Midjourney
Prompt metadata on direct downloads, and why Discord copies differ. Open the Midjourney guide.
Why the guides are not interchangeable
Search pages love a single sentence: “this AI writes metadata, strip it.” That sentence is true for a Firefly download and a ChatGPT download, shaky for Midjourney, and false as a promise for Grok or a CapCut video. The pages in this directory say which of those is which, and they say when the honest answer is “inspect the file.”
Two mechanisms get confused. Metadata, including C2PA, XMP, and PNG text, is what this remover deletes. Pixel watermarks, notably SynthID on many Gemini and Imagen images, are what it cannot delete. A guide that mentions both is not hedging. They are different parts of the file.
What we are willing to say
| Tool | Often in the download | Do not assume |
|---|---|---|
| ChatGPT / DALL-E | C2PA on the real download | A screenshot still has it |
| Firefly | C2PA Content Credentials | Every Adobe JPEG is Firefly |
| Photoshop Generative Fill | C2PA if the export keeps it | A normal retouch has a credential |
| Lightroom | EXIF, XMP, GPS if included | AI Denoise always writes C2PA |
| Midjourney | Prompt in PNG text or XMP on many direct downloads | Discord copies still have it |
| Gemini | SynthID in the pixels when Google embedded it | Metadata removal clears SynthID |
| Canva Magic Studio | Content Credentials on some AI downloads | Every design export is signed |
| Meta AI | A credential on some downloads | An in-app post is a file you can edit |
| CapCut | A still you can inspect | The MP4 is cleaned here |
| Grok | Whatever the checker finds | A universal C2PA block |
Open a guide
ChatGPT and DALL-E
C2PA on the download OpenAI gives you. Open the ChatGPT guide.
Gemini
SynthID stays in the pixels. Open the Gemini guide.
Firefly
Content Credentials and digital source type. Open the Firefly guide.
Canva
Magic Studio versus an ordinary design export. Open the Canva guide.
Photoshop
Generative Fill on the exported JPEG, not the PSD. Open the Photoshop guide.
Lightroom
GPS passthrough and Generative Remove. Open the Lightroom guide.
CapCut
Stills only. The video is out of scope. Open the CapCut guide.
Meta AI
Downloads versus posts made inside the app. Open the Meta AI guide.
Grok
Inspect first. No assumed credential. Open the Grok guide.
Use the directory
- 1
Match the export, not the vibe
Choose the tool that wrote the file. A photo edited in Photoshop and captioned in Canva needs the export you will upload, which might be Canva’s.
- 2
Read the limits on that page
Gemini’s limit is SynthID. CapCut’s limit is video. Grok’s limit is that we will not invent a credential.
- 3
Clean with the tool on that page
It is this same remover. JPG, PNG, and WebP pixels stay put.
- 4
Then match the network
The platform directory covers the upload side.
Questions
Do all of these tools write C2PA?
No. Firefly and ChatGPT downloads are the confident C2PA cases. Midjourney is more often PNG text or XMP. Grok and CapCut stills should be inspected rather than assumed.
Which page if I do not know the tool?
Use the metadata checker. It reports what this file contains without you naming the generator.
Can I clean a batch from mixed tools?
Yes. Up to 30 images. The report is per file, because a Firefly JPEG and a screenshot will not look the same.
Does the directory remover work offline after the page loads?
Cleaning runs locally in the browser. The page itself is static. The image is not uploaded.