Lantern is still a work-in-progress, so various aspects of the website, FAQs, GH, Discord, are still yet to be fleshed out. If you’d like to contribute, please feel free to join our Discord!
What is Lantern
Lantern is an open-source tool that does two things:
1. Check if your work appears in publicly-posted AI training datasets.
2. Set alert to notify you when your work appears in new datasets.
We aim to regularly scan new publicly-posted AI training sets for the works you choose to receive notifications about, and will send you an email when there is a match.
As an open-source tool, you can inspect Lantern’s code, contribute, or take the code and use it to check against datasets of your own choosing, without using our website.
How does Lantern work?
Simply put, the your image is checked against other images in datasets for a visual match. When you decide to search for your work, a one-way “fingerprint” for the image is generated using the PDQ fingerprinting algorithm. This happens on your device, and your image does not get uploaded to our servers.
Images that look the same will have the same fingerprint, and there is no way to reconstruct the image from the fingerprint. We then match the fingerprints against one-another in the dataset, where you can see the search results, and also set up watchlist alerts.
For the images you set up watchlist alerts for, when we update with new publicly-posted AI training datasets, if there is a match, you will receive an email alert.
Is my work safe with you?
Lantern does not save or store a copy of your art by default. Instead, we create a one-way fingerprint of your image that becomes a string that acts like an ID. This identifier lets us search for your work in datasets, but there is no way to reconstruct your image from the ID in reverse.
You can optionally choose to allow Lantern to show a small thumbnail of your work for a better dashboard experience, which requires your permission to toggle on.
Where does Lantern source datasets to scan?
Lantern scans the AI websites Hugging Face and ModelScope for new datasets. We also manually add datasets to our index, if you are aware of a new dataset that is not on Hugging Face/ModelScope that you are concerned about, please send us a message and we can look into it.
What do I do if Lantern finds my work in a dataset?
Lantern’ will’s scan results will provide you with a link to the source dataset. There, you can submit a takedown/DMCA notice. Lantern does not automatically submit takedowns for you, you must submit these yourself. Separately, we recommend you register your work with the copyright office.
How broad is Lantern’s search?
While we aim to search as many publicly-posted AI training datasets as we can. Large companies like OpenAI, Google, and Anthropic use their own internal datasets rather than sets published on platforms such as Hugging Face. So Lantern is an attempt to notify you when your work appears in these public datasets. It is not possible for us to scan non-public datasets.
Is Lantern free?
Lantern is free to use, with generous watchlist limits to prevent abuse. Please message us on Discord if you need to scan more works.