Changelog

New

• A new look at the top. The Foxglove wordmark, with the mascot peeking from behind it. On the website an “Under construction” sign hangs from it. Now and then the mascot visits other parts of the page; turn that off with Mascot visits in the About tab (Reduce Motion on your device also keeps it still).
• A batch takes up to 20 images. If you add more, the first 20 that open are kept and the app says how many were left out.
• When a HEIC, HEIF, AVIF, TIFF, PSD, JPEG XL or camera raw file doesn’t open, the message names the format and says to save it as JPEG or PNG. The logo picker says when a file doesn’t open.
• A picture too large to save in this browser is flagged before processing, with a button that turns on Limit Resolution. (The limits were measured on a Mac in Chrome and WebKit; the iPhone limits come from WebKit’s source, not a device.)
• Docs checks that a picked file is a PDF, and says so if it isn’t, if it’s locked, or if it can’t be read. Claims left out of the PDF are named, with the reason.
• Docs: pasted text can be in Greek, Cyrillic, Armenian or Georgian letters, and characters Docs can’t print are named before the build. The made-up claims it writes are in English.

Changed

• The Docs button says “Foxglove it”, like the button on Images.
• Large batches use less memory: the app no longer keeps a full-size decoded copy or a base64 copy of each picture, and the grids show small previews. (Measured on a Mac in Chrome and WebKit, not on an iPhone.)
• While a batch is being processed, new pictures aren’t added, and the app says so. Pictures still opening when you tap Foxglove it are processed with the rest.
• Docs keeps single line breaks in pasted text. In uploaded PDFs with cropped, print-bleed or rotated pages, the hidden claims and the printed claim now go inside the visible page, and the printed claim reads upright at the bottom (tested on generated test PDFs only).

Fixed

• Picking a PDF over 30 MB no longer leaves the previous PDF in use.
• Inspect labels AVIF files as AVIF. The same file can be picked again in the logo and Inspect pickers.
• The page’s pictures and the Docs tools load at addresses that end in a slash (like /docs/).

A test release of the website. The iPhone app on the App Store is still version 1.1.1 and doesn’t have these changes. Anything marked Testing is new and still being tested. Send ideas and suggestions with Send feedback in the app’s About tab, or by email to info@cweird.com, and if you find a bug, let me know and send a screenshot if you can!

New

• A record in every file. Every saved file carries the name you enter, a copyright line, your terms and a machine-readable “do not train” statement in its EXIF and XMP fields, with a choice of no AI training or search indexing only, a line saying how you made the image (a digital creation, a photograph of real work or a scan), and a receipt with the file’s fingerprint. Location, serial numbers, hidden thumbnails and edit history are dropped; the capture date and camera model are kept unless you turn that off.
• Visible signature Testing — turn it on and Foxglove draws your name (or your own text) and, if you add one, your logo into the picture. Drag to place it; pinch or use Size to resize it.
• Inspect tab Testing — replaces Verify. Open any image to see what its metadata carries (creator, rights, “do not train”, note, camera, date), its fingerprint and any hidden message Foxglove can read from the pixels; paste a receipt to check whether it is the same file.
• JPEG and WebP copies Testing — save PNG, JPEG and WebP in any combination, each with the record and its own receipt. The JPEG also carries the older IPTC copyright fields, and its transparent areas turn white. WebP appears where the browser can save it.
• Files for your own website Testing — pick what AI may do with your site and where it is hosted, and Foxglove writes the files to paste in: robots.txt rules, header rules, page tags, a tdmrep.json and a policy page. They are requests and declarations; only an edge rule such as Cloudflare’s bot blocking turns a crawler away.
• Docs tab Testing — paste text or choose a PDF, and Foxglove makes a PDF with made-up claims hidden in its text and one in small visible print, plus an answer key you keep. Before you get the PDF, Foxglove reads it back with a text extractor to check that the claims are in it. “Check a submission” looks for those claims in text you paste. A tool that reads the pages as images sees only the visible claim. It all runs in your browser, and the answer keys stay there.
• PDFs in the main upload box — A PDF picked in the Images tab’s upload box opens in the Docs tab, one at a time.
• Receipts kept on this device Testing — Each picture you save or share keeps its receipt in this browser on this device. The Inspect tab lists them, downloads them all (JSON and CSV) and says when an opened file matches one. Nothing is sent.
• Feedback form — Send feedback in the About tab opens a form for ideas, suggestions and bugs, with room for screenshots. It sends through FormSubmit to info@cweird.com and says “Sent” only after FormSubmit confirms.

Changed

• Hidden message — paused: it used Foxglove’s own format, which only Foxglove could read, and social media uploads might erase it. It’s switched off while a sturdier mark is planned: TrustMark, an open invisible watermark from Adobe that other programs can read. It isn’t in the app yet. Pictures saved earlier with a hidden message can still be checked in the Inspect tab.
• The noise pattern (Perturbation) is gone: the website no longer adds a pattern to the pixels.
• Before/After reads “Original / Foxgloved”.
• The text on the site describes what each feature does. Statements about effects on AI training and the comparison with Nightshade are removed.

A major bug made it into the June 9 web release. If you foxgloved an image, downloaded it, and re-uploaded it to Verify, you could get "No message found." This release addressed the two causes below.

Fixed

"No message found" on freshly foxgloved images. Two causes. First, the v1.1 selective encoder was so strict about avoiding visible artifacts that most real artwork didn't have enough textured regions to carry the message at all — and the embed was skipped. Second, on very dark or very bright flat regions, the smooth block-edge blending couldn't push pixel values past pure black or pure white, so some embedded bits read back flipped.

A long message plus artist credit could never embed. The hidden message can carry 124 characters total, but the message box allowed 280 — and your artist credit is appended on top of that. With a name and handle filled in, even the default message exceeded what any image can physically hold, so embedding failed on every image, in every mode. The message is now trimmed to fit, cutting your own text before your artist credit, and the character counter shows the remaining budget as you type or fill in your credit.

Changed

• The embed is now checked before you get it. Foxglove reads the message back from the processed pixels before reporting it embedded. If the textured-only embed doesn't fit or doesn't read back, it allows flatter regions, and if no attempt reads back, the results screen says the message couldn't be embedded.
• Visible signature is applied before the hidden message, so adding your signature can no longer overwrite parts of the message.

The trade-off, updated: v1.1 chose "no visible artifacts" over "message always embeds," and skipped flat images. That was the wrong balance — the message is the point. Now, by default, the textured-only embed is tried first, and only if it doesn't fit or doesn't read back does Foxglove fall back to embedding in flatter regions. If no attempt reads back, the notice on the results screen says so.

Fixed

Yellow/colored patches on flat artwork. If your art had skies, white backgrounds, or large flat color regions, the hidden message was creating faint visible patches. The encoder was trying to embed the message in every part of the image — including flat areas where any modification has nowhere to hide. In this build the encoder embedded only in textured regions, and a mostly flat image got a notice on the results screen that no hidden message was embedded. (1.1.1 changed this; see above.)

PNG transparency was being destroyed. Transparent PNGs (logos, illustrations with cutouts) were losing their transparency. Two stacked bugs: our pipeline was forcing every pixel to fully opaque, and the iOS save flow was routing through a third-party library that strips alpha. This build stopped forcing pixels opaque, which keeps transparency in web downloads. The iPhone save path that keeps transparency was connected in 1.1.1 (Build 8).

Added

• Max mode toggle — lets the hidden message use flatter regions. Off by default. It may show faintly on plain backgrounds.
• Proactive feedback — when a hidden message can't be embedded, the results screen tells you, including how many images in a batch were affected.
• "Coming next" card in About — Style cloaking, Android version, JPEG-resilient stego, ML-powered perturbation (since removed).
• Bug report link in About — pre-fills your version and platform.
• Splash screen with name and tagline.
• Verify tab now lists possible reasons when no message is found, instead of only saying "no message detected."

Fixed

• Crash when accessing camera (missing camera usage description).

Changed

• App Store subtitle changed.

Added

• Block-based steganographic encoding for hidden messages.
• "Always on" labels for metadata stripping.
• Better error handling — no more stuck button if processing fails.

Fixed

• PNG description metadata wasn't being embedded (silent IEND search bug).
• Character counter on the poison message showed wrong default value.