Watermark AI images at the source
Add an invisible watermark to AI images as you create them, and link each one to your records. Etchv detects its own watermarks, not AI in general.
7 days · 200 credits · No card required
{
"output": "gen_7f3a",
"model": "image-v4",
"watermark_id": "cf81d8a0cf…503cab",
"published": true
}- At
- generation or delivery
- 256-bit
- digest of your reference
- Your
- records keep prompts and context
- Known
- watermarks only; not a general AI detector
Link each image to a record you keep
Start inside your own system, right after an image is created. Give it a reference code that means nothing on its own, watermark it, and save the watermark ID Etchv returns with your record of that image. Do this before you publish.
Your record can hold the model version, the time it was made and anything else you need. The watermark carries only a hash (a one-way fingerprint) of the reference you send, not the prompt or the history. Keep prompts and personal details in your own records.
- output_id
- gen_7f3a
- model
- image-v4
- created_at
- 2026-09-26T14:02Z
- prompt
- kept private
- watermark_id
- cf81d8a0cf36…503cab
Choose where to watermark
Watermark after the last edit you control, then test the steps you can’t avoid, such as resizing when the image is shown. A big edit after watermarking can stop the watermark from being detected.
Decide whether you need one watermark per image or one per person you send it to. One per image links every copy to the same creation. One per person also tells you who received it, as long as you keep that list. The same reference always gives the same watermark.
- Generate
- Your edits
- Delivery resize
- Publish
Embed after the steps you control. Evaluate the transformations that follow.
Treat a match as a lookup, and plan for “unknown”
When Etchv finds a watermark, look it up in the records you trust. A known watermark links the image to that record. An unknown watermark, no watermark and a partial result each need their own message in your app. None of them shows that an image is real or made by a person.
Etchv detects its own watermarks. It can’t tell whether any image was made by AI. The confidence score says how sure Etchv is about the bits it read, not how likely the image came from a given model or whether your record is true.
Keep the records behind the watermarks trustworthy
This only works if your records are reliable. Control who can write and change them, keep the creation events you need, limit who can watermark, and log every change.
Content Credentials and invisible watermarks do different jobs. Content Credentials are signed notes about how an image was made and edited, using the C2PA standard. Etchv does not issue them. If you need both, test how each survives the way you share images.
Beautifully invisible.
Distinctly yours.
Start free and watermark your first image, PDF or video in minutes.
7 days · 200 credits · No card required