OpenAI is adding new watermarks to DALL-E 3

Key Points:

  • Watermarks from C2PA will appear in images generated on ChatGPT website and API for DALL-E 3 model.
  • Check provenance of images generated by OpenAI platforms through websites like Content Credentials Verify.
  • Adding watermark metadata to images has “negligible effect on latency” and will not affect image quality.


OpenAI’s latest DALL-E 3 image generator is getting a snazzy new makeover with watermarks from the Coalition for Content Provenance and Authenticity (C2PA). These watermarks will be popping up on images churned out by the ChatGPT website and the DALL-E 3 model API, making reaching for the metadata as cool as finding extra fries at the bottom of your takeout bag. Mobile users, get ready to feel the love by February 12th, as these watermarks sport both visible CR symbols and invisible metadata components, adding a touch of mystery to your digital experience.


But, before you splash into a pool of watermark excitement, know that this party’s reserved for still images only; videos and texts need not apply. OpenAI assures us that these watermarks won’t cramp the style of image generation, causing only a minuscule latency blip and a slight bump in image size. The C2PA, playing the role of the watermark cheerleader, boasts a star-studded lineup including Adobe and Microsoft, pushing for these digital birth certificates of sorts to certify whether your content is born from human hands or the spark of AI brilliance.


In a swirl of AI-generated content, Adobe whipped up a Content Credentials symbol, now a trendsetter as OpenAI jumps on the bandwagon, sticking them on DALL-E 3 creations like jewels on a crown. Meanwhile, Meta’s also throwing its hat in the ring, swirling tags like confetti on AI-generated content across its social media realms. It’s all part of the odyssey to fulfill one of the Biden administration’s AI executive order goals: separating the real McCoy from the AI daydreams.


However, don’t bet the house on watermarks being the silver bullet against digital falsehoods. OpenAI smartly notes that these metadata tags can vanish quicker than pickles at a kids’ party, especially on social media platforms that play fast and loose with metadata. Snapping a screenshot? Say goodbye to that digital fingerprint. Despite the hurdles, OpenAI is confident that championing these provenance-establishing methods and pushing users to decode these signals will pump up the trustworthiness of your digital information. So, get ready to dive into the digital watermark pool; the water’s fine, and the symbolism game is strong.



Prompt Engineering Guides



©2024 The Horizon