Check JPG, PNG, WebP, and the standard DOCX package location for supported C2PA evidence. PDF and DOCX also receive general readable metadata checks. Processing stays in your browser.
BeforeShare detects supported C2PA containers, IPTC AI declarations, and readable generator hints locally. It does not detect or remove SynthID or other content-level watermarks; use the relevant provider verifier when you need to check those signals.
The absence of an explicit marker cannot prove that a file was not AI-generated. Metadata can be absent, removed earlier, or stored in an unsupported location.
Inspection and cleaning run locally in your browser. No account is required and the file content is not sent to our server.
BeforeShare reports deterministic file evidence. Use these primary sources to interpret what a marker can — and cannot — establish.
Defines the provenance container and trust model. BeforeShare detects supported container evidence but does not validate its signature or trust chain.
Read the C2PA specificationSynthID and similar provider signals are not ordinary metadata. They require the provider’s compatible verifier.
Read the SynthID documentationCompare current provider claims, open provenance, invisible signals, visible labels, and the limits of a negative result.
Open the Markdown evidence briefPlatform behavior changes by product, model, file type, export path, and date. Inspect the exact downloaded file; these provider-specific notes explain what BeforeShare can remove and what can remain.
Current supported OpenAI image exports include C2PA metadata and a SynthID watermark, although coverage can vary. BeforeShare can inspect and remove supported C2PA and ordinary metadata without re-encoding the image; it cannot detect or remove the embedded SynthID signal. Keep the original if provenance matters and use OpenAI Verify when you need provider verification.
Provider source: OpenAI provenance signalsAdobe says it automatically applies Content Credentials to assets whose pixels are fully generated with Firefly. BeforeShare can remove the supported credential container from the local JPG, PNG, or WebP copy, but it does not validate the credential and cannot erase any durable copy stored in Adobe’s Content Credentials cloud.
Provider source: Adobe Content Credentials overviewMicrosoft documents Content Credentials for AI-generated images from Copilot and Designer. BeforeShare reports supported C2PA container evidence and readable generator hints, then removes the supported container from the cleaned copy. It does not verify the signature, issuer, or trust chain.
Provider source: Microsoft Designer FAQGoogle-generated images can use SynthID, a content-level watermark rather than ordinary file metadata. BeforeShare can remove standard metadata containers that happen to be present, but it does not detect or remove SynthID. A clean local metadata result is therefore not a negative SynthID result.
Provider source: Google Imagen documentationAUTOMATIC1111 can store prompts, negative prompts, seeds, samplers, CFG values, and model details in PNG text. ComfyUI can store the prompt and entire workflow JSON, including in animated PNG comf chunks. BeforeShare now shows those fields as AI evidence and removes them from the cleaned copy; save the original first if you need to reproduce the workflow.
Provider source: AUTOMATIC1111 PNG infoProvider source: ComfyUI metadata writerDo not assume a platform always adds—or never adds—a particular field. Export formats and downstream editors change. BeforeShare checks the actual JPG, PNG, WebP, PDF, or DOCX for supported C2PA, EXIF, XMP, IPTC, PNG text, prompt, model, creator-tool, and document fields, without guessing the source from appearance.
Images may contain prompt, model, creator-tool, timestamp, or C2PA-container evidence. PDF and DOCX files may contain general document properties. Generic software metadata alone cannot prove that AI created a file.
C2PA is a provenance standard for media files. BeforeShare reports supported container evidence and readable generator hints; it does not validate signatures, trust chains, or every assertion.
Some generators and workflows store prompt text, negative prompts, seeds, samplers, model names, checkpoints, LoRA names, or workflow JSON in XMP or PNG text fields.
A visible label, logo, signature, or platform badge is part of the image pixels. Metadata cleanup does not remove visible content.
The visible subject, dimensions, edits, and crops remain. JPEG, PNG, and WebP image data is copied without re-encoding, so the picture and its quality are unchanged; only metadata segments are removed.
Supported EXIF, XMP, IPTC, PNG text, creator-tool, timestamp, model-setting, and content-credential containers are removed. The receipt re-inspects the output for removable fields the detector supports.
If the image visibly contains a watermark, signature, prompt text, username, or private detail, remove that separately before sharing.
JPG, PNG, and WebP files can carry EXIF, XMP, IPTC, PNG text chunks, Content Credentials, creator-tool fields, prompts, seeds, and model names.
AI-generated PDFs commonly keep Title, Author, Subject, Creator, Producer, Keywords, timestamps, and sometimes XMP metadata.
DOCX outputs can include creator, last modified by, company, revisions, editing time, template, app version, and a C2PA 2.4 manifest at the standard package location.
Drop your AI-generated image into the tool above. BeforeShare reads the file in your browser, reports C2PA content credentials and other readable metadata, then creates a clean copy without supported metadata containers. No upload or software installation is required.
Content Credentials use C2PA provenance data. For supported JPG, PNG, and WebP images, BeforeShare creates a copy without supported metadata containers. JPEG, PNG, and WebP are all cleaned at the container level without re-encoding.
Some ChatGPT or DALL-E image exports may contain C2PA or creator-tool metadata, depending on the product, settings, and export path. Inspect the actual file above, then remove supported containers and review the post-clean receipt.
BeforeShare does not detect proprietary content-level watermarks from ordinary metadata. It checks supported C2PA, IPTC, prompt, model, and creator-tool fields locally. Use the relevant provider verifier for SynthID, Amazon, or another provider-specific watermark, and treat a negative result as inconclusive.
Do not remove or hide a required disclosure, provenance marker, platform label, evidence record, or contractual notice. Rules vary by jurisdiction and use. Keep the original and get legal advice when the EU AI Act, China’s labelling rules, or another obligation may apply.
No. BeforeShare removes hidden metadata only — C2PA manifests, content credentials, EXIF, XMP, and IPTC fields. Visible watermarks, logos, signatures, or text baked into the image pixels are part of the picture itself and require image editing to remove.
Never. BeforeShare processes everything locally in your browser using JavaScript. Your file stays on your device — we cannot see it even if we wanted to.