DALL-E, Midjourney, and Stable Diffusion (and tools built on them) all write metadata into their outputs. When you upload those images to Instagram, Pinterest, or TikTok, that metadata triggers the "Made with AI" or "AI Info" label. The three tools don't embed the same things in the same way. This guide compares what each does and how removal differs.
DALL-E / ChatGPT (OpenAI)
What it embeds: OpenAI exports typically include C2PA content credentials. C2PA is a cryptographic provenance standard that platforms scan for explicitly. DALL-E images may also carry XMP and EXIF software tags identifying the tool.
Why it matters: C2PA is one of the strongest triggers for AI labels. Platforms are tuned to detect it. Removing C2PA (and any XMP/EXIF) from DALL-E exports is usually enough to stop the label.
Format: Exports are often PNG or JPEG. Metadata lives in standard EXIF/XMP and in C2PA manifests. A tool that strips C2PA, XMP, and EXIF handles DALL-E outputs well.
Midjourney
What it embeds: Midjourney writes XMP generation parameters — creator tool, model version, seed, and sometimes prompt-related data. PNG exports can include additional chunks. EXIF software fields often reference Midjourney.
Why it matters: Platforms look for XMP AI tags and EXIF software identifiers. Midjourney doesn't use C2PA by default, so the main triggers are XMP and EXIF. Stripping both removes the signal.
Format: PNG and JPEG. Metadata is mostly XMP + EXIF. Some PNGs have extra chunks. A remover that handles XMP, EXIF, and PNG text chunks covers Midjourney.
Stable Diffusion (including ComfyUI, Automatic1111)
What it embeds: The most metadata-heavy of the three. PNG text chunks (tEXt, iTXt) often contain full workflow data: prompt, negative prompt, seed, steps, CFG, model name. XMP and EXIF are also common. ComfyUI may embed node/graph data.
Why it matters: Platforms scan for both XMP and PNG chunks. Stable Diffusion's chunk data is a direct signal. Removal must strip PNG chunks as well as XMP/EXIF, or the label can still appear.
Format: PNG is common; JPEG is less so. For PNG, a tool that removes PNG text chunks in addition to EXIF/XMP/C2PA is important.
Side-by-side summary
| Tool | Main metadata | Strongest trigger | Removal focus |
|---|---|---|---|
| DALL-E | C2PA, XMP, EXIF | C2PA | Strip C2PA + XMP + EXIF |
| Midjourney | XMP, EXIF, PNG chunks | XMP | Strip XMP + EXIF (and chunks) |
| Stable Diffusion | PNG chunks, XMP, EXIF | PNG chunks + XMP | Strip chunks + XMP + EXIF |
One workflow for all three
Use a single metadata remover that supports:
- C2PA removal (for DALL-E and any C2PA-enabled export)
- XMP stripping (all three)
- EXIF stripping (all three)
- PNG text chunk removal (Midjourney PNGs, all Stable Diffusion PNGs)
Process your export once before uploading. The same workflow works for DALL-E, Midjourney, and Stable Diffusion; the tool just needs to handle all the formats each can write.
Summary
DALL-E leans on C2PA; Midjourney on XMP and EXIF; Stable Diffusion on PNG chunks plus XMP and EXIF. Platforms use all of these to apply AI labels. A remover that strips C2PA, XMP, EXIF, and PNG chunks handles all three. For a free, in-browser option that removes these markers from DALL-E, Midjourney, and Stable Diffusion exports, you can use Remove AI Label before you post — no account needed, and files are processed locally.