app.ionelul.com tools

Turns down background noise (fans, traffic, hum, keyboard) in a recording and keeps the voice, with RNNoise on this machine. Takes any audio or video ffmpeg reads; writes WAV, MP3, M4A, FLAC or OGG by the output extension (default <name>-clean.wav), or, for a video, the same video with the cleaned sound when the output is a video file. The sound comes out mono. Needs ffmpeg installed.

Makes a barcode as SVG (sized in mm, prints at exact size) or PNG (at a DPI, written into the file): EAN-13, EAN-8, UPC-A, UPC-E, ISBN, Code 128, Code 39, Code 93, ITF-14, Interleaved 2 of 5, GS1-128, Codabar, Data Matrix, GS1 Data Matrix, PDF417 and Aztec. EAN/UPC check digits are added when left out.

Encodes text or any file to Base64 (optionally URL-safe or wrapped at 76 columns, e.g. for data: URLs and e-mail), decodes Base64 back to text or to a file, and percent-encodes or decodes URLs.

Finds and decodes every QR code, Micro QR, Data Matrix, Aztec, PDF417 and 1D barcode (EAN, UPC, Code 128/39/93, ITF, Codabar, DataBar) in a PNG, JPEG, GIF or BMP image, and explains the content: Wi-Fi login, contact card, event, payment, GS1 fields, product code or link (with its domain).

Converts a colour (hex, rgb(), hsl(), hwb(), hsv(), cmyk(), lab(), lch(), oklab(), oklch(); not CSS names) to every format. With `background`, gives the WCAG 2 contrast ratio, the AA/AAA verdicts and, when it fails, the nearest colour that passes. Can add a tint/shade scale and harmonies.

Converts an amount between currencies at the European Central Bank reference rates (published each working day, via the Frankfurter API). Only the rate list is fetched; the amount stays on this machine.

Typesets a CV with XeLaTeX (TeX Live 2026) on this machine and writes the PDF. Give structured `cv` data for the Classic, Modern or Academic template (Academic: Latin Modern with small caps, a skills table, education grouped by school), or your own LaTeX in `tex`/`texFile`. Projects and awards take one line as a paragraph, several as bullets. Only bundled packages are available (fontspec, geometry, xcolor, enumitem, titlesec, hyperref, tabularx, multicol, paracol, fontawesome5 and their dependencies); fonts: Linux Libertine, Biolinum, Latin Modern. The first call downloads the engine (~21 MB) from app.ionelul.com and caches it; the CV is never sent anywhere.

Finds personal data in a .docx or text file (or in `text`) and replaces it: names after titles and court roles (Mr., Frau, domnul, reclamantul…), addresses, national ID numbers checked by their control digit (CNP, PESEL, EGN, TC Kimlik, DNI/NIE, codice fiscale, NIR, SSN, NINO, rodné číslo), ID card and passport numbers, dates of birth, e-mails, phones, IBANs and cards, licence plates and IPs; companies and places on request. The same person always gets the same label ([Person 1]). Word files keep their formatting; author fields are cleared. Rules only (no AI model), so names with no title before them are found only when they also appear after one, or via `terms`. The originals are never returned: only counts, so the result can be shown to a model safely. Run it before reading a confidential file.

Reads a Romanian e-Factura (UBL 2.1 XML, or the .zip downloaded from ANAF SPV) and checks that its totals add up (EN 16931 BR-CO rules). Does not validate against the ANAF schema.

Turns an e-Factura XML (or every invoice in an ANAF .zip) into a readable A4 PDF.

Makes the usual favicon set from a logo (JPEG, PNG, WebP, AVIF or JPEG XL): favicon.ico (16, 32, 48 px), 16/32 px PNGs, apple-touch-icon, the Android 192/512 icons and site.webmanifest, centred on a square with optional padding and background, plus the <link> tags to paste into <head>.

Hashes files (streamed, so large files are fine) and optionally verifies them against an expected hash or a checksum file (sha256sum, md5sum or BSD format).

Calories and macros of a packaged product from its EAN/UPC barcode (8, 12, 13 or 14 digits), from Open Food Facts. Only the code is sent. Use code_read first to get the code from a photo.

Looks up everyday foods (fruit, vegetables, grains, dairy, meat, fish, legumes, nuts, fats, sweets, drinks, dishes) in a built-in list of about 170, with Romanian dishes, and gives their calories and macros per 100 g and a usual portion. Names match in the chosen language, in English or in any other. Offline.

Joins JPEG, PNG, WebP, AVIF or JPEG XL images into an animated GIF, in the order given, each fitted (or cropped to fill) on a frame of one size.

Re-encodes JPEG, PNG, WebP, AVIF and JPEG XL images with the Squoosh codecs (MozJPEG, OxiPNG, libwebp, libavif, libjxl): convert between formats, compress at a quality, and shrink to a maximum width/height or a percentage (Lanczos3, aspect kept, never enlarged). Metadata is not carried over.

Shows what a photo reveals: GPS position (with a map link), camera and lens, date taken, author and copyright, software and exposure, from EXIF, XMP and IPTC in JPEG, PNG, WebP, HEIC, AVIF and TIFF. Use image_strip_metadata to remove it.

The main colours of a JPEG, PNG, WebP, AVIF or JPEG XL image (k-means in OKLab, transparent pixels ignored), largest share first, as hex, RGB and OKLCH.

Cuts the subject (people, products, animals, objects) out of a JPEG, PNG, WebP, AVIF or JPEG XL photo with the ormbg AI model, on this machine. Writes a transparent PNG, WebP or AVIF, or the subject on a new colour; can crop to the subject. The first call installs Transformers.js with ONNX Runtime (about 450 MB) and downloads the model (45 MB) into the cache.

Removes EXIF (GPS, camera, dates), XMP, IPTC and comments from JPEG, PNG and WebP without re-encoding: pixels stay byte for byte. Rotated photos keep only their orientation.

Traces a JPEG, PNG, WebP, AVIF or JPEG XL into an SVG of filled paths (imagetracerjs). Best for logos, icons, drawings and scans with few colours. Presets: logo (8 colours), detailed (48), mono (black and white), gray, smooth.

Enlarges JPEG, PNG, WebP, AVIF or JPEG XL images 2x or 4x with Real-ESRGAN (AI that restores detail), on this machine. Transparency is kept. The first call installs ONNX Runtime (about 113 MB) and downloads the model (5 MB) into the cache.

Puts images into one PDF, one per page, in the order given: A4 or Letter (orientation from each image, or fixed) with a margin, or pages the size of each image. JPEG and PNG go in untouched; WebP, AVIF and JPEG XL are converted to JPEG at quality 90.

Makes an A4 invoice, proforma or quote PDF: seller and buyer, dates, line items with VAT rates, totals with VAT per rate (rounded per rate, as accountants check them), payment details, notes and an optional PNG/JPEG logo. Labels and number/date formats in 14 languages. Not an e-Factura (no XML is made).

The last day of a procedural or legal time limit, with each counting step explained. Systems: ro-civil (Romanian Civil Procedure Code art. 181-183, clear days: neither first nor last day counts), ro-criminal (CPP art. 269, same), ro-custody (CPP art. 271, preventive measures: both ends count, no holiday extension), ro-civil-code (Civil Code art. 2551-2553, limitation periods), eu (Regulation 1182/71), generic (calendar or working days). A last day on a weekend or public holiday moves to the next working day. Public holidays are computed for any year for: RO, BG, DE, AT, FR, BE, NL, ES, IT, PT, PL, CZ, SK, HU, GB, IE, US (GB = England and Wales, US = federal).

Writes a ready-to-send letter citing the right articles: GDPR requests (gdprAccess art. 15, gdprRectification art. 16, gdprErasure art. 17, gdprObjection art. 21, gdprPortability art. 20, gdprConsent art. 7(3)), a formalNotice to pay (in Romanian it cites art. 1522 Civil Code and art. 1014 CPC) or a powerOfAttorney, in any of 14 languages, as PDF, .docx or plain text. Fields left empty show as ________ and are listed in the reply.

Builds a text logo as SVG: a shape with initials (circle, rounded, square, hexagon, diamond, shield) beside or above the name and tagline, or the text alone. Colour presets: forest, ocean, sunset, mono, royal; any colour can be overridden. The text uses common system fonts, so it may look slightly different on other machines.

Placeholder text: paragraphs, sentences or words, optionally starting with the classic "Lorem ipsum dolor sit amet" and wrapped in <p> tags.

Converts Markdown (GitHub flavour: tables, task lists, fenced code) to an HTML fragment, or to a complete styled HTML page titled after the first heading.

Adds up the calories and macros of foods from food_search (by id) in grams or portions, item by item and in total. For packaged food, pass its values per 100 g from food_barcode as a custom item.

Daily calories, protein, carbohydrates, fat, fibre and water for an adult, from the Mifflin-St Jeor resting energy times an activity factor, with 250-750 kcal taken off to lose weight or added to gain (never below 1200 kcal for women, 1500 for men). Also BMI. Estimates, not medical advice.

Reads the text in JPEG, PNG, WebP, AVIF or JPEG XL images (scans, screenshots, photos of documents) with Tesseract on this machine. Languages: eng, ron, deu, fra, spa, ita, hun, por, nld, pol, ces, bul, ukr, tur (combine them, e.g. ["ron","eng"]). Low-confidence guesses are dropped; Romanian ş ţ become ș ț. The first use of a language downloads its model (2-15 MB) from app.ionelul.com and caches it; images never leave this machine.

Converts .docx, .xlsx, .pptx or .odp to PDF without Office installed. Layout is redrawn, so it is close to but not identical with Office output; GIF/WebP/BMP pictures show as placeholders.

Generates passwords, passphrases (EFF long word list, 7,776 words) or PINs with a cryptographically secure random generator, without modulo bias, and reports their entropy and strength.

Adds a text watermark and/or page numbers, sets metadata, and protects with a password. With `password` and no `protect`, the output is an unlocked copy.

Writes values into fillable fields (names from pdf_form_read). Diacritics are supported.

Lists the fillable (AcroForm) fields of a PDF with their type, current value and options.

Page count, page sizes, metadata, encryption and form fields of a PDF.

Joins PDFs into one, in the order given. Each input can pick its pages ("1-3, 5, 8-").

Builds a new PDF from the pages listed, in that order (pages left out are deleted), optionally rotating some of them.

Places a signature, stamp or logo image (a transparent PNG works best; JPEG, WebP, AVIF and JPEG XL too) on chosen pages, at a corner or at exact millimetres, keeping its proportions. This is a visible signature, not a cryptographic (digital certificate) one.

Splits a PDF into several files: one per range ("1-3, 4-6") or every N pages.

Finds tables in a PDF from its text layout and writes them to .xlsx (or .csv). Works on statements, price lists and reports; not on scans.

Rebuilds the text of a PDF as an editable .docx: paragraphs, headings, bold and italic. Tables and columns come out as paragraphs; scanned pages are skipped.

Makes a QR code as SVG (styled dots and eyes, any size) or PNG (square modules) from text, a link, Wi-Fi login, e-mail, phone, SMS, contact card, location or calendar event. Fields per type: url: url; text: text; wifi: ssid, password, security, hidden; email: to, subject, body; phone: phone; sms: phone, message; vcard: firstName, lastName, phone, email, company, title, website, street, city, zip, country; location: lat, lng; event: title, start, end, location, description. UTF-8, so diacritics and emoji work.

Suggests built-in recipes (breakfast, lunch, dinner, snack; Romanian dishes included) whose serving fits the calories left, preferring protein-rich ones, with nutrition per serving, ingredients and steps in the chosen language.

Writes a JavaScript regular expression, without AI: from examples of what should (and optionally should not) match, worked out run by run and checked on every example; or from a description, picked from a library of common patterns (email, URL, IP, IBAN, Romanian CNP/CUI/plates/phones, dates, times, money, UUID...); or from a format such as "DD.MM.YYYY" or "AA 99 AAA" (9 a digit, A an upper-case letter, a lower-case, X letter or digit, DD/MM/YYYY/HH/mm date and time parts, \ for a literal).

Runs a JavaScript regular expression on a text: every match with its position and capture groups (numbered and named), and optionally the text with a replacement applied.

Adds missing Romanian diacritics (ă â î ș ț) from a dictionary of ~70,000 words, fixes the old cedilla letters (ş ţ to ș ț), or removes diacritics. With `add`, words that can be spelled more than one way get the most common form and are listed with their alternatives.

Transcribes speech from any audio or video file ffmpeg can read with OpenAI Whisper, on this machine: plain text, SRT or WebVTT subtitles, or text with timestamps. About 100 languages, detected automatically; can also translate the speech into English. Models: tiny (41 MB), base (77 MB), small (249 MB); bigger is slower and more accurate. Needs ffmpeg. The first call installs Transformers.js with ONNX Runtime (about 450 MB) and downloads the model into the cache.