HumanStamp
A creator tool that embeds an invisible 'made by human' watermark into your photos and posts — AI vision models see a decoy while your real audience can verify the work is human-made.
Photographers, illustrators, and writers who want to prove their work is human-made in the AI era
- Invisible motion watermark that survives JPEG compression and resizing
- Browser extension that verifies whether images you encounter online are human-made or AI-generated
- 'Human-made' badge exportable to Instagram, X, and portfolio sites
- Tracker showing where your watermarked images appear online and whether they've been AI-scraped
Ghost Font's 141-point Hacker News reception proves strong demand for AI-resistant visual communication; creators are increasingly worried about AI scraping and impersonation, and the original article explicitly notes this gap
Ghost Font HN thread confirmed at ~193 points / 143 comments (creator/AI-safety cultural moment), Glaze/Nightshade got MIT Tech Review Innovator-of-the-Year coverage, and Google DeepMind shipped SynthID — strong adjacent signal that creators want human-vs-AI signaling.Ghost Font: A font that humans can read but AI cannot | Hacker News ↗The AI lab waging a guerrilla war over exploitative AI ↗
No direct competitor builds an invisible 'made-by-human' watermark with a human-verification side; Glaze/Nightshade are defensive style-poisoning (free, academic), SynthID flags AI-output (opposite direction), AIWatermark/PixelSeal do ownership not human-authenticity.Glaze - Protecting Artists from Generative AI ↗SynthID — Google DeepMind ↗
Comparable tools bifurcate: Glaze/Nightshade are free (university-backed), while B2B provenance tools like AIWatermark sell 14-day trials to enterprises — no validated creator WTP for a human-proof watermark, and photographers are notoriously price-sensitive.AIWatermark℠ - Invisible Watermarks for Images, Videos & Audio ↗Nightshade: Protecting Copyright ↗
HN thread itself documents AI models (GPT-5.x) already cracking the underlying perceptual trick via temporal averaging, and an open-source 'remove-ai-watermarks' repo strips SynthID/C2PA — this is an arms race the tool will likely lose as vision models gain pixel-perfect resolution.wiltodelta/remove-ai-watermarks - GitHub ↗Ghost Font: A font that humans can read but AI cannot | Hacker News ↗
Top HN commenters showed the exact attack (temporal averaging across 4 frames, ffmpeg + OCR) defeats this category of perception-trick in production; building the embedder is doable, but the core 'AI sees decoy, humans verify' promise is fragile to standard image processing.Ghost Font: A font that humans can read but AI cannot | Hacker News ↗