AI Metadata Editor
Inspect supported AI-related image parameters, edit selected fields like prompts, seeds, and sampler settings, independently verify the output, and download your updated image.
Drop your image here
or choose a file from your device
Understanding the Three AI Metadata Tools
Our platform provides three dedicated tools designed for specific user needs when managing AI generation metadata:
Audit container headers to understand if an image contains AI parameters, provenance manifests, or generator footprints.
Selectively change prompts, seeds, model checkpoints, or creator credits while preserving the rest of the file intact.
Strip all AI parameters, prompts, negative constraints, and node graphs completely from the image container.
What Fields Can Be Edited (and What Is View-Only)?
- Prompt: Positive text prompt in PNG parameters or ComfyUI node graphs.
- Negative Prompt: Negative constraint text.
- Seed: Random initialization seed number.
- Model / Checkpoint: Base checkpoint filename.
- Sampler & Scheduler: Sampler algorithm (Euler, DPM++ 2M, etc.).
- Steps & CFG Scale: Denoising steps and guidance scale.
- Creator / Author: Dublin Core author or PNG Artist chunk.
- Title & Description: Asset caption and title tags.
- C2PA Content Credentials ManifestsCryptographic provenance manifests are digitally signed with private keys. Editing a signed claim invalidates the hash and breaks the signature. Use C2PA Remover to strip if necessary.
- Proprietary Camera MakerNotesBinary vendor-specific data blocks cannot be safely modified without manufacturer-specific encoders.
How to Edit AI Metadata in 4 Steps
- Select your image: Drop any supported PNG or JPEG file into the workspace above.
- Review detected fields: Examine the extracted prompt, seed, model, and sampling parameters.
- Modify target values: Type updated values for the fields you want to change, and review the pending changes summary.
- Apply & Download: Click "Apply Changes & Verify" to write the updated chunks, review the independent second-pass verification table, and download your updated file.
Technical Scope & Boundaries
Metadata editing updates file container headers only. It does not re-render the image, change visible visual pixel elements, or run generative diffusion models. Unmodified metadata tags and image pixel bitstreams remain 100% mathematically identical.
How We Verify: Lossless Binary Slicing & Independent Two-Pass Verification
We do not ask you to trust that metadata was removed. Every cleaning operation preserves pixel streams without re-encoding and independently validates the output binary before download.
100% Client-Side Processing
All byte manipulation executes locally in browser memory via typed arrays (Uint8Array). Zero image bytes, filenames, or metadata values are transmitted to remote servers.
Bit-for-Bit Pixel Preservation
Rather than re-encoding via canvas, the engine splices container markers directly. JPEG entropy scan data (SOS) and PNG raster chunks (IDAT) remain bit-for-bit identical with zero generational loss.
Two-Pass Rescan Verification
After cleaning, the output binary is re-parsed through the full scanning pipeline. The tool verifies each targeted metadata field independently to prove it is absent from the exported container.
Empirical Test Lab Fixtures
Our parsing and cleaning behavior is tested against synthetic and real-world fixtures spanning GPS IFDs, camera serials, ComfyUI/SD parameters, and 10 C2PA signature scenarios.
Scope & Technical Boundaries: What Metadata Removal Can and Cannot Do
What this tool removes: Embedded container headers and text metadata chunks (EXIF, IPTC, Adobe XMP, PNG ancillary text chunks, ComfyUI workflow graphs, Stable Diffusion prompt parameters, and C2PA JUMBF provenance manifests).
What this tool does not do: It does not alter visual pixel content, remove rasterized visual stamps/watermarks burned into the image pixels, or strip frequency-domain steganographic watermarks (such as Google SynthID or Adobe Digimarc). Furthermore, third-party social media and messaging platforms frequently re-encode uploads, stripping existing metadata or embedding their own platform headers.