IdeaWell

Project ideas from Hacker News discussions.

Show HN: Weedout – Safari extension that hides YouTube AI-labeled videos

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Three Prevalent Themes in the Discussion

  1. Desire to filter or block AI‑generated content
    Users repeatedly expressed a wish to keep AI‑labeled videos out of their feeds to improve the YouTube experience.
  2. “I wish I had this but for every website and media surface” – thatmf
  3. “Well, AI flags this content so it's an easy target without heuristics - that's why I've started there.” – masteranza
  4. “Having the option to choose whether or not to see AI content was the first thing I hoped they'd add a filter for when it was first introduced.” – MarioMan

  5. “Finally, a way to reclaim my YouTube feed from the AI sludge.” – vist_orn

  6. Criticism of YouTube’s AI labeling as inaccurate or overly broad
    Many commenters pointed out that the “Made with AI” label is too general, often misapplied, and fails to distinguish between trivial AI use and genuine AI‑slop.

  7. “I’ve found that the AI label doesn’t necessarily mean AI slop, it just means there’s a non zero amount of AI in the video.” – Schnitz
  8. “Yep, it's super useless. The label is so general. Was AI used for one image in the 10 minute video? Is the voiceover AI? Is the entirety of it generated with AI? No way to know.” – f3408fh
  9. “One problem is that algorithms(1) to detect AI generated content are surprisingly bad… non‑AI generated animated/cartoon content is much more likely to be mistakenly labeled as AI…” – dathinab

  10. “I think instead of ‘AI or not’, that something more nuanced is needed – perhaps a rating system inspired by the MPAA's…” – AmazingEveryDay

  11. Preference for human‑created content, especially aversion to AI voiceovers
    Several participants said they instantly reject videos with AI‑generated narration, valuing human voices even if imperfect, and view AI voice work as a quality signal.

  12. “My pet peeve is the AI generated voices… once I detect that the voiceover is AI generated that shit get instablocked forever.” – karim79
  13. “Many content creators on YouTube may not have English as their first language… I'd rather listen to broken English read by a human than an AI voice over.” – mrweasel
  14. “Depends. I take an AI generated voice over a french guy trying to talk english every day! Same for japanese… I really try, but I find these two speaker types really hard to follow and stay alert.” – lynx97

🚀 Project Ideas

AIFilterX – Cross-browser Granular AI Content Blocker

Summary

  • A browser extension that lets users filter YouTube (and other video platforms) by specific AI‑generated components: voiceover, visuals, thumbnails, metadata, etc.
  • Provides toggles for each AI type and a community‑sourced whitelist/blacklist to improve accuracy beyond YouTube’s coarse “Made with AI” label.
  • Core value: gives fine‑grained control over what AI content you see, reducing noise while preserving legitimate AI‑assisted videos.

Details

Key Value
Target Audience YouTube viewers who dislike AI slop but accept limited AI use (e.g., voice‑only, thumbnail‑only)
Core Feature Per‑category AI filters (voice, visual, thumbnail, metadata) + custom rules + sync across Chrome, Firefox, Safari
Tech Stack WebExtension Manifest V3, Options page in React/Vue, optional Rust/WASM module for local heuristic scoring, Web Storage API for sync
Difficulty Medium
Monetization Revenue-ready: Freemium (free basic filters, $2.99/mo for premium rule sync, cloud‑based community lists, and advanced analytics)

Notes

  • Commenters asked for “more nuanced… a rating system” and the ability to block only certain AI uses (karim79, masteranza, Amazin­gEveryDay).
  • Cross‑browser demand: several users wanted Firefox/Safari ports (ssernikk, lukeasch21, masteranza).
  • Community‑sourced lists already appreciated (mschuster91’s AiSList); AIFilterX would extend that idea with per‑type tagging.

EchoGuard – On‑device AI Voice Detector & Blocker

Summary

  • A lightweight client‑side tool that runs in the browser (via Service Worker) or as a desktop app to detect synthetic speech in audio tracks and automatically mute, lower volume, or flag the segment.
  • Operates completely offline, preserving privacy and avoiding reliance on platform labeling.
  • Core value: instantly removes the most off‑putting AI element—AI voiceovers—while leaving human‑narrated content intact.

Details

Key Value
Target Audience Listeners of podcasts, educational videos, and YouTube content who find AI voices irritating or untrustworthy
Core Feature Real‑time synthetic speech detection using a compact ML model; configurable actions (mute, volume duck, visual indicator)
Tech Stack TensorFlow.js / ONNX Runtime in a Service Worker, WebAssembly‑accelerated model (e.g., Whisper‑based voice‑activity + CNN classifier), IndexedDB for model cache
Difficulty High (model optimization, real‑time audio pipeline)
Monetization Hobby (open‑source) – can accept donations or offer a $4.99 one‑time “pro” model update pack for enhanced accuracy

Notes

  • Multiple commenters expressed instant aversion to AI voices (karim79, mrweasel, ungreased0675) and wished for a “small local model that detects AI voices and blocks the content.”
  • Privacy‑focused users would appreciate client‑side detection that never sends audio to external servers.
  • The approach sidesteps the inaccuracies of YouTube’s AI label (dathinab, f3408fh) by doing its own voice‑synthesis detection.

AIQ – Community AI Content Rating System

Summary

  • A platform where users rate videos on multiple AI‑usage dimensions (e.g., voice AI, visual AI, script AI, music AI) using a simple 0‑5 scale; ratings are aggregated and displayed as an AIQ badge.
  • Browser extension reads the badge and lets users filter or sort by AIQ thresholds (e.g., hide anything >3 on visual AI).
  • Core value: replaces the vague “Made with AI” label with a transparent, crowd‑sourced rating system akin to MPAA ratings, enabling nuanced preferences.

Details

Key Value
Target Audience Consumers who want detailed insight into how AI was used in a video before deciding to watch
Core Feature Per‑dimension AI ratings (voice, visual, audio, text) aggregated per video; extension overlay shows AIQ score and offers filter controls
Tech Stack Firebase Firestore for rating storage, Cloud Functions for aggregation, WebExtension (MV3) to inject scores and UI, optional GraphQL API for third‑party apps
Difficulty Medium
Monetization Revenue-ready: Free tier with basic ratings; $1.99/mo for premium features (custom weight‑sets, export of personal rating history, ad‑free experience)

Notes

  • AmazingEveryDay suggested an MPAA‑style rating system for AI content; AIQ directly implements that idea.
  • Users complained about the binary label being “super useless” (f3408fh) and wanted to know what part of the video used AI (Schnitz, PixelForg).
  • Community‑driven approach mirrors the enthusiasm for shared lists (mschuster91) while adding granularity that could reduce false positives/negatives.

Read Later