# Feyard — Full Content > A curated matrix of 26 free, browser-first tools — AI image, 3D, CAD, audio, video, text, PDF and OCR utilities plus Sudoku. AI processing runs on-device via WebGPU with WASM fallback; classic pipelines run entirely in local TypeScript — no servers, no sign-up, no file uploads. Feyard is an indie-built suite of micro-tools that run entirely in the browser. Each tool does one thing superbly: background removal, image compression, upscaling, photo restoration, photo-to-3D depth reconstruction, web page cleaning, speech-to-text, subtitle generation, text-to-speech, vocal removal and denoising, PDF RAG search, PDF summarization, PDF tooling (merge/split/rotate/compress/reorder/encrypt/watermark/OCR/render/assemble), OCR, bitmap vectorization, CAD tracing, 3D model optimization, point-cloud downsampling and conversion, font inspection and subsetting, AI text humanizing, and Sudoku. Powered by Transformers.js, WebGPU, three.js, Depth Anything V2, Tesseract.js, Kokoro-82M, Whisper, Swin2SR, RNNoise, @jsquash WASM codecs, and a local-first architecture. Available in 8 languages: English, 中文, Deutsch, Français, Español, Italiano, Nederlands, Português (BR). --- ## Remove BG AI background removal. 100% on-device via WebGPU. Three model tiers, an alpha-safe edge brush, batch ZIP export. A pure client-side AI background remover. No uploads, no servers, no sign-up — your photos never leave your device. Full-resolution output, no watermark, free for commercial use. ### Features - **100% private**: Your images never leave your device. No uploads, no servers, no tracking. - **3 model tiers**: Fast (MODNet ~26 MB) for portraits and mobile; Sharp (BiRefNet Lite ~183 MB) for products, animals, logos and hair; Sharp · GPU runs the same BiRefNet Lite weights on WebGPU — identical edges, usually several times faster. - **Alpha-safe edge brush**: Touch-ups write into the alpha channel, so they survive a background swap, a shadow or a re-export at 4K. Feather, shrink/grow and contrast sliders cover the rest. - **Batch to ZIP**: Up to 500 images in one run, exported as a single ZIP with your choice of background, format and output size. ### How it works 1. **Drop an image**: Drag & drop, paste, or browse. PNG, JPEG, WebP, AVIF or BMP. 2. **AI removes the background**: The model runs locally on WebGPU or WASM and produces a precise alpha matte. 3. **Refine and compose**: Feather, shrink/grow and contrast sliders, a manual brush with undo/redo, 8 backgrounds, 5 shadows and 8 aspect ratios. 4. **Download**: Transparent PNG, JPEG or WebP at up to 4096 px, or copy the cutout to the clipboard. ### FAQ - **Is my image uploaded to a server?** No. Everything runs in your browser. The model downloads once from our CDN and is cached locally. - **Which tier should I pick?** Fast (MODNet) for portraits and mobile; Sharp for products, animals and hair; Sharp · GPU for the same edges faster when your browser supports WebGPU. - **How long does it take?** A typical photo processes in 1–3 seconds with WebGPU. The first run downloads the model — ~26 MB for Fast, ~183 MB for the Sharp tiers. - **Is the output free for commercial use?** Yes — no watermark, no credits, no export cap. The weights are MODNet (Apache-2.0) and BiRefNet Lite (MIT). URL: https://feyard.com/products/remove-bg/ --- ## Upscaler AI image upscaling & denoising. 2×/4× via WebGPU, batch queue, two engines. Two on-device AI engines — Swin2SR for photos and Real-ESRGAN anime6b for anime, illustrations, and line art (Pro) — with 2×/4× upscaling and denoise, batch processing, optional GFPGAN face enhancement (Pro), and PNG/WebP/JPEG export. Nothing is uploaded. ### Features - **100% private**: Your images never leave your device. All processing happens locally. - **WebGPU accelerated**: Inference runs on your GPU when available, with automatic WASM fallback. - **Two AI engines**: Swin2SR rebuilds noisy real-world photos (trained on BSRGAN degradation); Real-ESRGAN anime6b sharpens anime, illustrations, and line art (Pro). - **2× / 4× modes**: Choose 2× for speed or 4× for maximum detail. Denoise-only mode also available. - **Batch queue**: Up to 4 images per run on the free plan, 24 on Pro — download everything as one ZIP. - **Face enhance**: Optional GFPGAN v1.4 pass restores portrait detail after upscaling (Pro). - **Formats**: Export PNG, WebP, or JPEG with adjustable quality. ### How it works 1. **Upload images**: Drag & drop, paste, or browse. JPG, PNG, or WebP — single files or whole batches. 2. **Pick a scale and engine**: Choose 2× or 4× super-resolution, Standard (Swin2SR) or Anime (Real-ESRGAN). Models run locally via WebGPU. 3. **Download the results**: Save files one by one or export the whole batch as a ZIP in PNG, WebP, or JPEG with adjustable quality. Models cache for instant reuse. ### FAQ - **Is my image uploaded to a server?** No. Everything runs in your browser. The models download once and are cached locally. - **Which browsers support WebGPU?** Chrome 113+, Edge 113+. Others fall back to WebAssembly. - **How long does it take?** With WebGPU, 1–5 seconds per image depending on size. The first run downloads the model (~200 MB). - **What's the max input size?** Free accounts process images up to 2048 px on the long edge, Pro up to 4096 px — up to 4 images per batch on the free plan and 24 on Pro. URL: https://feyard.com/products/upscaler/ --- ## Restore AI photo restoration studio. Repair, sharpen, colorize, fill & ID photos, on-device. Repair, sharpen, colorize, fill, and reframe — plus batch processing — a full AI photo studio that runs entirely in your browser. Your photos never leave your device. ### Features - **Six tools, one studio**: Restore, sharpen, black-and-white colorization, fill, ID photos, and smart crop — every photo fix in a single page, with batch queues for Restore, Sharpen, and Colorize. - **100% private**: Old family photos are personal. Everything runs on your device — no uploads, no servers. - **WebGPU accelerated**: Inference runs on your GPU when available, with automatic WASM fallback. - **Real AI models**: Swin2SR super-resolution, GFPGAN face revival, LaMa inpainting, DDColor colorization, and BiRefNet/MODNet matting — locally. ### How it works 1. **Drop in a photo**: Drag & drop, paste, or browse. JPG, PNG, or WebP. 2. **Pick a mode**: Restore old photos, sharpen blur, colorize black-and-white shots, brush away objects, swap ID backgrounds, or crop smart. 3. **Compare & download**: Slide between before and after, then export PNG/JPEG/WebP at a chosen quality — or run a whole folder as a batch and download one ZIP. ### Modes - **Restore**: Scratches, noise & color revived (Swin2SR ~200 MB), optional GFPGAN faces (~340 MB) with a 0–100% strength slider, and a brush for scratch/damage inpainting via LaMa (~208 MB). - **Sharpen**: Super-resolution & denoising via tiled Swin2SR 2×/4×. - **Colorize**: DDColor-tiny (~129 MB, Apache-2.0) predicts chroma in Lab space while keeping the original luminance, so output resolution equals input resolution. Free at full resolution, no watermark. Already-color photos are detected and blocked to avoid color shifts. - **Fill**: LaMa inpainting erases objects (~208 MB). - **ID Photo**: Cutouts & passport sizes (BiRefNet/MODNet matting ~176 MB). - **Crop**: Subject-aware smart framing. ### Limits - Restore/Sharpen input: up to 2048 px long edge on Free, 4096 px on Pro (larger drops are pre-scaled with a notice). - Colorize and the scratch-repair brush always run at the photo's full resolution, free. - Batch (Restore/Sharpen/Colorize): up to 4 images on Free, 16 on Pro; serial processing, individual downloads or one ZIP. ### FAQ - **Is my photo uploaded to a server?** No. Every model runs in your browser. Models download once from Hugging Face and are cached locally. - **What does each mode do?** Restore repairs scratches/noise/color and revives faces. Sharpen upscales 2×/4×. Colorize adds natural color to black-and-white photos with DDColor. Fill removes objects via inpainting. ID Photo cuts out subjects. Crop frames automatically. - **How large are the models?** Swin2SR ~200 MB, GFPGAN ~340 MB, LaMa ~208 MB, DDColor ~129 MB, matting ~176 MB. Each downloads only when first used. - **Which browsers are supported?** Chrome 113+ and Edge 113+ get WebGPU. Safari and Firefox fall back to WASM. URL: https://feyard.com/products/restore/ --- ## Depth 3D Photo → 3D motion video + interactive 3D. On-device depth AI, MP4/WebM and GLB/OBJ/STL/PLY/USDZ export. A free 3D photo maker. AI estimates depth entirely on-device, turns any picture into a cinematic parallax video or an interactive relief mesh/point cloud, and exports MP4/WebM, GLB, OBJ, STL, PLY or USDZ — no uploads, no sign-up, no watermark. ### Features - **Cinematic 3D motion**: Four camera presets (dolly in, sway, orbit, pull out) turn a photo into a parallax clip in 16:9, 1:1 or 9:16, with 3–10 s duration and ping-pong looping. Free exports watermark-free 720p/8 Mbps; Pro renders 1080p/16 Mbps. MP4 on H.264-capable browsers (Safari, recent Chrome), WebM elsewhere. - **100% on-device & private**: Depth estimation, mesh building, brush edits and video rendering run in your browser via WebGPU with automatic WASM fallback. Photos and video frames never leave your device. - **Real depth AI**: Depth Anything V2 Small — a state-of-the-art monocular depth model (~50 MB WebGPU, ~27 MB WASM, Apache-2.0) — maps every pixel to scene distance. Free runs 518px depth maps; Pro runs 700px. - **Real 3D files, not just a clip**: Export a textured relief mesh or vertex-colored point cloud as GLB (embedded texture), OBJ, STL, PLY, USDZ (AR Quick Look on iPhone/iPad), plus the grayscale depth map as PNG. - **Manual depth brush (Pro)**: Raise, lower or smooth depth on areas the model under-separated, with restore that paints the estimate back; edits feed both geometry and motion, never the original photo. ### How it works 1. **Add a photo**: Drag & drop, paste, or browse. JPG, PNG, or WebP. 2. **AI estimates depth**: Depth Anything V2 runs locally via WebGPU or WASM in seconds and caches after the first run. 3. **Orbit, animate, export**: Rotate the interactive 3D result and tune the depth, then play a cinematic move and export MP4/WebM, or download GLB, OBJ, STL, PLY, USDZ and the depth PNG. ### FAQ - **What is a 3D photo motion video?** A short cinematic clip where a virtual camera travels through the photo's real depth. The move is deterministic — no pixels are invented, faces and products stay exactly as shot, and nothing is uploaded. - **Is my photo uploaded to a server?** No. Depth inference, mesh building, brush edits and video rendering all happen in your browser; the exported video is a purely local download. The model downloads once and is cached. - **Which video sizes and aspect ratios are supported?** 16:9, 1:1 and 9:16 at 30 fps: 1280×720 / 720×720 / 720×1280 free, 1920×1080 / 1080×1080 / 1080×1920 on Pro. All presets, looping and durations are free. - **What is the difference between free and Pro?** Free includes all presets, aspect ratios, looping, watermark-free 720p video, all 3D formats and 518px depth maps. Pro adds 700px depth, 1080p/16 Mbps motion video and the manual depth brush. - **Is this real 3D like TripoSR or Meshy?** It is honest 2.5D — front-facing geometry from one photo, like classic Facebook 3D photos; unseen backs stay flat rather than hallucinated. In return it runs in seconds on your device and exports standard 3D files. - **Which browsers support WebGPU?** Chrome 113+, Edge 113+, and other Chromium browsers. Safari and Firefox fall back to WebAssembly with identical output. URL: https://feyard.com/products/depth-3d/ --- ## Cleaner Web page → clean Markdown. Readability + Turndown, on-device. Turn any web page into clean, readable Markdown. Readability + Turndown, 100% on-device. ### Features - **100% private**: Page content is processed entirely in your browser. Nothing is uploaded. - **Readability first**: Mozilla's Readability extracts the main content, stripping ads and clutter. - **Turndown Markdown**: Clean, semantic Markdown output via Turndown. Ready for Obsidian, Notion, or anywhere. - **Copy or download**: One click to copy to clipboard or download as a .md file. ### How it works 1. **Paste a URL**: Enter the page URL you want to clean. 2. **AI extracts content**: Readability identifies the main article, stripping navigation, ads, and sidebars. 3. **Copy your Markdown**: Get clean Markdown output. Copy or download as a file. URL: https://feyard.com/products/cleaner/ --- ## Speech-to-Text Whisper speech-to-text. Upload audio or record, on-device. Whisper speech-to-text, fully on-device. Upload audio files or record live — no servers, no limits. ### Features - **100% private**: Your audio never leaves your device. All transcription happens locally. - **WebGPU accelerated**: Inference runs on your GPU when available, with WASM fallback. - **Whisper model**: Multilingual speech recognition with automatic language detection. - **Upload or record**: Drag & drop audio files (MP3, WAV, M4A) or record live from your microphone. ### How it works 1. **Upload or record**: Drag & drop an audio file, or record live. MP3, WAV, M4A, and more. 2. **AI transcribes**: Whisper runs locally, transcribing your audio to text with timestamps. 3. **Copy or download**: Copy the transcript or download as a text file. The model caches for reuse. URL: https://feyard.com/products/stt/ --- ## RAG Search Client-side PDF RAG. Extract, embed, and retrieve on-device. Client-side PDF RAG. Extract, embed, and retrieve — all on-device, no servers. ### Features - **100% private**: Your documents never leave your device. All processing happens locally. - **On-device embeddings**: Text embeddings generated locally via Transformers.js. No API calls. - **Chunk & retrieve**: Automatic text chunking and vector similarity search for precise retrieval. - **Cited answers**: Answers include source citations so you can verify the context. URL: https://feyard.com/products/rag/ --- ## Vectorize Bitmap to SVG/DXF. Convert logos, icons & illustrations to crisp layered vectors. Convert logos, icons, and illustrations to crisp, scalable vectors. 100% on-device via a custom pure-TypeScript tracing engine. ### Features - **100% private**: Your images never leave your device. All processing happens locally. - **Custom engine**: Median-cut color quantization, single-pass crack-following contour tracing, Douglas-Peucker simplification and Catmull-Rom Bézier curve fitting — pure TypeScript, no native dependencies, no server-side processing. - **Layered full-color tracing**: 2-32 colors with presets for logos, clip art, photos and line art; despeckle and blur pre-processing. - **Editable palette**: Recolor or hide individual color layers before exporting; automatic background detection and removal. - **SVG & DXF export**: Layered SVG (evenodd fills) free for everyone; per-color-layer DXF (R12/AC1009) for CAD workflows on Pro. ### How it works 1. **Upload an image**: Drag & drop a bitmap image (PNG, JPG, etc.) to vectorize. 2. **Tune the trace**: Pick a preset or adjust colors, detail, smoothness, despeckle and blur; the engine re-traces live. 3. **Download SVG/DXF**: Get a crisp, scalable vector file. No quality loss at any size. URL: https://feyard.com/products/vectorize/ --- ## Voice Synth On-device text-to-speech. 28 neural voices via Kokoro-82M. On-device neural text-to-speech powered by Kokoro-82M. 28 voices, zero servers. ### Features - **100% private**: Your text never leaves your browser. All synthesis happens locally. - **WebGPU accelerated**: Inference runs on your GPU when available, with WASM fallback. - **Kokoro-82M**: A lightweight, high-quality TTS model with 28 neural voices. - **28 voices**: Choose from a variety of male and female voices across multiple accents. ### How it works 1. **Type or paste text**: Enter the text you want to convert to speech. 2. **Pick a voice**: Choose from 28 neural voices. Adjust speed and pitch if needed. 3. **Play or download**: Listen in your browser or download the audio as a WAV file. URL: https://feyard.com/products/tts/ --- ## PDF Summary Summarize & extract from PDFs with on-device LLMs. Summaries, key points, and entity extraction from your documents — powered by on-device LLMs, no servers. ### Features - **100% private**: Your documents never leave your device. Inference runs locally via WebGPU. - **Two open models**: Choose SmolLM2 1.7B or Qwen2.5 1.5B — small, fast, and remarkably capable. - **Long-document aware**: Long PDFs are summarized part by part, then distilled into a final brief. - **Structured output**: Markdown-ready summaries, key points, and entity lists you can copy anywhere. URL: https://feyard.com/products/pdf-summary/ --- ## PDF Tools A free offline PDF toolbox — ten everyday operations in one tab. A single page that hosts the ten most-requested PDF operations: merge, split, rotate, compress, reorder, delete, encrypt, watermark, render to images, and assemble images into a PDF. Every step uses pdf-lib, pdf.js and Tesseract.js — the file never leaves the browser. ### Features - **100% offline**: Parsing, modification, recompression, encryption, OCR, image rendering — all in your browser. No upload endpoint, no telemetry, no account. - **Ten operations, one toolbox**: Merge · Split · Rotate · Compress · Page Manager (delete + reorder) · PDF→Images · Images→PDF · Encrypt · OCR · Watermark. - **Generous free tier**: 50 MB per file, 10 PDFs per merge, 30 pages per image export, 5 pages per OCR batch. Unlimited runs, no daily cap, no watermark on output, free for commercial use. - **Pro tier**: 200 MB per file, 50 PDFs per merge, 600 pages per export, 600 pages per OCR batch, higher-fidelity 3× rendering. - **OCR built-in**: A subset of the tool uses Tesseract.js WebAssembly to recognize text in scanned PDFs, in 12 languages, with confidence and timing per page. URL: https://feyard.com/products/pdf-tools/ --- ## Sudoku Classic Sudoku with AI logic hints, step-by-step solver, and printable packs. A minimal, distraction-free Sudoku with AI logic hints, a step-by-step solver, and printable puzzle packs. Generated entirely in your browser. ### Features - **Minimal, elegant UI**: Thoughtful typography, soft contrast and dark mode support — so the grid stays the focus for hours. - **AI logic hints**: Hints explain the rule — naked single, hidden single, and more — so you get better. - **Printable packs**: Six-puzzle packs across four difficulties with answer keys, ready to print or save as PDF. - **Step-by-step solver**: Stuck on a paper puzzle? Enter it and walk through every placement with reasoning. ### How it works Three modes, one engine. Pure TypeScript — runs in your browser, always offline-capable. 1. **Play**: Pick a difficulty and start solving. Keyboard, touch, or the number pad. 2. **Learn**: Ask for a hint whenever you're stuck. It reveals exactly why one digit is forced in one cell. 3. **Print**: Take puzzles offline. Six puzzles per pack, one answer page — ideal for travel or the classroom. ### Difficulty levels - **Easy**: 40 givens — approachable for beginners. - **Medium**: 33 givens — a pleasant challenge. - **Hard**: 28 givens — requires advanced techniques. - **Expert**: 24 givens — for seasoned solvers. ### Sub-pages - [Sudoku Solver](https://feyard.com/products/sudoku/solver/): Enter any grid and get the solution with step-by-step explanations. Each placement is annotated with its solving technique (Naked Single, Hidden Single). Advanced techniques fall back to search and are flagged. - [Printable Sudoku Packs](https://feyard.com/products/sudoku/printable/): 16 curated packs (4 difficulties × 4 packs) with 6 puzzles each and a separate answer key page. Also includes a random pack generator that produces fresh puzzles on-demand using cryptographically random seeds. ### Technical details The Sudoku engine is written in pure TypeScript and runs entirely client-side. Puzzle generation uses a seeded mulberry32 PRNG with FNV-1a hashing for deterministic, reproducible packs. Uniqueness is verified via MRV backtracking with a candidate bitmask. The solver annotates human-readable techniques (Naked Single, Hidden Single) before falling back to search for hard/expert puzzles. URL: https://feyard.com/products/sudoku/ --- ## Compress Free online image compressor. Batch JPEG, PNG, WebP & AVIF — MozJPEG WASM, no uploads. Free online image compressor. Batch compress JPEG, PNG, WebP & AVIF with MozJPEG WASM codecs — 100% in your browser, no uploads, no sign-up. ### Features - **100% private**: Your images never leave your device. No uploads, no servers, no tracking. - **WASM codecs**: MozJPEG, OxiPNG, WebP & AVIF — professional C encoders compiled to WebAssembly. - **Batch & convert**: Compress many at once. Convert between JPEG, PNG, WebP, and AVIF in a click. - **EXIF stripped**: Metadata is removed automatically. Optional pre-scale caps large dimensions. ### How it works 1. **Add your images**: Drag & drop, paste from clipboard, or browse. Batch as many files as you like. 2. **Pick format & quality**: Keep the original format or convert. Tune quality and an optional max dimension. 3. **Download**: Grab each file or download all at once. TinyPNG-class savings, fully on-device. ### FAQ - **Is this a TinyPNG alternative?** Close. We use MozJPEG + OxiPNG + WebP/AVIF WASM codecs from Google's Squoosh project. JPEG results are nearly indistinguishable from TinyPNG; PNG is optimized losslessly. - **Are my images uploaded to a server?** No. All compression runs in your browser via WebAssembly. Nothing is ever sent anywhere — there is no backend. - **Which formats can I convert between?** Inputs: JPG, PNG, WebP, AVIF, and GIF (first frame). Outputs: keep-original, JPEG, PNG (lossless), WebP, or AVIF. AVIF generally gives the best ratios but is slower to encode. URL: https://feyard.com/products/compress/ --- ## Subtitle Studio Video/audio → timed SRT, VTT or ASS subtitles with word-level editor and bilingual output. Turn any video or audio into timed SRT, VTT or ASS subtitles — with a word-level editor and a second language, entirely in your browser. ### Features - **100% private**: Your footage never leaves the device. Decoding, transcription and translation all happen locally. - **WebGPU accelerated**: Transcription runs on your GPU when available, with a WebAssembly fallback. - **Word-level timing**: Timestamped Whisper models give per-word timings for tight cues and karaoke export. - **Bilingual output**: Translate into a second language and export both tracks stacked in either order. ### How it works 1. **Drop a video or audio file**: Drag in MP4, WebM, MP3, WAV or M4A. It is decoded locally — nothing is uploaded. 2. **Whisper writes the cues**: Speech is transcribed in overlapping windows, then normalised to broadcast timing rules. 3. **Edit, translate, export**: Fix cues in the studio, add a translation, then download SRT, VTT, ASS, TXT or JSON. ### FAQ - **Are my videos uploaded anywhere?** No. The file is decoded, transcribed and translated inside your browser. Only the subtitle file you download leaves the tab. - **Which subtitle formats can I export?** SRT, WebVTT, ASS (with optional karaoke tags), plain text and structured JSON — as the original language, the translation, or both stacked. - **How does the bilingual translation work?** Compact Marian (OPUS-MT) models translate the cues, pivoting through English when there is no direct model. Models are cached after the first run. URL: https://feyard.com/products/srt/ --- ## Vocal Studio Vocal remover and audio denoiser. RNNoise WASM, 100% on-device. Vocal remover and audio denoiser powered by RNNoise (WASM). Separate vocals from background music, or clean up noisy recordings — 100% on-device, zero uploads. ### Features - **100% private**: Your audio never leaves your device. All processing happens locally via WebAssembly. - **Vocal separation**: Split any stereo track into vocals and accompaniment using center channel extraction. Perfect for karaoke or sampling. - **Noise reduction**: RNNoise removes background hiss, fan noise, and environmental sounds while preserving clear speech. - **WAV export**: Download processed audio as high-quality WAV files. Use vocals, instrumentals, or denoised tracks anywhere. ### How it works 1. **Upload audio**: Drag & drop an audio file or browse. MP3, WAV, OGG, M4A, FLAC — any format your browser supports. 2. **Process locally**: Choose vocal separation or noise reduction. RNNoise runs in your browser via WebAssembly — no uploads. 3. **Download results**: Preview the processed audio, then download as WAV. Vocals, instrumentals, or clean voice — take what you need. ### FAQ - **Is my audio uploaded to a server?** No. All processing happens in your browser via WebAssembly. Your audio files never leave your device. - **How does vocal separation work?** Vocals are typically panned to the center of a stereo mix. Using phase cancellation (center channel extraction), we isolate the center signal as vocals and the side signal as accompaniment. This works best on stereo recordings. - **What audio formats are supported?** MP3, WAV, OGG, M4A, FLAC, and any format your browser can decode via the Web Audio API. URL: https://feyard.com/products/vocal-studio/ --- ## OCR Image to text. Tesseract OCR via WASM, 100+ languages, zero uploads. Tesseract OCR, fully on-device via WebAssembly. Upload an image or paste a screenshot — 100+ languages, zero uploads. ### Features - **100% private**: Your images never leave your device. All recognition happens locally via WebAssembly. - **100+ languages**: English, Chinese, Japanese, Korean, and major European scripts — switch on the fly. - **Word-level boxes**: Every recognized word comes with a bounding box and confidence score, overlaid on the image. - **Export anywhere**: Copy plain text, or download as TXT, JSON (with positions), or searchable hOCR. ### How it works 1. **Upload or paste**: Drag & drop an image, paste a screenshot, or browse. PNG, JPG, WebP, BMP, and GIF. 2. **OCR runs locally**: Tesseract recognizes text in your browser via WebAssembly, with word positions and confidence. 3. **Copy or export**: Copy the text or download as TXT, JSON, or hOCR. The engine caches for reuse. ### FAQ - **Is my image uploaded to a server?** No. Everything runs in your browser. Tesseract OCR runs locally via WebAssembly — your images never leave your device. - **Which languages are supported?** 100+ languages via Tesseract, including English, Chinese (simplified & traditional), Japanese, Korean, and major European scripts. Switch language from the dropdown before recognizing. - **What image formats work?** PNG, JPG/JPEG, WebP, BMP, and GIF all work. Drag & drop a file, paste a screenshot, or browse to upload. URL: https://feyard.com/products/ocr/ --- ## AI Humanizer Make AI text read human. Rewritten on your device. Free AI writing and rewriting tool. Paste any draft — it scores how machine-like it reads, rewrites it entirely in your browser, and shows you every word it changed. No uploads, no sign-up. ### Features - **100% on-device & private**: Three engines — Chrome Built-in AI, a local on-device model, or an offline rules engine — all run in your browser. Nothing is uploaded, ever. - **See the score**: Six signals — sentence rhythm, predictability, vocabulary, active voice, opener variety and AI phrasing — combine into one score, shown before and after. - **Word-level diff**: Every edit is highlighted against your original. Nothing gets reworded behind your back, and you decide what to keep. - **Four modes, three tones**: Humanize, polish, expand or shorten. Keep it academic, professional or casual — or leave the register exactly as it was. ### How it works 1. **Paste your text**: Drop in a draft, an essay, or a paragraph of model output. It is analysed locally, instantly, with no download required. 2. **Read the score**: The panel shows which signals look machine-made and flags the exact phrases that give it away. 3. **Rewrite and compare**: Pick a mode and intensity, then watch the score drop and the diff fill in. Copy it, download it, or send it back to the editor. ### FAQ - **Does this beat AI detectors?** No tool can promise that, and we won't pretend otherwise. This is a writing tool: it lowers the surface signals detectors look for — uniform sentence length, predictable phrasing, tell-tale vocabulary — so your draft reads more like you. Use it on writing that is already yours. - **Where does my text go?** Nowhere. All three engines run in your browser. The on-device model is downloaded once and cached; after that the tool keeps working offline. - **Do I have to download anything?** No. The offline rules engine works the moment the page loads, with zero downloads. Chrome's built-in AI needs nothing either, if your browser has it. The local model is optional — choose it for higher-quality rewrites. - **Which languages work?** Rewriting follows the language of your input, and the interface ships in ten. The AI-score panel is tuned for English and Chinese; other languages still get full text statistics, just without the phrasing analysis. - **Will it change what I meant?** It shouldn't. Every engine is instructed to preserve facts, numbers and names, and the diff view lets you check each change before you accept it. Treat the output as a strong second draft, not a finished one. URL: https://feyard.com/products/humanizer/ --- ## Diagram Studio Text to diagram. Drawn by AI, on your device. Free AI flowchart and mindmap generator. Describe what you want in plain language — a small on-device model writes the Mermaid code and renders a clean diagram live. No uploads, no sign-up, export to SVG or PNG. ### Features - **100% on-device & private**: Diagram generation runs entirely in your browser through Chrome's built-in AI or a local Qwen 1.5B model. Your prompts are never sent to a server, ever. - **Eight diagram types**: Flowcharts, mindmaps, sequence, class, state, entity-relationship, Gantt and git graphs — pick a type or let the model infer it from your prompt. - **Live streaming render**: Watch the diagram draw itself token by token. The Mermaid is parsed and rendered as it streams, so you see the shape emerge in real time, then refine the code inline. - **Edit, export, own it**: The generated Mermaid code is fully editable — tweak any node or edge, then export as crisp SVG vector or high-resolution PNG up to 4×. Nothing is locked behind a paywall. ### How it works 1. **Describe your diagram**: Write a sentence or paste a spec — "user signup and onboarding flow", "e-commerce database schema", "agile sprint workflow". Pick a diagram type, or let the model decide. 2. **Watch it draw**: The on-device model writes valid Mermaid line by line. The canvas renders it live, catching and repairing common syntax slips as it streams. 3. **Edit and export**: Tweak the Mermaid code directly in the editor — add nodes, relabel edges, swap the theme. Export as SVG for vector use, or PNG at up to 4× resolution. ### FAQ - **Is my prompt sent to a server?** No. Generation runs entirely in your browser, either through Chrome's built-in AI (zero download on supported browsers) or a local Qwen 1.5B language model. Your prompts never leave your device. - **Do I need to download anything?** Only optionally. Chrome 127+ with the built-in LanguageModel API needs no download at all. The local Qwen model is a one-time ~1 GB download that is then cached for offline use. You can also edit Mermaid code directly without any AI. URL: https://feyard.com/products/diagram/ --- ## Video Compressor Compress video. No upload. No watermark. No limit. Free online video compressor. Shrink MP4, MOV, MKV and WebM with native WebCodecs — H.264, VP9 and AV1 re-encoding runs entirely in your browser, starts instantly, and never uploads a byte. ### Features - **100% private**: Your video is demuxed, decoded and re-encoded inside this tab. It never travels to a server — there is no server in the pipeline. - **Instant, no size cap**: Skip the upload wait and the 250 MB limits. Multi-gigabyte files start compressing immediately, and nothing is ever queued remotely. - **Quality you control**: Choose a quality level or an exact target size, cap resolution to 1080p or 720p, frame rate to 30 or 24 fps, and keep or drop audio. - **H.264, VP9 & AV1**: Export H.264 MP4 for universal playback, VP9 WebM for the web, or AV1 for files roughly 45% smaller at the same visual quality. ### How it works 1. **Add your video**: Drag & drop, paste, or browse. MP4, MOV, MKV, WebM, AVI, FLV and MPEG-TS are inspected locally in seconds. 2. **Choose your settings**: Pick an output codec, a quality level or an exact target size, plus resolution and frame-rate caps. The estimated size updates live. 3. **Compress & download**: Hardware-accelerated encode with live progress and speed readout, then compare before/after and save the watermark-free file. ### FAQ - **Is the compressed video watermarked?** Never. There is no account, no paywall and no watermark at any quality setting — the output is entirely yours. - **Are my videos uploaded anywhere?** No. A small WebAssembly demuxer and the browser's native WebCodecs encoders do all the work on your device. Nothing is transmitted, and the tool keeps working offline after the first engine load. URL: https://feyard.com/products/video-compress/ --- ## Eraser Erase watermarks and objects without uploading a thing. Paint over any watermark, logo, date stamp or unwanted object — or click it once with Pro Magic Select. Feyard rebuilds what was underneath with LaMa inpainting and a fast-marching repair engine, and Auto mode routes every mark to the right engine — entirely inside your browser. ### Features - **LaMa large-mask inpainting**: A generative model with 51M parameters reconstructs the structure hidden behind the mark instead of smearing the surroundings into a blur. - **Magic Select (Pro)**: Click an object and SlimSAM (Apache-2.0, ~14 MB quantized / 21 MB fp16, one-time download) segments it on-device with positive and Alt/right-click negative points; apply the highest-IoU candidate into your mask, or Esc out. - **Auto engine (default)**: Every run measures the mark — bounding box up to 96 px long edge and under 1% coverage routes to zero-download fast-marching repair; anything larger goes to LaMa. Fast and AI HD stay selectable manually. - **Same-position batch (Pro)**: Paint one shared mask on a reference photo and erase it from up to 10 images in a run; the mask is resampled per image at the same relative position, per-item failures stay isolated, and cleaned PNGs download as a ZIP. - **Nothing leaves your device**: Inference runs on WebGPU or WebAssembly inside this tab. No upload, no queue, no account, no image history. ### How it works 1. **Import a photo**: Drag & drop, paste or browse. JPEG, PNG and WebP are decoded locally; the free plan pre-shrinks files above 2048 px on the long edge, Pro accepts up to 4096 px. 2. **Paint what to remove**: Use the brush (adjustable size and hardness), rectangle or lasso to cover the mark — or click it once with Magic Select on Pro. Undo and redo cover every stroke, and a five-step result history can roll back whole eraser runs. 3. **Erase and export**: Only the masked region is sent to the model as a padded region of interest; the result is feather-blended back into the full-resolution original. Export PNG, JPEG or WebP, or batch-download a ZIP on Pro. ### FAQ - **Do the models need to download?** Yes — once each. LaMa is around 210 MB and Pro Magic Select downloads SlimSAM once, about 14 MB in WASM (21 MB for the WebGPU build); both are cached in your browser's IndexedDB. Fast mode never downloads anything. - **What does Auto mode decide?** It measures each painted mark: marks at most about 96 px on the long edge covering under 1% of the image go to instant, download-free fast-marching; everything else goes to LaMa. You can still pin Fast or AI HD manually. - **How large an image can I erase?** Up to 2048 px long edge for free (several competitors cap free uploads at 720p), 4096 px with Pro; larger inputs are pre-shrunk. Only the padded mask crop reaches the model, so memory stays predictable. - **Can I remove the same watermark from many photos?** Pro Batch paints the shared mark once on a reference photo and queues up to 10 images; the same relative-position mask is resampled onto each photo and the cleaned PNGs come back as a ZIP. Everything is still processed locally. - **Are my photos uploaded anywhere?** No. Decoding, masking, inference and export all happen in the page. The only network requests are the one-time model downloads, and once they are cached the tool works offline. URL: https://feyard.com/products/eraser/ --- ## CAD Trace Turn a part photo, hand sketch, or scanned drawing into an editable CAD file. Drop in a raster image and get native LINE, ARC, CIRCLE and POLYLINE entities on separate layers — geometry you can open and edit in AutoCAD, Fusion 360, LibreCAD or Inkscape. A custom contour and curve-fitting engine runs entirely in the browser; there is no AI model and nothing is uploaded. ### Features - **Real CAD entities, not a traced picture**: The linework is binarized and contoured locally, then fitted to straight lines, arcs and full circles with corner detection and endpoint snapping. Remnants that are neither stay as lightweight polylines on the LINES layer. - **Four-point perspective correction**: For photos of sheets and whiteboards, drag four corner handles onto the drawing corners; the keystoned quad is rectified before tracing so sides come out horizontal and vertical. - **Real-world scale calibration**: Click two endpoints of a known feature, type its real length (mm, cm, m or inches), and every measurement and exported coordinate uses that scale. - **DXF R12 + SVG export**: DXF (AC1009) with LINES / ARCS / CIRCLES layers and ACI colors for CAD software, plus SVG for Illustrator, Figma, Inkscape and the web. - **Free preview, metered export**: Importing, tracing, tuning, perspective correction and calibration are free and unlimited. Each DXF or SVG download uses one of 3 monthly free export credits; Pro is unlimited. ### How it works 1. **Add the drawing**: Drag & drop, paste or browse (JPG, PNG, WebP), or start with a built-in sample. Pick the input mode — sketch, scan or photo — which sets tuned threshold and fitting defaults. 2. **Trace and tune**: A Web Worker runs grayscale conversion, Otsu/adaptive thresholding, morphology, Moore-boundary contour tracing and line/arc/circle fitting. Adjust simplification tolerance, minimum segment length, corner angle and arc recognition with live re-runs; cancel any run instantly. 3. **Calibrate and export**: Optionally fix perspective and set one known dimension. Toggle layers, compare against the source with a split slider, then download DXF or SVG. The file is generated in this tab; the image never reaches a server. ### FAQ - **What file do I get?** A DXF in the R12 (AC1009) format — real LINE, ARC, CIRCLE and POLYLINE records on three layers — plus an SVG. Both open in virtually every CAD and vector package. - **What can I feed it?** Hand-drawn sketches, black-and-white scans, and high-contrast part photos. Clear linework with even lighting works best; shaded renders and faint pencil produce more freeform polylines instead of clean primitives. - **How much does it cost?** Previewing is free and unlimited. Free accounts get 3 DXF/SVG exports per month; Pro adds unlimited exports. There is no watermark. - **Is my drawing uploaded?** No. Thresholding, contour tracing, arc fitting and file serialization all execute locally. The only network request during export is a small JSON counter for the monthly credit. - **How accurate is it?** Entities are fitted within a configurable pixel tolerance and near-miss endpoints snap together, making dimensions self-consistent after calibration. It is a conversion tool, not a parametric modeler — verify critical tolerances against the original. URL: https://feyard.com/products/cad-trace/ --- ## 3D Model Optimizer Compress GLB and glTF files entirely on your device — geometry, textures, cleanup and validation, with a split before/after 3D preview. Drop in a model and the same glTF Transform pipeline used by professional asset pipelines runs in a local Web Worker: unused and duplicate assets are removed, vertices welded, geometry simplified and compressed with Meshopt or Draco, textures resized and color maps re-encoded to WebP. Every output is read back and validated before download. ### Features - **Meshopt or Draco geometry compression**: Meshopt decodes extremely fast for three.js and Babylon.js scenes; Draco reaches higher ratios for mobile and AR. Attribute quantization (KHR_mesh_quantization) adds further savings either way. - **Triangle reduction with rig protection**: A target-percentage slider runs error-aware edge simplification. Meshes bound to a skin or driven by morph targets are detected automatically and left untouched so bones and blend shapes never break. - **Lossless cleanup and welding**: Prunes unreferenced nodes/materials/textures/accessors, merges duplicates, and welds coincident vertices that cause cracks and defeat simplification. - **Texture optimization without surprises**: Proportional downscaling at a chosen max edge (4096/2048/1024/512). Base color and emissive maps go WebP only when the bytes are smaller; normal, occlusion and metallic-roughness maps stay PNG; KTX2 textures pass through byte-for-byte. - **Asset report and 3D comparison**: Size, triangles, vertices and texture counts before/after, graded compatibility issues, and a draggable split viewer with wireframe, grid and animation controls. ### How it works 1. **Add the model**: Drag & drop or browse GLB/glTF, or OBJ (with MTL), STL and PLY. Separated glTF bundles — .gltf, .bin, textures — are selected together and treated as one asset. 2. **Pick a preset or go custom**: Web (Meshopt, 2048 WebP), AR (Draco, 2048 WebP), Compatibility (no geometry compression), Aggressive (Meshopt, 35% triangles, 1024 WebP). Every control stays editable afterwards. 3. **Optimize and download**: The queue processes sequentially; cancel any batch instantly. Output is a self-contained GLB or a glTF ZIP; multiple models download individually or as one ZIP. ### FAQ - **What files can I optimize?** GLB and glTF natively, with OBJ, STL and PLY converted automatically. FBX, USD, STEP and Collada are not accepted — convert them to glTF first. - **What are the batch limits?** Free handles up to 10 models totaling 100 MB per batch; Pro raises this to 100 models and 500 MB. A single file is always accepted and sign-up is never required. - **Will reduction break my rigged characters?** No — skinned meshes and morph-target meshes are skipped automatically. Use your DCC tool for authored LODs on those meshes. - **Is my model uploaded?** No. Decoding, optimization, encoding and the 3D preview run in browser memory; there is no upload endpoint. Only the open-source Draco/KTX2 decoder runtimes are fetched from a CDN when the preview first needs them. - **Can I use results commercially?** Yes — no watermark, no attribution requirement, including files produced on the free plan. URL: https://feyard.com/products/model-optimizer/ --- ## Point Cloud Downsample, crop, filter and convert PLY/PCD/XYZ point clouds in your browser — 100% on-device, no upload. A free, on-device point cloud studio that opens PLY, PCD and XYZ scans, lets you crop, downsample, recolor and filter outliers, and writes a clean PLY, XYZ or GLB point cloud without ever uploading the source. Built for surveyors, LiDAR and SLAM researchers, photogrammetry hobbyists, drone mappers, and anyone who needs to make a million-point scan a tenth of its size before sharing it. ### Features - **Five downsamplers**: Voxel-grid keeps uniform density (one centroid point per cube), random 1-in-N for previews, every-Nth as a deterministic stride for time-aligned series, color-histogram to merge nearby colors and keep vegetation, road and façade regions intact, and farthest-point sampling that preserves global shape for sparse SLAM and small-object scans. All five are bit-exact reproducible. - **Crop, align, normalize**: Six sliders set min/max for X, Y and Z with a wireframe preview of the retained box. Align-to-Z-up computes the principal direction and rotates Y-up scanner output to Z-up; centering and normalizing move the cloud to the origin at a fixed scale. - **Statistical outlier removal**: One-pass k-NN filter (defaults k=8, sigma=2.0, tuned for terrestrial LiDAR; raise to k=20–30 and sigma=3.0+ for sparse SLAM clouds) drops isolated returns in a typed-array pipeline — a 10-million-point scan filters in under a second on a modern laptop. - **Color by source, height, intensity or density**: Local density (k-NN distance, normalized) surfaces under-sampled regions of a SLAM cloud as bright halos. - **Format bridge**: Auto-detects binary little-/big-endian and ASCII PLY (vertex colors, normals, intensity, alpha), ASCII and binary PCD, and ASCII XYZ from magic bytes. GLB export uses the Khronos KHR_points extension — viewable in three.js, Babylon.js, Blender and any web 3D viewer. - **Limits**: Free handles a single file up to 200 MB or 50 million points with the full pipeline; Pro raises that to 1 GB or 500 million points with 20 files per batch. Memory peaks at roughly 2× the source size, and outputs are byte-exact reproducible. ### How it works 1. **Drop your scan**: Drag & drop, paste or browse a .ply, .pcd or .xyz file — the parser streams it once into a flat Float32Array. 2. **Process**: Pick a downsampler, dial in the crop box, align the axes and filter outliers with live 3D preview. 3. **Export**: Save PLY (binary LE or ASCII), XYZ or GLB with KHR_points — no watermark, commercial use included on both plans. ### FAQ - **Is my scan uploaded?** No. Parsing, processing and export all run in your tab; there is no server in the pipeline and the tool keeps working offline after the first load. - **How large a file can I open?** Free: 200 MB or 50 million points per file. Pro: 1 GB or 500 million points, 20 files per batch. A single file is always accepted. - **Will the same settings give the same output?** Yes — every operation is deterministic; the same input and settings always yield byte-identical output, so downstream pipelines see a stable artifact. URL: https://feyard.com/products/point-cloud/ --- ## Font Forge Inspect, preview and extract font glyphs — without uploading the font. Font Forge is a free, browser-based font inspector and glyph extractor. Open a TTF, OTF or WOFF file and it parses the font entirely on your device — no upload, no server, no account. Built for designers, front-end developers, and anyone who needs to preview a font, pull a single glyph as SVG, or ship a smaller web font. ### Features - **Full glyph grid**: The table directory is walked locally and the name table is read for family and copyright metadata; every glyph lands in a searchable grid filterable by character, glyph name, glyph index or Unicode point (e.g. U+4F60). - **Glyph detail and SVG export**: Click any glyph for its vector outline, metrics (advance width, bounding box) and Unicode point, then download a clean SVG or copy the path data to your clipboard. - **Live preview**: Type any text and the actual outlines in the file are rendered as vector paths — what you see is exactly what the font contains. - **Subsetting**: Paste the text you actually use and subset keeps only those glyphs, rebuilding a valid font you can drop into @font-face — for Chinese and other large scripts this can cut a multi-megabyte font down to a few kilobytes, the single biggest win for web font performance. - **OTF and WOFF2 output**: Download the subset as OTF or WOFF2, plus drop-in @font-face CSS with the font base64-embedded — or export every glyph as individual SVG files in one ZIP, handy for pulling icons out of an icon-font package into a component library. - **WOFF2-aware input**: WOFF2 upload is detected and rejected with a clear message; TTF, OTF and WOFF are supported. - **Limits**: Free handles files up to 20 MB; Pro raises the limit for whole-font libraries and variable fonts. ### How it works 1. **Drop a font**: TTF, OTF or WOFF is parsed in your tab via opentype.js; nothing is uploaded. 2. **Inspect and subset**: Browse or filter the glyph grid, preview live text rendering, and subset to just the characters you need. 3. **Export**: Download single-glyph SVGs, a glyph ZIP, a subset font (OTF/WOFF2) or the @font-face CSS — saved straight to disk. ### FAQ - **Are my fonts uploaded?** No. Parsing and export happen in your tab; check the Network panel — once the page is loaded there is no traffic while you work. - **Can I subset any font I find?** Font Forge only re-packs the glyphs you already have; it grants no license. Only subset fonts you are authorized to use, and respect the original license when redistributing the result. - **Does it work offline?** Yes — after the first load the tool runs entirely locally, which also makes it safe for licensed or confidential fonts. URL: https://feyard.com/products/font-forge/ --- --- ## RAW Studio Develop camera RAW files — without uploading them. RAW Studio is a free, browser-based camera RAW developer. Open a DNG, CR2, CR3, NEF, ARW, RAF, ORF, PEF or RW2 file and it is decoded entirely on your device — no upload, no server, no account. Built for photographers who want a Lightroom-class develop workflow without the install or the subscription. ### Features - **Nine RAW formats**: DNG (uncompressed, lossless JPEG, deflate), Canon CR2/CR3, Nikon NEF/NRW, Sony ARW, Fujifilm RAF, Olympus ORF, Pentax PEF and Panasonic RW2, parsed by a pure-TypeScript engine — TIFF and ISO-BMFF containers, lossless JPEG (SOF3) entropy decoding, 8/10/12/14/16-bit CFA extraction. - **Correct color**: DNG black/white levels, CFA patterns, embedded color matrices and as-shot neutral drive an exact camera-to-sRGB transform with Bradford adaptation to D65 — the same data Adobe's converter uses. Camera-original files fall back to a generic camera profile. - **Desktop-class develop controls**: exposure, contrast, highlights, shadows, whites, blacks, temperature, tint, vibrance, clarity and dehaze — all in linear light, all non-destructive, with a before/after split view and a live RGB histogram with clipping indicators. - **Curves, HSL, split toning**: a per-channel tone curve editor, eight-band HSL with hue/saturation/luminance per color, highlight/shadow split toning, film grain and vignette. Ten presets plus a one-click Auto (gray-world white balance with a gentle tone lift). - **Detail and lens**: two demosaic algorithms (fast bilinear, quality gradient-corrected), unsharp masking with radius and threshold, luminance and chroma noise reduction, chromatic-aberration and distortion correction. - **Geometry and batch**: 90° rotation, flips, straighten with auto-crop, aspect-ratio crops, and a filmstrip queue of 10 files (free) or 100 (Pro) — each file keeps its own non-destructive settings. - **Export**: full sensor resolution with the quality demosaic at export time; JPEG, PNG and WebP at any tier with quality and long-edge controls, plus 16-bit uncompressed TIFF on Pro. No watermark at any tier; every export is licensed for commercial use. - **Limits**: Free handles RAW files up to 80 MB and batches of 10; Pro raises the limits to 400 MB per file, 100 files per batch, higher-resolution previews and 16-bit TIFF export. ### How it works 1. **Drop your RAW files**: the engine auto-detects container, compression and bit depth, and reads the embedded color profile. 2. **Develop non-destructively**: every slider is a reversible instruction — the sensor data stays untouched in the worker. 3. **Batch with the filmstrip**: queue files, jump between them, and carry your look across the set. 4. **Export at full resolution**: render with the quality demosaic and encode in your browser straight to disk. ### FAQ - **Is RAW Studio really free?** Yes — the free tier is the complete RAW workflow: every format, every adjustment, every preset, full-resolution export with no watermark and a commercial license, unlimited edits. Free processes files up to 80 MB and batches of 10. Pro raises the limits to 400 MB per file and 100 files per batch and adds 16-bit TIFF export. - **Are my RAW files uploaded anywhere?** No. The container parser, lossless JPEG decoder, demosaic, develop pipeline and encoder all run in a Web Worker inside your browser. Zero network egress — verify it in the Network panel or disconnect and keep editing after the first load. - **Which RAW formats are supported?** DNG, CR2, CR3, NEF/NRW, ARW/SRF/SR2, RAF, ORF, PEF, RW2 and uncompressed TIFF CFA files. Foveon X3F, lossy sRAW/mRAW and vendor-proprietary Huffman compression are not supported — converting those to DNG first is the reliable path. - **Will my colors match Lightroom?** For DNG files, yes — the embedded color matrix, calibration illuminant and as-shot neutral build an exact camera-to-sRGB transform. For camera-original files without an embedded matrix a generic profile is applied, so hues stay natural but saturation may differ slightly. - **What are the limits of a browser RAW editor?** Heavy AI denoising, tethered capture and catalog management still belong on the desktop; decode, demosaic, tone, color, detail, geometry and export all run here at full quality. URL: https://feyard.com/products/raw-studio/ --- ## Comparisons ### remove.bg alternatives remove.bg retires its standalone site on 1 December 2026 (09:00 CET). The web tool moves into Canva — you must open the Canva editor and be signed in — the self-serve API moves to Leonardo.Ai, and the Photoshop, Figma, GIMP and Sketch plugins plus the Zapier / Make automations are discontinued with no successor. Prepaid credits do not carry over. Feyard is a drop-in replacement for the common case: three model tiers (Fast / Sharp / Sharp · GPU), a manual edge brush that edits the alpha channel, batch export of up to 500 images to one ZIP, no uploads, no account, no watermark, and full-resolution output licensed for commercial use. Compared at a glance: | Capability | Feyard | remove.bg | Canva | Photoroom | Adobe Express | | --- | --- | --- | --- | --- | --- | | Where the photo is processed | In your browser | remove.bg servers | Canva servers | Photoroom servers | Adobe servers | | Account required | No | No (web), yes (API) | Yes | Yes | Yes | | Free full-resolution export | Yes | No — 0.25 MP preview | No — Pro plan | No — watermarked | Limited | | Commercial use of free output | Yes | No — personal use only | Plan-dependent | Paid plans | Paid plans | | Manual edge brush | Yes, edits the mask | Magic Brush (AI) | Editor brush | Yes | Editor brush | | Batch | Up to 500 → ZIP | Paid plans only | No | Paid plans | Limited | | Price | $0 | ~$0.20–$0.60 per image | Subscription | Subscription | Subscription | URL: https://feyard.com/alternatives/remove-bg/ --- ## Privacy & Terms - [Privacy Policy](https://feyard.com/privacy/): How Feyard handles data — all processing is on-device, no servers, no tracking. - [Terms of Service](https://feyard.com/terms/): Terms of use for Feyard tools. --- ## Roadmap Feyard publishes a quarterly roadmap. Currently shipping: - **Q2 2026 (Done)**: Feyard launch, Remove BG, Upscaler, Restore, global theme system. - **Q3 2026 (Current)**: Speech-to-Text, Voice Synth, RAG Search, Vectorize shipped. Next: VocalFlow, QuickScribe, Polyglot. - **Q4 2026 (Upcoming)**: SightLine summaries, DocMuse writing, team workspaces. - **Q1 2027 (Planned)**: CodexLens, ChatThread, AgentOS private preview.