Mushich](https:\u002F\u002Fx.com\u002FAmirMushich)\n\nOfficial model and research sources:\n\n- [Meta AI publication](https:\u002F\u002Fai.meta.com\u002Fresearch\u002Fpublications\u002Fa-foundation-model-of-vision-audition-and-language-for-in-silico-neuroscience\u002F)\n- [TRIBE v2 model page](https:\u002F\u002Fhuggingface.co\u002Ffacebook\u002Ftribev2)\n- [TRIBE v2 GitHub repository](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Ftribev2)\n- [CC BY-NC 4.0 license](https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby-nc\u002F4.0\u002F)\n\nAll inference is run through the official TRIBE v2 model as released by Meta on Hugging Face under the CC BY-NC 4.0 license.\n\nThis repository is a non-commercial community prototype built around the official Meta TRIBE v2 model. TRIBE v2 is used under the CC BY-NC 4.0 license, and Meta remains the owner of the model, its weights, and associated research materials. This project is not affiliated with, endorsed by, sponsored by, or officially connected to Meta. It does not claim ownership over TRIBE v2, Meta research materials, model weights, brand assets, uploaded videos, or third-party materials. No sales, paid distribution, sublicensing, or commercial use are intended.\n\n## What it does\n\n- Reviews one uploaded video as a deep analysis.\n- Compares 2-4 uploaded versions side by side.\n- Shows a predicted response-over-time curve.\n- Shows a 3D brain activity visualization.\n- Explains the major brain zones used by the model.\n- Adds practical editing recommendations.\n- Exports JSON and PDF reports.\n- Uses a local Whisper speech layer for transcript\u002Ftiming hints.\n- Optionally uses Ollama for local recommendation copy rewriting when a supported local model is available.\n\n## What it is not\n\n- It is not an official Meta application.\n- It is not affiliated with, endorsed by, sponsored by, or officially connected to Meta.\n- It is not a commercial product.\n- It is not a guaranteed virality predictor.\n- It does not measure one specific real viewer.\n- It does not claim ownership over TRIBE v2, Meta research assets, uploaded videos, or any third-party materials.\n\nSee [NOTICE.md](NOTICE.md) for the non-commercial notice and official source links.\n\n## Main idea\n\nThe response graph is the primary tool.\n\nUse the curve first, then use the video player, brain map, and recommendation cards to understand what is happening around the marked timestamps.\n\n## Workflows\n\n### A. Compare 2-4 versions\n\nUse compare mode for several versions of the same creative. The goal is not to blindly pick one full video. The goal is to find the strongest sections across versions and use them as an editing map for the next cut.\n\n![Compare workflow](docs\u002Fassets\u002Fworkflow-compare.svg)\n\nTypical use:\n\n1. Upload 2-4 variants of the same ad\u002Fvideo idea.\n2. Compare the overlaid curves.\n3. Mark which version has the strongest hook, middle hold, transitions, and later useful section.\n4. Build a new edit from the best-performing blocks.\n5. Re-run the new edit against the current leader.\n\n### B. Improve one video\n\nUse solo mode when you only have one cut. Look for real dips in the graph, click the timestamp, inspect nearby frames, and test one edit at a time.\n\n![Solo workflow](docs\u002Fassets\u002Fworkflow-solo.svg)\n\nTypical edits:\n\n- Cut or shorten slow fragments.\n- Speed up a section that drags.\n- Move the main action or caption earlier.\n- Make the subject larger or clearer.\n- Remove visual clutter.\n- Add a new beat before the graph drops.\n\nFull workflow notes: [docs\u002FWORKFLOWS.md](docs\u002FWORKFLOWS.md)\n\n## Project structure\n\n```text\n.\n|-- app.py                         # FastAPI app and report routes\n|-- bootstrap_models.py            # First-launch dependency and model preparation\n|-- tribe_runtime.py               # TRIBE v2 model loading and inference wrapper\n|-- official_report.py             # Official-output report layer\n|-- review_engine.py               # Local recommendation and comparison logic\n|-- brain_visualization.py         # Brain heatmap visualization data\n|-- report_localization.py         # UI\u002Freport copy layer\n|-- pdf_report.py                  # Chrome-based HTML-to-PDF export\n|-- templates\u002Findex.html           # Main web UI\n|-- static\u002Fvendor\u002F                 # Local browser dependencies\n|-- runtime_media\u002F                 # Local runtime uploads\u002Freports, ignored by Git\n|-- docs\u002FINSTALL_WINDOWS.md        # Windows setup guide\n`-- docs\u002FTROUBLESHOOTING.md        # Common issues\n```\n\n## Quick start on Windows\n\nFor non-technical users:\n\n1. Click the green `Code` button on GitHub.\n2. Click `Download ZIP`.\n3. Extract the ZIP into a normal folder, for example `Desktop` or `Downloads`.\n4. Open the extracted folder.\n5. Double-click `Start_TRIBE_Review.cmd`.\n6. Keep the black terminal window open while the app prepares itself.\n7. When the terminal says setup is complete, close it.\n8. Double-click `Start_TRIBE_Review.cmd` again.\n9. The app should open in your browser.\n\nThe first launch can take a while because the app downloads and installs everything it needs. Later launches are much faster.\n\nQuick command version:\n\n```powershell\nStart_TRIBE_Review.cmd\n```\n\nIf the browser does not open automatically, open:\n\n```url\nhttp:\u002F\u002F127.0.0.1:8000\n```\n\nFull setup notes: [docs\u002FINSTALL_WINDOWS.md](docs\u002FINSTALL_WINDOWS.md)\n\n## Requirements\n\nRecommended local setup:\n\n- Windows 10\u002F11 64-bit\n- Python 3.11\n- 16 GB RAM\n- Modern 8-core CPU or better\n- NVIDIA GPU\n- 6 GB VRAM minimum, 12 GB+ preferred\n- 30 GB+ free disk space, preferably on SSD\n\n## Runtime data and privacy\n\nUploaded videos and generated reports are written to `runtime_media\u002F`.\n\nThat folder is intentionally ignored by Git. Do not commit runtime media, reports, transcripts, tokens, logs, model weights, or cache folders.\n\n## Development checks\n\nRun a syntax check:\n\n```powershell\npython -m py_compile app.py bootstrap_models.py tribe_runtime.py speech_runtime.py official_report.py review_engine.py report_localization.py pdf_report.py brain_visualization.py runtime_setup.py\n```\n\nRun a smoke test with a local video:\n\n```powershell\npython smoke_test.py C:\\path\\to\\test-video.mp4\n```\n\n## License and use\n\nNo open-source license is granted yet. Treat this repository as private, all-rights-reserved, non-commercial evaluation code unless a license is explicitly added later.\n\nReview the official TRIBE v2 license and all third-party licenses before any redistribution or public use.\n\nTRIBE v2 model weights are not included in this repository. They are downloaded from the official Hugging Face model page by the local setup flow and remain subject to the official TRIBE v2 license.\n","该项目是一个用于分析短视频广告的私有本地Web应用程序，基于Meta TRIBE v2模型进行预测并提供编辑友好的审查结果。其核心功能包括运行TRIBE v2官方推理路径、可视化预测的大脑响应曲线和热图，并提供实用的剪辑建议来比较不同版本和发现时间线中的薄弱环节。此外，它还支持多版本视频对比、3D大脑活动可视化、主要大脑区域解释以及JSON和PDF报告导出等功能。该应用适用于需要深入分析和优化短视频广告效果的场景，如营销团队在制作或调整广告内容时使用。",2,"2026-06-11 02:30:58","CREATED_QUERY"]