AI label check See whether a picture says it was made or edited with AI: Content Credentials (C2PA), the IPTC source type, the generator’s name and saved prompts. Read in your browser; nothing is sent.
Developer tools v0.15.0
What’s new in Developer tools Close Version 0.15.0 latest The encoder and decoder knows three more formats: Base58 (as in Bitcoin addresses), Ascii85 (Base85, as in PDF files, with or without its <~ ~> marks) and Z85 (ZeroMQ). Each has its own page, for example /encode-decode/base58. The hash generator makes SHA3-256 and SHA3-512 hashes too, with HMAC and a salt like the others, worked out in your browser. Pages: /hash/sha3-256 and /hash/sha3-512. Version 0.14.0 The AI label check now reads the picture's pixels too, in your browser: - Stable Diffusion's invisible watermark (SD 1, SD 2 and SDXL from their official tools): when it is found, it is a strong sign of AI, even after the labels were removed. - The trace a camera sensor leaves in the colours of a real photo; resizing and compression wipe it, so only finding it counts. "Ask the model": an open AI detector (MIT licence) runs in your browser and estimates how likely the picture is AI. Your picture is not sent; the model (about 15 MB) is downloaded once from Hugging Face. It is right about three times in four, so it counts only as a weak clue. Version 0.13.1 The AI label check says when a file carries no metadata at all (re-saved, converted, edited or a screenshot), so an "unclear" result explains itself: the file has nothing left to say. Version 0.13.0 The AI label check sums up its clues: "leans towards AI", "leans towards a real photo" or "unclear", with a signed AI step or a trusted camera signature deciding on its own. Expert views for pictures: a noise map and a frequency spectrum, drawn in your browser from the centre of the picture, with what to look for. An aid, not a verdict. Videos (MP4, MOV, up to 200 MB): their signed Content Credentials are checked too. Version 0.12.1 The AI label check takes a picture dragged and dropped onto it, as well as a chosen one. Version 0.12.0 The AI label check now verifies Content Credentials: whether the signature still matches the file, who signed it (OpenAI, Adobe, a camera maker…), the tool, and whether a step was made with AI. A valid camera signature counts towards a real photo. More clues, each marked strong, medium or weak: camera data and exposure, sizes AI generators often use, and a saved editing history. "Look further" links to Google SynthID, Google Lens and TinEye. Version 0.11.0 New tool: the AI label check. Choose a picture to see whether it says it was made or edited with AI: Content Credentials (C2PA), the IPTC source type, the generator's name or a saved prompt (Stable Diffusion, ComfyUI). Only the file's labels are read, in your browser; finding none does not prove a person made it, and the page says so. Version 0.10.1 The control panel shows the tools' numbers again: uses, results copied or downloaded and errors in the last 30 days, signed-in members using the tools, feedback waiting, and saved settings. Version 0.10.0 The file converter reads and writes Excel workbooks (.xlsx). Open a workbook, choose its sheet (or all sheets at once) and turn it into JSON, CSV, YAML, XML, Markdown, HTML or SQL; or make an Excel file from your data, with numbers kept as numbers and the first row frozen. Formulas are read as their last results and never calculated; macros never run. Dates come out as dates (2026-10-24), and merged cells, pictures and macros are pointed out. RSS and Atom feeds become a clean list of entries (title, link, published, author, summary), the same whatever the kind of feed, with dates in one standard form. New pages such as /convert/xlsx-to-json, /convert/csv-to-xlsx and /convert/xml-to-csv. Version 0.9.0 New tool: the file converter. Turn JSON, CSV, XML and YAML into each other, or into a Markdown table, an HTML table or SQL INSERT statements. The format of what you paste or open is recognised, and so is the CSV delimiter (comma, semicolon, tab or pipe). Nested data becomes columns such as address.city and comes back nested; numbers, true/false and empty cells become values only when nothing is lost, so 007 and very large numbers stay exactly as they were. XML attributes, repeated elements and CDATA are kept; entities from a DOCTYPE are never read. See the result as text or as a table, copy or download it, use it as the next input, or open a JSON or SQL result straight in its own tool (passed inside your browser, never sent). Every pair has its own page, for example /convert/json-to-csv or /convert/csv-to-sql. Version 0.8.0 Compare two SQL scripts at /sql/compare: the differences line by line, with the changed words inside a line marked, side by side or one under the other. Jump from difference to difference, and see how many lines were added, removed and changed. By default the formatted scripts are compared, so a different layout does not count. You can also ignore upper and lower case, spaces and comments. Presets for the SQL formatter: start from a built-in style (compact, leading commas, lower case, tabs) or save your own rules under a name. Export them as a small file and import your team's. Presets stay in your browser. Version 0.7.0 New tool: the SQL formatter. Paste a query or a whole script and it is laid out clause by clause, with subqueries, CASE and procedure bodies indented. MySQL, T-SQL (SQL Server), PostgreSQL and standard SQL are recognised by themselves, or you choose one; strings and comments stay exactly as you wrote them. Fold any block with the arrow on its left, or fold everything at once. Comment a block out with "--" and back in with one click, or add your own comment above any line. Choose -- lines or /* */ blocks under Options. Search the result as text, as a whole word or with a regular expression, and jump from match to match; folded blocks open by themselves. Choose upper or lower case keywords, the indent, where commas and AND / OR go, and blank lines between statements. Copy the result or download it as a .sql file. Own pages: /sql/mysql, /sql/tsql, /sql/postgresql and /sql/standard. Version 0.6.1 The toolbar and the row of tools now stay at the top while you scroll, so every tool is one click away from anywhere on the page. Version 0.6.0 New tool: the image decoder. Paste Base64, a data URI, hex or Base32, or a whole HTML, CSS or JSON snippet: the pictures inside are found, shown on a checkerboard and named by their real type, with their size and dimensions. A wrong claimed type is pointed out, and a file that is not a picture (a PDF, a ZIP) is named and can be downloaded. Download the picture, copy it as a data URI, or convert it to PNG, JPEG or WebP with the quality you choose. "Image to Base64": choose a picture and get its Base64, data URI, HTML <img> and CSS url(). Own pages: /image/base64-to-image and /image/image-to-base64. Version 0.5.0 New tool: the hash generator. MD5, SHA-1, SHA-256, SHA-384, SHA-512 and CRC32 of text or of a file (up to 100 MB, read in your browser), several at once, as hex, HEX, Base64 or Base64URL. Add an HMAC key or a salt, or derive a key from a password with PBKDF2. Paste an expected hash to see whether it matches. Keys and salts are masked and never leave the page. Every algorithm has its own page, for example /hash/sha256 or /hash/hmac. A row under the toolbar now links to every tool, so you can switch in one click. The usage statistics page for admins opens again. Version 0.4.0 Check a JWT's signature: paste the shared secret (as text, Base64 or hex) for HS256, HS384 and HS512, or the public key as PEM or JWK for RS, PS, ES and EdDSA tokens. You see at once whether the token was signed with that key and left unchanged. The secret or key stays in the page: it is masked, never saved and never sent. Version 0.3.0 Every format of the encoder and every JSON feature now has its own page (for example /encode-decode/base64, /encode-decode/jwt, /json/validator), so you can go straight to it. Something did not work? Each tool has a "Tell us" form. You can choose to include a sample of your input; it is off by default, you see and can edit exactly what is sent, and it is kept for 90 days for the DevQuake team only. We now count which tools and buttons are used and whether they worked, to improve them. What you type, paste or open is still never sent. Version 0.2.0 New tool: the encoder and decoder. Paste something encoded and it recognises the format and decodes it, saying how sure it is: Base64, Base64URL, Base32, hex, URL encoding, HTML entities, Unicode escapes, Punycode domain names, data URIs, Basic authorization headers and JWTs. When it cannot tell, it says so and lets you choose the format, with the reasons the nearest ones did not fit. Results that are encoded again can be decoded again, step by step. Encode text into any of these formats, with options such as the text encoding, Base64 padding and line length, or how hex bytes are separated. A JWT shows its header, its content and its times in your time zone, and warns when it has expired or has no signature. A URL is taken apart, with a table of its parameters you can edit. Decoded pictures and files show a preview or their type, and can be downloaded. Version 0.1.0 The developer tools are here, starting with the JSON formatter and validator: format, minify and check JSON, see the exact line and column of a mistake, and explore it as a tree whose paths you can copy. Everything runs in your browser: nothing you type, paste or open is sent anywhere or stored. Your browser remembers only your options, and when you are signed in you can keep them in your DevQuake account. All tools About DevQuake