The invisible provenance Microsoft embeds in AI images made in Paint (Cocreator) and Windows Photos — a server-issued GUID hidden in the pixels plus a signed C2PA manifest, applied whether or not you turn on the visible mark.
Last verified · 2026-08-24 · by Moe Ameen
Microsoft Paint AI Watermarking refers to the provenance signals Windows attaches to images generated by Paint's Cocreator feature and the Photos app's Image Creator / Restyle tools. Reverse-engineering research published on August 20, 2026 documented two layers that most users never see. The first is an invisible, pixel-level watermark carrying a server-issued GUID — a 16-byte identifier written across the image as roughly 144 bits using Microsoft's content-adaptive InvisMark method, imperceptible while you look at the picture but recoverable by a matching detector. The second is a signed C2PA (Content Credentials) manifest attached to the file that records the same GUID in a provenance assertion, so a checker can confirm the image was AI-generated and whether the metadata was later altered.
The part worth understanding before you rely on any of it: the GUID is issued by Microsoft, not generated on your PC. Even when Cocreator produces the image on-device — for example on a Copilot+ PC's NPU — the prompt is sent to a Microsoft moderation endpoint that returns a revised prompt and the watermark identifier, which is then embedded in the finished image. So the mark ties an image back to the Microsoft-side request that made it, and "local" generation still involves an online round-trip for moderation and provenance signing.
This is separate from the opt-in visible watermark Microsoft added to Paint and Photos in 2026, which is off by default and offers "Never," "Always," or a prompt-each-time choice when you save. The invisible GUID and the C2PA manifest are reported to apply to AI generations regardless of that visible setting — turning the visible badge off does not remove them. Microsoft has attached C2PA provenance to its AI image output since 2023; the newer finding is the hidden in-pixel GUID and its link to a server-issued identifier.
The driver is transparency regulation. The EU AI Act's Article 50 rules took effect on August 2, 2026 and press generative-AI providers to make output carry a detectable, machine-readable mark — the same force behind [Claude's content watermarking](/ai-tools/claude-content-watermarking) and Google's SynthID. Because Microsoft has not published a detailed public spec of the invisible behavior in Paint, treat the mechanics here as independent reverse-engineering findings and confirm anything load-bearing before you build a compliance process on it.
The watermark labels the image; it does nothing to make that image on-brand, sized for each feed, or published — and as more of your tools start marking their output, the useful discipline is a single, deliberate disclosure point rather than a different one bolted onto every app. That is where [Kompozy](/) fits. It is a full AI content generation and multi-platform publishing engine, so a provenance-aware image workflow becomes concrete: Kompozy generates images through OpenAI's gpt-image and Google Gemini and lays them into brand-exact [Carousel Posts](/glossary/output-buckets), quote cards, and face-locked [Persona Photos](/glossary/persona-shorts) via [HyperFrames](/glossary/hyperframes), then fans that one source into a blog, a newsletter, captioned video, and native posts across the eight social platforms plus blog and email — each passing a per-post review gate where your standard AI-disclosure line goes in once and rides every asset.
Two honest specifics, because provenance is the whole subject here. First, Kompozy's own generated images come from gpt-image and Gemini, not Paint — so they carry those providers' provenance, not Microsoft's invisible GUID or C2PA tag. Second, if you bring a Paint- or Photos-made image into Kompozy as a source asset, Microsoft's mark travels with that file; Kompozy neither strips it nor claims to. The value is not evading anyone's watermark — it is running a disclosed pipeline where you know exactly which tool touched which asset, and where the output is a finished, scheduled set of posts instead of a single PNG you still have to place by hand.
Two things, per reverse-engineering research from August 20, 2026: an invisible, pixel-level watermark carrying a server-issued 16-byte GUID (via Microsoft's InvisMark method), and a signed C2PA provenance manifest that records the same GUID. Both are reported to be applied to AI generations in Paint (Cocreator) and Photos regardless of the opt-in visible-watermark setting.
The image can be generated on-device on a Copilot+ PC, but the prompt is still sent to a Microsoft moderation endpoint that returns a revised prompt and the watermark GUID. So the identifier is issued server-side and the finished image goes through online provenance signing — "local" generation is not fully offline.
There is no confirmed user toggle for the invisible GUID or the C2PA manifest — the visible-watermark setting ("Never," "Always," "Ask every time") governs only the visible badge. Because Microsoft has not published a detailed spec, the safe assumption is that anything generated in Paint or Photos carries provenance. The durable strategy is clean disclosure, not evasion.
No. Kompozy generates images with OpenAI gpt-image and Google Gemini composited through HyperFrames, not Paint, so they carry those providers' provenance rather than Microsoft's GUID or C2PA tag. If you import a Paint image into Kompozy as a source, that file keeps its Microsoft mark — Kompozy does not strip it.
It fits the industry-wide move to label synthetic media, driven largely by the EU AI Act's Article 50 transparency rules that took effect on August 2, 2026. Anthropic (Claude), Google (SynthID), and Suno have made comparable moves for text, image, and audio.