If you are here from my last post, “Why Real Photos Get An AI Label On Facebook Or Instagram“, then this is part 2 for you. I told you guys I will be sharing the logic here for making AI generated images come out of the watermarks?
Well YES, IT’S THIS POST, and what I will talk about will be 100% with proof. I am not talking out of the blue.
Tested With The Picazor Pony And What SynthID Said About It
Now I will show you with this example. This is PICAZOR, the FB downloader, and its PONY mascot is generated with AI. No harm in generating these images, obviously AI is a helping hand, and this is the reason the tech sector is getting layoffs. Earlier the PICAZOR team might have paid $50 for this, and now with their founder’s idea AI got it done.


Well, the use of AI apart, now see this image below. SynthID detected it in a few seconds, saying that:
SynthID was detected. This media was made or edited with OpenAI.


I Asked GPT To Trace The Same Image In SVG
Now see the claim I was making. I copied that image and pasted it in GPT (ASTRA 6.2 model), ok, and remember OPEN AI is part of the SynthID program. So well, now it generated me the SVG, and you can see the output.


Now SynthID won’t pick a direct SVG generated image, obviously, so what I do is either download it and convert it to any other popular format, or screenshot the image. Correct?
Now one thing that you might be thinking is that I could have simply just converted the direct Picazor image, the primary one that SynthID detected at first as AI generated. Well yes, I tried that, and it easily detects it. I changed the format of the same Picazor image and checked again, and SynthID was still detected. The reason is that this mark is not kept in the format or the file name. It is mixed into the picture itself, so wherever the picture goes the mark goes with it. The original Picazor image was generated via GPT, other such images are made in Google Labs, and the image I copied from Picazor was in .webp, by the way.
Converting or tracing a raster image into vector SVG instructions (lines, curves, shapes) reconstructs geometry rather than preserving original pixel-level watermark patterns.
Well, now coming to the results of the SVG traced one. Same image when SVG traced, and see the output, what does it say:
SynthID was not detected. It is unlikely that this media was made with AI from a SynthID partner.


Then I Gave The Same Image To Grok
Now I gave this image to Grok, asked it and it replied. See, same image from Picazor, control C, and SynthID said:
SynthID was not detected. It is unlikely that this media was made with AI from a SynthID partner.


Now here SynthID is legit, because xAI is not partnering. What I meant here is the claim that I made about avoiding AI labels, and this is the logic of it.
What Weight It Gets In Facebook, ChatGPT Or Google
Now I don’t know what weight and power it gets in the FB algorithm, or in ChatGPT answers when showing results, or in the Google SERP. As far as being an SEO myself, I am aware that GOOGLE pushes first hand and first experience material in the SERP, like they want high quality content, not the programmatic or LLM generated one.
That debate apart, see how a paid ad silences these quality raters, when something comes from outside the top 100 results and sits over even the AI Overviews, regardless of content quality or of being AI generated. I question you, Google, on this. If we pay money we simply bypass all your SEO things, the Google helpful guidelines and those SpamBrain things, right??
Well, leave that, it’s GOOGLE, they have been doing it for years.
Why The Traced Image Has No Watermark
Now do you see, this is how the logic of watermarking actually is. Once these models reply with the image, they put in the watermarking thing that I explained in the last post.
Let’s say you asked AI to generate an image of a horse eating grass. What the AI does is draw the horse, the grass and the field dot by dot, and while it is drawing it mixes its own invisible mark into those same dots. So the mark is born with the image.
Now once you ask any model to trace it, what it does is it is not generating that image again. It is now coding, the code level work: a line from here to there, a circle here, fill this with blue. There are no dots in it for the mark to live in.
Now the same logic is for META. I feel this spost was much focused on the Google thing, so let me tell you that META has no direct tool like SynthID by DeepMind of Google, but the logic is exactly the same. META reads the note that the AI tool writes into the file, the one I explained in the last post, and the traced image is a new file with no such note in it. So you can also avoid the AI generated label from META on your social media posts at Threads, Instagram or FB.
Now in future even in the codes AI models might add the watermarking, but for now it’s not there.
And for GROK, please note, Grok can satisfy you for SynthID or bypass META detection, but xAI is adding their own watermarks. So if you post these on X or similar, X is aware of AI. And if in future these big tech partner with each other, like the two rivals ChatGPT and Google DeepMind did, so xAI and META, or xAI and Google, can do so.














