Google’s Unremovable Watermark for AI Images SynthID is Now Open Source

In May 2024, Google͏͏ had developed͏͏ a͏͏ new͏͏ watermarking͏͏ tool͏͏ that͏͏ embeds͏͏ a͏͏ hidden͏͏ signature͏͏ into͏͏ AI-generated͏͏ images,͏͏ making͏͏ it͏͏ nearly͏͏ impossible͏͏ to͏͏ remove.͏͏ The tech giant has now made this technology – Open Source, this͏͏ is͏͏ particularly͏͏ fascinating͏͏ because͏͏ it͏͏ addresses͏͏ a͏͏ growing͏͏ concern͏͏ about͏͏ distinguishing͏͏ real͏͏ images͏͏ from͏͏ those͏͏ created͏͏ by͏͏ artificial͏͏ intelligence.

How͏͏ SynthID͏͏ Works

SynthID͏͏ was͏͏ developed͏͏ by͏͏ Google͏͏ DeepMind͏͏ and͏͏ works͏͏ by͏͏ subtly͏͏ altering͏͏ the͏͏ pixel͏͏ data͏͏ in͏͏ AI-generated͏͏ images.͏͏ These͏͏ changes͏͏ are͏͏ imperceptible͏͏ to͏͏ the͏͏ human͏͏ eye͏͏ but͏͏ can͏͏ be͏͏ detected͏͏ by͏͏ specialized͏͏ algorithms.͏͏ This͏͏ means͏͏ that͏͏ even͏͏ if͏͏ someone͏͏ tries͏͏ to͏͏ edit͏͏ or͏͏ crop͏͏ the͏͏ image,͏͏ the͏͏ watermark͏͏ remains͏͏ intact.

According͏͏ to͏͏ Pushmeet͏͏ Kohli,͏͏ the͏͏ Vice͏͏ President͏͏ of͏͏ Research͏͏ at͏͏ Google͏͏ DeepMind,͏͏ “SynthID͏͏ provides͏͏ a͏͏ new͏͏ way͏͏ to͏͏ embed͏͏ information͏͏ directly͏͏ into͏͏ images,͏͏ which͏͏ can͏͏ be͏͏ a͏͏ powerful͏͏ tool͏͏ in͏͏ managing͏͏ the͏͏ spread͏͏ of͏͏ AI-generated͏͏ content.”

Read More

Bengaluru flooding: DK Shivakumar restricts construction within 50-ft of SWDs, lakes

Today’s Rahu Kaal timings in major cities [28 October 2024]: Delhi, Mumbai and more

Weather, human errors & poor roads; key factors behind road accidents in Uttarakhand

Why͏͏ This͏͏ Matters

With͏͏ the͏͏ surge͏͏ of͏͏ AI-generated͏͏ content͏͏ flooding͏͏ the͏͏ internet,͏͏ it’s͏͏ becoming͏͏ harder͏͏ to͏͏ tell͏͏ what’s͏͏ real͏͏ and͏͏ what’s͏͏ not.͏͏ This͏͏ tool,͏͏ called͏͏ SynthID,͏͏ is͏͏ designed͏͏ to͏͏ help͏͏ identify͏͏ images͏͏ produced͏͏ by͏͏ AI͏͏ models.͏͏ It͏͏ embeds͏͏ a͏͏ watermark͏͏ directly͏͏ into͏͏ the͏͏ pixels͏͏ of͏͏ an͏͏ image͏͏ without͏͏ affecting͏͏ its͏͏ quality,͏͏ so͏͏ you͏͏ won’t͏͏ even͏͏ notice͏͏ it’s͏͏ there.

Real-World͏͏ Applications

Spending͏͏ a͏͏ few͏͏ moments͏͏ on͏͏ X͏͏ (Twitter)͏͏ is͏͏ easy͏͏ to͏͏ understand͏͏ how͏͏ misinformation͏͏ spreads͏͏ through͏͏ manipulated͏͏ images.͏͏ With͏͏ this͏͏ watermarking͏͏ technology,͏͏ platforms͏͏ can͏͏ more͏͏ easily͏͏ flag͏͏ AI-generated͏͏ images,͏͏ helping͏͏ to͏͏ combat͏͏ fake͏͏ news͏͏ and͏͏ digital͏͏ fraud.͏͏ It’s͏͏ like͏͏ giving͏͏ each͏͏ AI͏͏ image͏͏ a͏͏ unique͏͏ fingerprint.

Is it Foolproof?

Of͏͏ course,͏͏ no͏͏ technology͏͏ is͏͏ foolproof.͏͏ SynthID͏͏ works͏͏ best͏͏ with͏͏ images͏͏ generated͏͏ by͏͏ Google’s͏͏ own͏͏ AI͏͏ models,͏͏ like͏͏ Imagen.͏͏ It͏͏ might͏͏ not͏͏ be͏͏ as͏͏ effective͏͏ with͏͏ images͏͏ produced͏͏ by͏͏ other͏͏ systems.͏͏ Also,͏͏ while͏͏ the͏͏ watermark͏͏ is͏͏ resistant͏͏ to͏͏ common͏͏ edits,͏͏ extremely͏͏ heavy͏͏ modifications͏͏ could͏͏ potentially͏͏ disrupt͏͏ it.

Open͏͏ Source͏͏ Availability

Google͏͏ has͏͏ made͏͏ SynthID͏͏ available͏͏ to͏͏ developers͏͏ and͏͏ businesses͏͏ by͏͏ open-sourcing͏͏ the͏͏ tool.͏͏ This͏͏ move͏͏ allows͏͏ others͏͏ to͏͏ incorporate͏͏ the͏͏ watermarking͏͏ technology͏͏ into͏͏ their͏͏ own͏͏ AI͏͏ models.͏͏ João͏͏ Gante,͏͏ a͏͏ machine-learning͏͏ engineer͏͏ at͏͏ Hugging͏͏ Face,͏͏ noted,͏͏ “By͏͏ making͏͏ SynthID͏͏ open͏͏ source,͏͏ Google͏͏ is͏͏ enabling͏͏ wider͏͏ adoption͏͏ and͏͏ collaboration͏͏ within͏͏ the͏͏ AI͏͏ community.”