Small improvements.
Useful changes.
Dated notes describe the build at that time. Later entries may extend earlier features or limits.
· All 38 tools, in seven languages
Every tool now has an interface in English, Simplified Chinese, Hindi, Spanish, French, Arabic and Bengali, including instructions, limits and recovery messages. The homepage, guides, privacy page and changelog also follow your language. The license overview links to the original notices.
Interface language does not translate your files or text. PDF-to-Word supports selected printed-text OCR models, including these seven languages; recognition can still misread characters and numbers. Review the source before exporting. Local browser checks cover the translated workflows; native-language, screen-reader and physical-device review remain open.
· Smaller PDFs with repeated images
Preserve quality and Balanced compression can now share identical embedded images and compatible transparency masks. Repeated image data is stored once when its bytes and supported settings match, without changing those pixels. Balanced can still optimize eligible JPEGs separately.
Sharing is on by default and can be switched off. Images with different settings, identity-related metadata or unsupported reference uses stay separate. The result reports how many image streams were reused; original downloads, size targets and page comparisons remain available.
· Keep your place with the keyboard
Image and PDF queues now keep keyboard focus with moved files and move it to a useful neighboring control after removal. PDF page controls follow pages across thumbnail groups; deleting every page leaves Restore ready to use. Word structure controls follow blocks between columns, and new paragraphs and tables open ready to edit.
Polite status messages describe the changes. These workflows have keyboard and automated accessibility checks; screen-reader and physical-device verification remain open.
· OCR in six languages
PDF-to-Word now reads printed English, French, German, Italian, Portuguese and Spanish scans. Choose one language for the selected pages; only that model loads from this site when OCR is needed. Recognition stays on your device, with editable text and structure review, DOCX and TXT downloads.
Language changes clear the previous result. Model-download failures now report an error promptly so you can retry. OCR can still misread accents, punctuation and numbers; compare each page before exporting. Language detection, translation and handwriting recognition are not provided.
· Choose HEIC photo metadata
HEIC-to-JPG can now keep the primary photo’s EXIF, including camera settings, capture dates and GPS when present, or remove source metadata (the default). Retained orientation, size and color tags describe the exported JPG. Standard EXIF thumbnails are removed; separate XMP items are not copied. Unsupported, damaged or oversized EXIF produces an explicit preservation error; metadata removal remains available.
· Text history and reviewed-text handoff
Text input editors now offer undo and redo for typing, find and replace, examples, file imports, and using a result as input. Keyboard shortcuts stay within the editor. History stays in memory with limits of 20 steps and four million retained characters; Clear discards it.
PDF-to-Word review can send its edited text directly to Word Counter, using the same selected plain-text or structure version and page order as the TXT download. The one-time transfer stays in tab memory and respects the counter’s one-million-character limit.
· PDF size targets and grayscale
PDF compression now offers an optional file-size target. It tries bounded quality settings from the original, reports actual bytes and whether the target was met, and retains the smallest tested result or original. Your selected compression mode stays in control. Download the untouched original alongside a smaller result.
Optional grayscale affects optimized JPEGs in Balanced mode or every page in Image-only mode. Size targets cannot guarantee a particular result; inspect readability and any meaning carried by color.
· Word fonts and text colors
PDF-to-Word structure mode now recovers supported font-family names and solid text colors in paragraphs and table cells. Review mixed styles, choose a font and color, or apply underline. Text edits retain unchanged styles. Missing font metrics use approximate text boxes so headings are retained.
Fonts are not embedded, and Word may substitute unavailable families. Transparent or patterned text and drawing text that cannot be matched reliably need manual color review. Source comparison remains essential.
· Editable Word headings and lists
Supported tagged PDF headings now retain their levels, and simple tagged lists export as native Word bullets, numbers, letters or Roman numerals. Nested items, continuation paragraphs and visible starts can be reviewed before download. Lists crossing source pages retain their visible numbering as separate lists. Custom labels remain editable text.
Independent W3C and PDF Association examples exercise headings, nested restarts and a list continuing at 23 on page two. Original typography, complex numbering and complete tagged reading order still need broader support.
· Network diagnostics with real measurements
IP observation, DNS checking and a bounded speed test now join the toolkit. DNS compares Cloudflare and Google responses with record values, TTL, response status and DNSSEC details. Reports can be copied or downloaded. Geographic probes require configured endpoints with verified runtime regions.
Speed tests begin only after Start, transfer capped synthetic data, count actual download and acknowledged upload bytes, and report HTTP latency before and during transfers. Cancellation stops active work. Local endpoints are labeled as local throughput; deployed Internet performance, ingress trust and IPv6 coverage still need verification. File and text processing continues to stay in your browser.
· Reviewable Word structure
PDF-to-Word now offers structure reconstruction with editable paragraphs, basic styles, one- or two-column regions, tagged or simple aligned tables, merged table headers and supported figure crops. Review blocks, edit cells, change styles, reorder content and crop missing figures from the source before exporting real DOCX content or text.
Upright printed-English scans can use recognized word positions to reconstruct simple tables. Source page dimensions join Letter and A4. Exact fonts, colors, positioning and pagination are not preserved; complex layouts and broader OCR support remain development work. Plain text and structure keep separate review edits.
· More control over PDF pages
Split, extract, rotate, organize, and PDF-to-images tools now join the toolkit. Merge shows page counts and offers thumbnail ordering, duplication, removal, and rotation. Split supports custom groups, fixed page counts, and odd/even separation with individual or ZIP downloads.
PDF compression adds named presets and a rendered original/result comparison. Page-image export offers PNG or JPG, resolution, quality, and grayscale. Bounded local previews respect page crop bounds, rotation, and physical units; rendering errors produce an explicit failure instead of silently omitting damaged image content.
· Images, from selection to export
New image crop, rotate/flip, compression, and WebP-to-PNG tools join the toolkit. Image resizing now supports percentages, dimension presets, ratio locking, and crop positioning. HEIC and WebP conversions add maximum dimensions, source previews, and actual file-size comparisons.
JPG-to-PDF adds automatic orientation, per-image rotation, filename sorting, a page-layout preview, separate PDF files with ZIP download, and optional PNG embedding for transparency. Fit and fill choices make page cropping explicit. All of these workflows process files locally.
· Colors and device checks, with more detail
New contrast, CSS gradient, and image palette tools join the toolkit. Color conversion now preserves decimal alpha in RGB and HSL, with editable channels and local swatch exports. Image sampling adds pixel coordinates, keyboard movement, magnification, paste support, and an explicit transparency backdrop.
Microphone tests now show a waveform, digital RMS/peak levels, reported processing settings, and downloadable recordings. Camera tests add preferred resolution, observed preview FPS, mirrored PNG/JPG snapshots, and fullscreen. Keyboard tests include navigation/numpad layouts, simultaneous-key counts, lock-state snapshots, and local reports.
· More room for text and code
Five new local tools handle Base64, URL encoding, SHA hashes, UUIDs, and text cleanup. The word counter now includes keyword density, speaking estimates, and word goals. Case conversion has 13 styles, JSON has a searchable tree with copyable pointers, CSS can be beautified, HTML has line-width and attribute controls, and Lorem Ipsum supports sentences and a classic opening.
Text workspaces now share literal find and replace, line wrapping, downloads, and a use-result-as-input action. Changes to input or settings clear previous results so they cannot be confused with a new operation.
· PDF text, ready for Word
A local text-conversion beta reads existing PDF text and printed English scans. Review each source page, correct the text, then export editable DOCX paragraphs with optional page breaks. Original tables, images, fonts, and complex layouts are not recreated. PDF results can continue directly into this review workflow.
· HEIC photos, made shareable
Convert the primary HEIC or HEIF still photo into a JPG locally, with quality controls, batch ZIP downloads, and direct handoffs to image resizing or PDF assembly. Supported color profiles are converted for SDR output. HDR and unsupported inputs receive explicit messages.
· Local PDF compression
Compress PDFs with preserving modes or an explicitly selected image-only rebuild. The result shows before/after size, and an unchanged original is returned when the candidate would not be smaller. New PDFs and merged documents can continue directly into compression.
· First development build
The ToolMellow workspace begins with local image resizing and conversion, PDF assembly, text/code tools, color tools, and device checks. Each tool has a dedicated page with instructions and explicit limits.
- In-memory handoffs from image results to JPG to PDF, and from new PDFs to Merge PDF.
- Processing workers with cancellation and bounded input sizes.
- Strict JSON formatting that preserves numeric literals and warns about duplicate keys.
- Account-free use and self-hosted processing assets.
On this date the site was a development preview, closed to search indexing. It opened to search engines on 1 October 2026. This page does not claim completed cross-browser certification.