Found Your Art on Have I Been Trained? What To Do Next
If Have I Been Trained lists your art, treat those older clean files as already exposed. You cannot pull them out of models that already trained on them. Use any opt-out the site still offers, protect every new public export before you post, and keep proof of your original uploads.
Have I Been Trained (often shortened to HIBT) is a public search people use to check whether an image appears in certain training-related datasets. A hit means a matching file showed up in that index. It does not mean every copy of your work everywhere, and it does not mean you can erase the image from models that already finished training.
What a hit actually means
A match on Have I Been Trained means the service found something close to your file in the datasets it searches. That usually points to an older public post, a scrape, or a mirror. It is a signal to act on future posts. It is not a delete button for past training.
When this is enough. Use the result to decide which accounts and folders still publish clean masters, and which ones need a protect step first.
When this is not enough. A blank search does not prove you were never scraped. Indexes change. New datasets appear.
What you can still do today
1. Save evidence. Keep the HIBT result page, the time you checked, and links to your original posts.
2. Use opt-outs that still exist. Follow Have I Been Trained and each platform for their current Do Not Train or removal forms. Write down what you submitted. Opt-outs vary and often only affect future collection, not models already trained.
3. Stop uploading unmarked masters. Keep the clean file offline. Export a smaller public version for social posts.
4. Protect the next export. Run Jingwei Protect on the file you will upload so it carries a visible mark and a JW claim Verify can read. A JW claim is machine-readable credit in the file (your name and how you want the work treated).
5. If style scraping still worries you, process a publish copy with Glaze or Nightshade first, then Protect, as separate steps. Jingwei complements those tools. It does not replace them.
6. Recheck Verify on the PNG you download before you post. Save the live URL, the time, and a screenshot after the post goes up.
What Jingwei can and cannot fix here
Jingwei helps on files you still control. Visible layers raise the cost of a clean wipe. The JW claim can travel with the file so later copies may still show your name and intent on Verify.
Jingwei cannot remove your work from a model that already trained on an older clean file. No watermark tool can rewind that.
Next step. Protect the next public export on jwprotect.com/protect, then spot-check it on Verify.
I found my art on Have I Been Trained. Can I delete it from AI models?
No. You can request opt-outs and removals where a service still offers them. You cannot reliably delete an image from models that already finished training on it. Focus on evidence and on protecting future uploads.
Should I still use Glaze after a dataset hit?
Yes if style learning from new posts is still a fear. Glaze and Nightshade sit on the training side. Jingwei sits on the posted-file side. Many artists keep both lanes.