Frequently asked questions

Visible-pattern review is bounded to tested legacy 48/96 px Gemini white-alpha overlays at documented corner positions. Review every candidate; weak matches require manual selection. LaMa generates plausible pixels inside a chosen mask, not recovered originals. BiRefNet-lite estimates a cutout matte with manual edge refinement; difficult hair, glass and low contrast can be imperfect. Text watermarks use local fonts. Prepare only the selected model explicitly; weights/runtime are cached, while your files, masks, text and results remain local and ephemeral.

Free inspection and the fixed synthetic example remain. Processing user files, edited exports and processed previews require the verified license. Tools cover bounded image transforms, JPEG/PNG XMP and PDF property editing, PDF inventory/extraction and rendering, comparisons, PCM/float WAV stream export, unfragmented MP4 tx3g subtitle conversion, tested native WebM video encoding, and text-only DOCX/PDF conversion. Format and device limits are technical constraints, not commercial quotas. No accounts or uploads are added.

Choose a working tool and inspect its exact support before processing. Metadata cleaning removes only the supported set; image editors intentionally change pixels. Each result states its checks and limits. No operation promises a file is free of every identifying trace.

Free inspection shows structure, support status, findings and proposed changes. The fixed generated PNG and JPEG examples in the Home demo can be inspected, cleaned, verified and downloaded free. This is bound to those exact synthetic examples, not a free-processing allowance for your own files. They do not demonstrate GPS, PDF or Office cleaning.

Once current resources are ready, selecting files inspects them automatically, once. The first time, choose Prepare explicitly; inspection then resumes automatically and no extra inspect action is needed. Cleaning never starts on its own, and you review the result before downloading.

25 MiB per file. JPEG Preserve validates encoded structure without requiring pixel decode. JPEG/PNG Fresh and HEIC primary photos are bounded to 50,331,648 pixels (50.3 MP), 8192 pixels per side. HEIC separately allows 64M primary coded-tile pixels and 20M discarded auxiliary coded pixels, at most 256 items, and a conservative 1536 MiB allocation guard. AVIF/TIFF retain their 40 MP subsets. These are refusal bounds, not a guarantee of browser memory safety. Full 48 MP exports have generated-fixture desktop evidence only, with substantial memory use even after cancellation/retry improvements. Physical iPhone memory, Photos/Files delivery and saving remain unverified. Later groups remain unread until you explicitly acknowledge saving or discarding current results and release the group. No file bytes persist or continue processing after closing the tab. Splitting a selection does not override per-file limits.

Supported files can be cleaned while inspection-only, unsupported or refused files are reported separately with a reason. Your originals are never modified.

Webshop images covers crop, size, format, an optional PNG logo and a measured size check for up to 10 images. Share photos inspects readable fields, then cleans, edits or only inspects, and asks you to review faces and visible text yourself, because metadata removal is not redaction. Check a client PDF lists properties and embedded assets, extracts text only when you ask, and offers no OCR or page management.

Two caveats: your host may keep ordinary request logs for the app itself, and billing is separate. Payment and license requests go to online services only when you choose them and never contain file contents, names, hashes or metadata.

PixelRinse does not persist selected files or outputs in its app cache. Downloads are saved by your browser to your device filesystem, outside the tab session. Closing a tab is not secure erasure: browsers and operating systems do not guarantee when memory is overwritten. Your original files on disk are never modified.

Faces, text, license plates and reflections remain in a metadata-cleaned copy. Pixel editors such as background removal, retouching and text watermarks deliberately change images, but do not guarantee anonymization or complete redaction. Use an appropriate redaction workflow and review the final saved file.

Cleaning creates a separate copy in memory, which you choose whether to download.

Current naming remains the default. Anonymous mode ignores custom prefixes and original basenames, preserves actual output extensions and uses consistent session names in individual downloads and new bundles. It does not rename internal archive entries, document links or originals, and it does not replace metadata cleaning. No original-name mapping is added.

Recipes are saved only when you choose Save, in this browser. They never contain files, the logo image, metadata values you typed or your confirmation of effects, and loading one never runs anything. Export and import use a small JSON file of up to 20 KB; Forget removes them all.

A proxy is a smaller viewing raster, not full-resolution detail. Its zoom is labeled as preview pixels. Clean/compress comparisons currently require supported still images within 8 MP; HEIC and larger images do not gain a hidden full-size decoder. Full-window viewers close with Escape. Originals are unchanged.

1 MiB is 1,048,576 bytes; 1 MB is 1,000,000 bytes. Current input limits remain 25 MiB per file, not 32 MiB. General photo tools and 8 MP editors have different pixel limits; 20 files/200 MiB is still the active cleaner-group budget. PDF and media tasks have additional page, time and output limits. For larger files choose a smaller working copy with a trusted local desktop tool; never rename a file to bypass checks.

Page renders default to 1200 px longest edge; opt-in 2400 px uses more memory. Up to 20 pages per run, 32 MiB total output, sequential rendering. JPEG/WebP are lossy. Every export must pass encoding and reopening checks; unsupported WebP stops with an error rather than silently returning PNG. Page renders flatten text, links and layers; they are not extracted original images. OCR and general PDF merge/split/page management are not implemented.

The image editor supports standard XMP in single-picture JPEG/PNG while other bytes, including EXIF/IPTC and rendering information, are unchanged and read-only. The PDF editor handles Info properties and supported catalog XMP, not every page/object packet. Signed/provenance-bearing and ambiguous packets may be read-only. A later explicit Clean operation can remove deliberately written values; it is not automatically applied after editing.

Unsupported filters, masked images, missing font programs and unusual resource graphs are inventoried with reasons. A font marked as a subset is not a full font, and extraction does not grant redistribution rights. Attachments remain inert and extraction does not clean metadata inside assets. Page-to-PNG rendering and text extraction are separate operations; scanned pages need OCR, which is unavailable.

DOCX-to-PDF does not preserve tables, images, headers/footers or original pagination; its current font subset cannot encode every language. PDF-to-DOCX creates an editable text-only document, not a faithful reconstruction or OCR. This existing answer link is retained. The former video part now has its own answer: video comparison checks five synchronized samples, not every frame or audio.

JPEG/WebP offer lossy quality settings and bounded target-size search. PNG defaults to lossless; optional RGB color reduction (32 levels/channel, up to 8 MP) deliberately changes colors without quantizing alpha. Compare actual bytes: output can grow. Crop and enlargement are off by default. Animation is refused; secondary JPEG photos need explicit primary-only consent. HEIC remains conditional on validated decoding and HDR/color/auxiliary effects, with PNG/JPEG output only. Processing and exports of your files still require the existing Individual Pro license; free inspection and fixed synthetic demos remain.

After you add a file, PixelRinse also states which class that exact file falls into and why, before anything is cleaned.

HEIC uses local WebAssembly for validated SDR grids and transforms. Bounded primary photos have local RGB matrix/TRC ICC, P3 and selected BT.2020 SDR/PQ/HLG color-managed paths. A licensed public iPhone 13 Pro ICC-grid/gain-map fixture has bounded primary-only conversion evidence; that gain map is discarded, not rendered as HDR. Other managed graphs remain conditional. PNG/JPEG exports are 8-bit SDR sRGB: gamut clips, HDR uses fixed 203-nit Reinhard mapping (HLG 1000-nit, gamma 1.2) and appearance can change. Alpha, sequences, independent photos, nested grids, non-neutral tile crops and unknown dependencies remain unavailable. 25 MiB per file. JPEG Preserve validates encoded structure without requiring pixel decode. JPEG/PNG Fresh and HEIC primary photos are bounded to 50,331,648 pixels (50.3 MP), 8192 pixels per side. HEIC separately allows 64M primary coded-tile pixels and 20M discarded auxiliary coded pixels, at most 256 items, and a conservative 1536 MiB allocation guard. AVIF/TIFF retain their 40 MP subsets. These are refusal bounds, not a guarantee of browser memory safety. Full 48 MP exports have generated-fixture desktop evidence only, with substantial memory use even after cancellation/retry improvements. Physical iPhone memory, Photos/Files delivery and saving remain unverified. Explicit consent and paid user-file rights are required. This is not unchanged-pixel, HDR-preservation, complete-removal, physical iPhone 17 Pro Max/Safari or universal-compatibility proof.

Re-encoding can change bytes and quality, so it is an explicit opt-in. PixelRinse never switches to it automatically.

Custom presets for retaining specific EXIF fields are listed as future work on the privacy page. Today PixelRinse removes the supported metadata set or leaves the file alone.

When Preserve cannot keep the picture correct, PixelRinse refuses with a reason rather than producing a subtly different file. Your original stays unchanged.

PixelRinse has no independently validated, browser-local SynthID image, video or audio detector/remover. A visible overlay being removed, a metadata packet disappearing or a file looking different is not proof about invisible SynthID. Local text research needs the original watermark configuration/key and tokenizer and does not validate an image tool. Pixel edits can invalidate C2PA assertions; they do not restore provenance or authenticity.

Missing or unsupported tracks are not reported as equal. A separate short-video conversion is a bounded realtime VP8/Opus WebM re-encode that can change frames, timing and quality; it is not a lossless copy. Matching samples do not prove whole-video, audio or byte equality.

Examples are that it parses, that supported metadata fields are gone and that retained content is intact. It describes those checks only, not every possible hidden trace.

Partial means extraction covered only part of the file because some content was unread or the scope was limited. Inspection removes nothing, and unavailable is not the same as empty.

Treat those areas as unknown and review them yourself before sharing.

Use the comparison and verification receipt to see what changed. File size may differ slightly, and preserved image pixels are meant to be unchanged.

Preview it before optionally downloading. It includes anonymous item IDs, outcome counts and limitations, not filenames, paths, actual metadata values, locations, hashes or license/payment details. Inspection-only files are not marked cleaned; not checked never means passed. It is not independent certification or guaranteed anonymity, and is never transmitted or automatically saved.

Merely loading the page is not enough. First-time preparation, updates, payments, payment checks and legacy license upgrades need a connection.

You can leave the tab open to review work. Download verified outputs first if you choose Load current app: acknowledge that selected files and results will be lost, then download app resources and reload this tab only. No files are restored. Your original files and license data are not deleted, and no automatic reload or site-data reset is performed.

Installing does not itself prepare offline resources, mobile browsers can evict site data under storage pressure, and private windows may not keep anything.

It covers build, connection and browser capability status, never files, file names, identifiers or licenses, and is never sent automatically. Do not email private files.

It is perpetual for the purchased major version and includes 12 months of updates; expiry of updates does not revoke it. PixelRinse adds no payment surcharge, but your bank, card issuer or wallet network may charge separate external fees that PixelRinse does not control.

Load purchase options in the unlock panel to see them; anything not configured stays unavailable. Checkout is hosted by the payment provider.

Open the unlock panel and choose Check payment for your saved order; a pending payment stays pending until the provider confirms. Never pay a second time for the same order.

Export a recovery file once unlocked and keep it yourself. Older licenses need Upgrade license while online, and orders from before hosted checkout can use the legacy recovery option, which only verifies an existing order.

PixelRinse does not promise unlimited devices. The license is stored locally, not in an account, and team use requires an agreement.

When the update period ends, cleaning keeps working. It does not cover a future major version, team use, or unlimited devices.

Up to 100 selected references / 1 GiB can be held in this tab, but each active group is limited to 20 files / 200 MiB. 25 MiB per file. JPEG Preserve validates encoded structure without requiring pixel decode. JPEG/PNG Fresh and HEIC primary photos are bounded to 50,331,648 pixels (50.3 MP), 8192 pixels per side. HEIC separately allows 64M primary coded-tile pixels and 20M discarded auxiliary coded pixels, at most 256 items, and a conservative 1536 MiB allocation guard. AVIF/TIFF retain their 40 MP subsets. These are refusal bounds, not a guarantee of browser memory safety. Full 48 MP exports have generated-fixture desktop evidence only, with substantial memory use even after cancellation/retry improvements. Physical iPhone memory, Photos/Files delivery and saving remain unverified. These safety limits are not usage-metering; no file persists or processes after closing this tab.