Windows desktop tool (ScreenKit.exe): screenshot, annotate, OCR, barcode/QR, long screenshot, screen/GIF recording, PDF workbench, ASR/TTS, LLM chat, translation, face, local HTTP API, LAN file transfer, and LAN/USB screencast with an Android companion.
Current version: 1.0.14 · GitHub Releases
Changelog: CHANGELOG.md (English and Chinese per version).
- Screenshots
- Features
- Dictionary
- Requirements
- Using ScreenKit
- Install features
- Configuration
- HTTP API
- CLI
- x86host
- Build from source
- License
| Area | Description |
|---|---|
| Screenshot OCR | Region capture → text OCR or barcode/QR; multi-monitor DXGI. Korean/English word spaces restored from visual gaps (no spaces between CJK). Optional overlay translation via LLM. The image bar is a compact WPF toolbar; items that do not fit move to the overflow menu. History lists screenshots/ as thumbnails, previews with pan and zoom, and annotates under that preview. With marks, Done, OCR, or Open saves a new screenshot and copies it. With no marks they use the original file and do not write a new one. Done, OCR, and Open then close the history window and show the picture on the main window; OCR also recognizes it. Copy and Ctrl+C put the current picture on the clipboard and leave the window open. |
| Annotate | WeChat-style tools: rect / ellipse / arrow / pen / text; dropdown next to confirm for copy-as image / file / path. |
| Long screenshot | Pick a scrollable window → auto-scroll stitch (no OCR). |
| Screen recording | Window or region → HUD → MP4 + optional system/mic audio; optional mouse cursor and click highlight. Codecs: x264 / x265 / AV1 (FFmpeg), Windows Media Foundation H.264 (MP4, no download). |
| GIF recording | Same region flow → 24 fps capture → preview (FPS, scale, palette) → silent GIF. |
| Clipboard | Paste image and OCR; Edit menu copy image / file / path; menu/tray sets on-capture copy mode. |
| Overlay text | Click-release selects one OCR block; empty click clears; drag-select stays in range. Ctrl+C copies. |
| PDF workbench | Open PDF → page OCR → edit lines → export searchable PDF. |
| Area | Description |
|---|---|
| ASR / TTS | Live captions: offline or streaming. Sherpa / SAPI / WinRT offline TTS and Edge online natural voices. With no saved engine, Windows speech is used when a voice is installed, otherwise Edge online. The Speech language list shows named languages first, then the remaining codes. SAPI and Windows speech list only languages that have a voice. Supertonic uses a fixed noise seed, so the same text sounds the same each time. |
| LLM chat | WeChat-style bubbles, Clear, mic, Speak / Auto speak, optional web + tmp/llm/ tools. |
| Translation | Local Opus-MT ONNX or a configured LLM; 20-trip round-trip stops early on a repeat; popup Ctrl+Alt+T. |
| Dictionary | Chinese, English, Japanese, and Korean headwords on a main-window tab. See Dictionary. dict.db is not in the release archive. Help → Install features can download it. |
| Face | InsightFace ONNX detect/compare; optional landmarks and gender/age. Models in facemodels/. |
| SAPI x86 helper | Sidecar x86host.exe for classic voices visible only to 32-bit processes. |
| Area | Description |
|---|---|
| PC file transfer | Same HTTP API port (1224) + UDP discovery 17531. File sync tab lists the sendfile/ inbox (no subfolders): list / thumbnails toggle, marquee and Shift/Ctrl multi-select, Ctrl+X/C cut/copy (cut items shown faded), paste, delete to Recycle Bin, drag to Explorer. Drop on the lower zone to send to the phone while connected; phone shares land in sendfile/. Install on phone checks and downloads the latest APK, then shows a LAN URL and QR. Web manager (same port): desktop /, phone /m (compact list, long-press multi-select, Upload/Camera/New); login to upload/manage; download only needs /f/…. Companion: android/ (com.whj.screenkit, launchers File transfer / Screencast). |
| HTTP API | JSON API (default port 1224). Allow LAN access opens every interface; otherwise it listens on 127.0.0.1 only. Main-window tab: call log + request builder. |
| Install features | Feature tree (installed items checked); add (green) / remove (red); Confirm installs/uninstalls. ONNX speech models and Windows speech are separate tabs. CN mirrors when locale is Chinese. |
| Image convert | Tools → Image convert: batch JPG/PNG/BMP, optional shorter-side cap (shrink only), rotate/flip; icon view with small thumbnails or a details list; live preview of the selected file after those settings; write to output/ next to each source, a chosen folder, or replace the source (permanently delete, or Recycle Bin). Optional: keep the original file when the result is still ≥ N% of the original size (default 80%; rotate/resize still writes the new file). |
| QR / barcode | Tools → QR / barcode: QR, Data Matrix, Code 128/39, EAN, UPC. One line of original text under the image. UTF-8, GBK, or Hex bytes (default UTF-8). |
| Batch rename | Tools → Batch rename: source list (add/remove/move, sort by name / modified / added); new names in a textarea. Everything / FastCopy patterns (%1/%2 shortest capture, # / ###); changing the list or rules recalculates names. |
| Hash | Tools → Hash: MD5 / SHA-1 / SHA-256; paste an expected hash to match. |
| Text tools | Tools → Text tools: Base64, URL, UTF-8/GBK hex, Unicode escape, case, whitespace, counts, smart JSON pretty-print (short arrays/objects stay on one line). |
| Password generator | Tools → Password generator: crypto-random; length, sets, skip 0OIl1, at least one of each class. Remembers last settings. Word lex tab: pick an LLM, then translations (including Literary Chinese, Ancient Greek, Latin, Sanskrit, Biblical Hebrew) plus pinyin / romaji / romanization. Variants tab: from a seed password, the LLM mostly translates into other languages and spells them in ASCII romanization (at most one English paraphrase; no extra symbols, digits, leetspeak, or word-reordering). |
| Network tools | Tools → Network tools: Ping, DNS, WHOIS, traceroute, ping-locate, proxy-locate, HTTP speed-locate. |
| Window manager | Tools → Window manager: list visible top-level windows, pick one by clicking, or enter an HWND. A green frame follows the window under the cursor while picking, and stays on the selected window. Pin in front or unpin (HWND_TOPMOST). |
| Screencast | Main window Screencast tab: cast this PC to another. Saved PCs stay in the left list; click one to fill the IP, or type an address / Scan LAN. Max size defaults to 1000×1000 (shrink to fit, never enlarge). Start cast first picks a region the same way as a screenshot (click a window or drag). Sending waits until Start is pressed again, so audio and its source (speakers, microphone, or both) can be changed. The red frame and toolbar can start, pause, or stop; pause keeps the connection and stops picture and audio until resume. The frame is not part of the picture. Tools → Screencast (same under the tray) still receives a phone or another PC, and can cast the whole desktop (quality 540p/720p/1080p). Audio plays continuously; the picture waits only as long as the sound is behind. The viewer is not always-on-top; picture can Fit or Fill the window. Rotating the phone keeps the viewer’s size, position, and maximized state; the picture fits or fills inside that window. Discovery shares UDP 17531 with file transfer; video uses WebSocket HTTP /cast (default 1224). Connect Android USB accessory (tray, file-sync tab, screencast window) opens “Waiting for Android USB accessory” and tells a phone already connected for file sync to enter accessory mode. Both sides show connected / waiting / not connected. It turns off after 2 minutes with no connection. Or USB cast (adb) with USB debugging. The phone app has a separate Screencast launcher icon. |
| Hotkeys | Toggle window · snap annotate · snap OCR · voice input · translate popup (configurable). |
| Modules | Settings → General turns OCR, TTS, ASR, chat, translate, face, and dictionary on or off (mod_*, default on). Off hides that page and stops its hotkey and tray entry. Settings → API turns each HTTP path on or off (http_ocr, http_tts, http_asr, http_translate, http_chat, http_face). Off returns code 810. |
| Main tabs | Settings → General can hide OCR / TTS / ASR / Chat / Translate / Face / HTTP / File sync / Screencast / Dictionary (tab_*_visible). Hiding a tab does not stop its hotkey. |
| Devices | CPU · NVIDIA CUDA · Intel DirectML; missing accel → CPU. |
| CLI | Batch OCR, list models / SAPI voices, probe CUDA, multi-monitor snap test. |
ScreenKit looks up headwords in a read-only SQLite file, dict.db, placed next to the program. The release archive does not include it. Help → Install features can download dict.7z and extract it there. Extracting downloads 7za.dll and SharpSevenZip.dll when they are missing. Opening the file needs e_sqlite3.dll. A lookup does not modify the file. The connection closes after 5 minutes without a search and opens again on the next one.
- Open it from the main window (Options → Dictionary). The tab is on by default (
tab_dict_visible). - The hotkey
hotkey_dictis off until you set it. It shows the main window and selects that tab, turning the tab on if it was hidden. Pressing it again while the tab is already showing hides the main window. It reads the clipboard and searches only when that text is one word: 1–4 Chinese characters, 1–20 English letters, Japanese with kana up to 12 characters, or 1–8 Hangul syllables. A sentence, a blank clipboard, or anything else only opens the dictionary. - Language filters are buttons in a row: All, Chinese, Japanese, Korean, English.
- The hit list is two lines (language and headword, then the gloss) with a speak button. The entry uses smaller type and colors pronunciation, part of speech, sense numbers, language labels, and examples. Each example has a speak button.
- Select text in the entry to open a popup outside the selection, with matching headwords plus Speak, Search, Translate, and Copy. Speak reads the whole selection. The popup closes when that window is no longer active.
- Search, a matching headword in the popup, or a word link in the entry opens another dictionary window and looks up that text. The page you were reading stays.
- Speak uses Settings → Dictionary. Each of Chinese, English, Japanese, and Korean has an engine (Auto / ONNX / SAPI / Windows speech / Edge), that engine's voice, and a rate from 0.5 to 2. Auto uses that language's Windows speech, or Edge online when none is installed. ONNX is chosen explicitly. Japanese and Korean headwords are spoken only up to the first comma. Audio is cached in
tmp/voiceand kept for 1 day. This does not change the Speech tab.
dict.db holds four dictionaries in one file. ScreenKit does not build the file.
- Chinese. Characters, words, and idioms from mapull/chinese-dictionary (a pinyin dictionary). License: MIT. The upstream project notes that the origin of some material is unclear.
- Japanese. Vocabulary senses from the JMdict/EDICT Dictionary Project (EDRDG). License: CC BY-SA 4.0. Redistribution has to keep the EDRDG attribution and follow that project's update terms.
- Korean. A derivative of the National Institute of Korean Language dictionary. License: CC BY-SA 2.0 KR.
- Windows 10/11 (x64)
- .NET Framework 4.8
- Build: Visual Studio / MSBuild with a
net48WPF targeting pack - Optional: NVIDIA GPU + CUDA matching
onnxgpu64; DirectML GPU foronnxdml64 - Optional (x264 / x265 / AV1 and GIF): FFmpeg 4.4 shared under
ffmpeg64/next to the exe. Media Foundation H.264 needs no download.
Capture a region (hotkey or menu). Result panel splits OCR / Barcode. Overlay translation uses a configured LLM when a dest language is selected. Annotate tools sit on the capture overlay; a region that spans monitors can be drawn on, moved, or resized on every screen it covers (move hot-zone is 10px outside the frame). The toolbar can open the shot with the default app. The Done dropdown copies as image, file, or path and remembers the default. Settings → Capture can cap the shorter side: if the smaller of width and height is over the limit, the saved file shrinks to fit and is never enlarged. That applies to screenshots/ and copy-as-file; OCR still uses the full image.
- Capture → Screen record (or GIF record): click a window or drag a region.
- HUD (drawn outside the capture area):
- Red frame; drag the 5px strip to move, or 8 grips to resize. Aspect is free before Start and locked afterwards unless
record_lock_aspect = false. - Floating control bar: left grip to drag; collapse; Options before Start; start/pause share one slot. Stays on the current monitor.
- Red frame; drag the 5px strip to move, or 8 grips to resize. Aspect is free before Start and locked afterwards unless
- Stop → save MP4 (Explorer selects the file) or open the GIF preview (output FPS 1–24, scale, palette) then save a silent GIF.
- Capture → Record options: codec (x264 / x265 / AV1 / Media Foundation H.264), FPS, CRF (x264/x265 only) or AV1 CRF (0–63, default 56; hidden for the other codecs), audio, shorter-side cap (shrink only, same rule as screenshots), record mouse / highlight clicks. AV1 needs libsvtav1 / libaom-av1 in
ffmpeg64and fails clearly with no fallback to x264. Media Foundation does not use FFmpeg. System H.264 writes AAC during capture at 44.1 kHz or 48 kHz, and does not mix a second file when saving. A savedrecord_codecofmjpegis read as system H.264.
GDI capture has no cursor: enable record mouse to overlay the pointer; highlight clicks draws a short yellow (left) / blue (right) / green (middle) ripple. GIF size grows quickly — use preview scale/FPS and the max-size limit. GIF still uses a max width and height box.
- On the PC, open the File sync tab (enable under Settings → API if needed; default on).
- Install on phone: check and download the latest APK from GitHub, then show LAN URL
http://<pc-ip>:1224/apkand a QR code (defaults to an internet-reachable NIC). Phone and PC on the same LAN; scan with a browser to install. The Windows release package does not include the APK. - First connection: the phone shows a waiting dialog (cancellable) until the PC allows pairing. The PC dialog stays always-on-top (also when the main window is in the tray).
- The upper list is the PC
sendfile/inbox (root only, no subfolder navigation; folders are rows). Double-click a file to open it with the system, or a folder to open Explorer. The right-click menu acts on the current selection (Open, Cut, Copy, Paste, Push, Delete) and does not clear a multi-selection when you click an already selected row. Use the toolbar to switch list / thumbnails; marquee, Shift range, and Ctrl multi-select; Ctrl+X/C/V cut/copy/paste (cut rows appear semi-transparent), Delete to Recycle Bin, or drag to Explorer. Dropping files onto the list imports them. Drop on the lower zone to send to the phone while the app is connected; otherwise nothing is sent. File/image paste goes into the inbox; text paste still goes to the right-hand pane. Phone share/upload always writes to PCsendfile/. - Both sides show in-progress transfers; the PC also has a transfer log. Text sync is one text box with Send, Copy, and Clear. The phone auto-connects by LAN IP only, not over USB.
- Web manager (same HTTP port): File sync → Web manager shows the LAN URL, QR, and login password. Desktop page
/, phone page/m. The page follows the browser language and has a Chinese / EN switch (saved in the browser). Sign in to upload, mkdir, rename, delete; optional Stay signed in. Phone UI: breadcrumb path, compact list (tap to open/download, long-press to multi-select), bottom Upload / Camera / New. Camera compresses in the browser before upload, using Settings → API (format, JPG quality, shorter-side cap). If the browser cannot compress, the PC applies the same settings. Reopen the page after updating the PC app, then take the photo. Zip/copy/rename/delete from the selection bar. A correct/f/<relative-path>URL downloads without login. Design reference:ScreenKit/SendFile/web/m-prototype.html.
Android details: android/README.md.
- PC Tools → Screencast (receive starts with the app; Settings → API can turn it off). Same Wi‑Fi, not a guest network. Launching this build ends leftover ScreenKit from another folder. Wi‑Fi/ADB media uses the same HTTP port as file transfer (
/cast); scan uses UDP 17531. - Phone File transfer / Screencast → Screencast. The page fills the saved PC IP; confirm it and tap Start cast, then allow screen capture. Use Saved PCs or Scan LAN to pick another PC. Stop cast appears only while casting. USB cast is a separate section below. The phone sends
helloand waits for the PC’shellobefore capture; the PC viewer opens only after that reply. Wi‑Fi media uses HTTP/cast(default 1224). The file-transfer listener on a LAN address completes that WebSocket handshake as well. If the PC app is closed, the handshake fails, the PC disconnects, or screen capture is denied, the phone shows a Cast failed dialog. After ~90s without an accessory it falls back to USB tethering. USB debugging cast usesadb reverseand needs USB debugging. Debug forward:adb shell am start -n com.whj.screenkit/.CastActivity --ez scst_adb_probe true. WiFi hello:--es scst_ip <PC LAN IP> --ez scst_wifi_probe true. USB-only test pattern:scst_usb_pat. - A view window opens in front, centered on the primary monitor. Tab overlay (type / resolution / fps) is off until you press Tab. Closing the window disconnects. Quality changes apply live. Landscape re-encodes at the same short-edge quality (does not crop a portrait frame). Stopping on the phone closes the PC window immediately (
GET /api/cast/stop). - PC-to-PC: the receiver keeps this app open (receive is on by default). On the sender, open the Screencast tab, click a saved PC to fill the IP (or scan the LAN, which includes this PC), set max width and height, tap Start cast, then click a window or drag a region. Tools → Screencast can still scan and cast the whole desktop using a quality preset.
Tools → Image convert: drop files, folders, or images; choose JPG / PNG / BMP; optional shorter-side cap (if the smaller side is over the limit, shrink to fit; never upscale); rotate 90/180/270° or flip the selected item. Icons shows a small thumbnail per file (List for details). The right pane previews the selected file after those settings. Convert-all writes next to each source under output/, into a chosen folder, or replaces the source (permanently delete, or Recycle Bin; convert-all asks first).
| Hotkey | Action |
|---|---|
Ctrl+Alt+O |
Toggle main window |
Ctrl+Alt+Q |
Screenshot annotate |
Ctrl+Alt+W |
Screenshot and OCR |
Ctrl+Alt+V |
Voice input (press again to stop) |
Ctrl+Alt+B |
Live caption |
Ctrl+Alt+T |
Translate popup |
Ctrl+Alt+P |
Cycle on-capture copy (image → file → path) |
Leave a hotkey string empty in Settings to disable it. The dictionary-tab hotkey is empty by default. Tray: left-click toggles the window; context menu has voice input, translate popup, clipboard OCR, on-capture copy mode, and exit. Closing the main window typically hides to tray. Switching the copy mode from the menu, tray, or Ctrl+Alt+P recopies the last screenshot and shows a toast at the bottom center of the screen. After a region is selected, the annotate bar shows the copy mode as a combo (for example “Copy as image p”). Choosing an item finishes the shot with that mode and remembers it. P on the bar cycles image / file / path and does not finish.
Options → Language → Chinese / English, or Settings → General (ui_lang = "zh" / "en"). Covers menus, Settings, OCR toolbar Pack/Lang names (ocr-display.json nameEn), Translate, Face, the translate popup, and the dictionary window.
- First launch may open the wizard (nothing selected). System H.264 stays checked and needs no download.
- Later: Help → Install features
- Features: installed items are checked; add (green) / remove (red) counts and sizes; Reset restores. Confirm installs and uninstalls. Speech packs are not on this tab. Dictionary downloads
dict.7zfrom a fixed URL (not the app update package). Extracting it downloads7za.dllandSharpSevenZip.dllwhen they are not already installed. Opening the database needse_sqlite3.dll. - ONNX speech models: TTS models with language filter; progress shows total batch size and downloaded bytes.
.tar.bz2packages are extracted in-process (no systemtar/bzip2); a junction-basedttsmodelsdirectory is supported. Japanese is Supertonic 3 (sherpa-onnx-supertonic-3-tts-int8-2026-05-11). Choose Japanese in the language list. That pack also speaks the other Supertonic 3 languages. - Windows speech: lists
Language.TextToSpeechpacks and the voices Windows reports for each language. Install or remove runs only when ScreenKit is an administrator; otherwise the DISM command is shown and can be copied. Restart ScreenKit before the speech page lists a newly installed voice.
- Features: installed items are checked; add (green) / remove (red) counts and sizes; Reset restores. Confirm installs and uninstalls. Speech packs are not on this tab. Dictionary downloads
- Using a feature that needs a missing package prompts to open the installer (e.g. OCR without any ORT → install
onnxcpu64).
| Runtime | Role | Typical size |
|---|---|---|
| onnxcpu64 | CPU ONNX Runtime (required for OCR if no GPU/iGPU ORT) | ~16 MB |
| onnxgpu64 | NVIDIA CUDA EP + CUDA/cuDNN (optional) | large |
| onnxdml64 | DirectML EP for iGPU (optional) | ~18 MB |
| OpenCV | Capture / image pipeline (OpenCvSharpExtern.dll) |
~61 MB |
| OpenCV video | opencv_videoio_ffmpeg4110_64.dll. Not used by OCR or recording. Help → Install features → OpenCV video |
~28 MB |
| ZXing | ZXing.dll. Barcode scan and QR images. Asked for when you use them |
~3 MB |
| SharpCompress | SharpCompress.dll. Extracts tar.bz2 speech and ASR packs |
~2.5 MB |
| SharpSevenZip / 7za | Extract the dictionary dict.7z |
~1.8 MB + ~0.4 MB |
| SQLite | e_sqlite3.dll. Opens dict.db |
~1.7 MB |
| ffmpeg64 | Screen record encode/mux | ~72 MB |
Download prefers CN mirrors when UI or system locale is Chinese.
Edge online TTS: 300+ Microsoft natural voices (including Korean), no model or API key. Requires Internet; listing and synthesis go to Microsoft Edge Read Aloud (unofficial; not paid Azure Speech). The configured HTTP proxy is honored.
Optional env vars for local full libraries (do not commit secrets/paths):
| Variable | Meaning |
|---|---|
WPF_OCR_CUDA_LIB |
Folder with full CUDA / onnxgpu64 DLLs |
WPF_OCR_FFMPEG_LIB |
Folder with FFmpeg 4.4 shared DLLs |
Settings live in config.toml beside the exe (Options → Settings / Record options). Tabs: General, OCR, Hotkeys, Dictionary, Speech, LLM, Translate, Capture, API. General has module switches and main-tab visibility. API can turn each HTTP path off. Options → Memory shows process memory and lists each loaded model and the ONNX runtime if one is loaded. Each row can unload that item from memory and forces a full GC. Unload on the ONNX row also frees the native CUDA, CPU, or DirectML libraries this process loaded, including while a Sherpa voice is using the GPU. The runtime loads on first use and unloads after the idle timeout. The status bar summarizes the loaded ONNX runtime and model count with their total file size. Click the bar to open that window.
| Section | Keys |
|---|---|
[ocr] |
pack, variant, device (Cpu / Gpu / IntelGpu), det thresholds |
[ui] |
hotkeys, tray, ui_lang, mod_*, tab_*_visible, update_check_days, http_proxy, capture_log, llm_log, screenshot_keep_days, screenshot_short, imgconv_* |
[http] |
OCR API (http_enabled, http_ocr and the other module flags, http_lan, port 1224), service mode, onnx_unload_min |
[sendfile] |
LAN file transfer ports, display name, web login password, paired devices |
[pdf] |
invisible text, raster DPI |
[asr] |
voice/live mode, polish, split, asr_llm |
[[llm]] |
OpenAI-compatible endpoints (name / url / key / model / think) |
[translate] |
local ONNX device, translate_llm, translate_llm_batch |
[record] / [gif_record] |
codec, CRF, audio, mouse overlay |
Do not commit a machine-specific config.toml.
Example config.toml
[ocr]
model_pack = "rapid-ch"
model_variant = "mobile" # pack variant id; a Chinese pack may store a Chinese label
device = "Cpu" # Cpu | Gpu | IntelGpu
det_limit = 960
det_thresh = 0.3
det_box_thresh = 0.5
use_cls = true
[ui]
hotkey = "Ctrl+Alt+O" # show / hide main window
hotkey_snap = "Ctrl+Alt+Q" # screenshot annotate
hotkey_snap_ocr = "Ctrl+Alt+W" # screenshot + OCR
# hotkey_translate = "Ctrl+Alt+T" # translate popup show/hide
# hotkey_dict = "" # dictionary tab; empty = off. Clipboard word only. Example: "Ctrl+Alt+D"
# dict_tts_zh_engine = "auto" # auto | onnx | sapi | winrt | edge; same keys for en, ja, ko
# dict_tts_zh_rate = 1.0 # 0.5–2; audio cache tmp/voice, kept 1 day
# dict_tts_zh_sapi = "" # local SAPI voice; empty = automatic
# dict_tts_zh_edge = "zh-CN-XiaoxiaoNeural"
# dict_tts_en_edge = "en-US-AriaNeural"
# dict_tts_ja_edge = "ja-JP-NanamiNeural"
# dict_tts_ko_edge = "ko-KR-SunHiNeural"
# hotkey_snap_copy = "Ctrl+Alt+P" # cycle copy as image / file / path
minimize_to_tray = true
capture_log = false # true → log/capture.log (DPI + save timings)
# llm_log = false # true → log/llm.log (polish HTTP; API key not written)
ui_lang = "zh" # zh | en
mod_ocr = true # false hides the page and stops its hotkey; default true
mod_tts = true
mod_asr = true
mod_translate = true
mod_chat = true
mod_face = true
mod_dict = true
tab_ocr_visible = true # main tabs; all default to true. Hide only; hotkeys stay
tab_tts_visible = true
tab_asr_visible = true
tab_chat_visible = true
tab_translate_visible = true
tab_face_visible = true
tab_http_visible = true
tab_sendfile_visible = true
tab_dict_visible = true
update_check_days = 7 # auto-check interval on startup (days); 0 = off
screenshot_keep_days = 3 # screenshot history; 0 = unlimited. Cleaned at startup only
screenshot_short = false # cap the shorter side (shrink only)
screenshot_short_px = 1080 # 16–16384
# http_proxy = false
# http_proxy_addr = "127.0.0.1:7897"
# ocr_translate_lang = "" # empty = off; LLM lang code (zh/en/ja/ko/…)
[http]
http_enabled = true
http_ocr = true # false returns code 810 for that path; default true
http_tts = true
http_asr = true
http_translate = true
http_chat = true
http_face = true
http_lan = true # false listens on 127.0.0.1 only
http_port = 1224
service_mode = false # keep engines warm; skips idle unload
onnx_unload_min = 5 # load ONNX on first use; unload sessions and the runtime after N idle minutes; 0 = never
[sendfile]
sendfile_enabled = true
# sendfile HTTP shares http_port (1224)
sendfile_udp_port = 17531
sendfile_name = "" # empty = machine name
sendfile_web_pass = "" # web manager login; empty = auto-generated on start
photo_fmt = "jpg" # phone app + web Camera: jpg | png
photo_jpg_quality = 60 # 1–100
photo_limit = true # cap the shorter side (shrink only)
photo_max_px = 2000 # 64–16000
# [[sendfile_device]] # paired phones (id / name / token)
[pdf]
pdf_invisible_text = true
pdf_dpi = 150
[record]
record_codec = "x264" # x264 | x265 | av1 | mf (mjpeg is read as mf)
record_fps = 24
record_crf = 28 # x264/x265 only, 0–51
record_av1_crf = 56 # AV1 only, 0–63 (56 ≈ half of x265 CRF28 size)
record_audio = true
record_audio_src = "Speakers" # Speakers | Mic | MicAndSpeakers
record_audio_kbps = 96
record_short = false # cap the shorter side (shrink only)
record_short_px = 1080 # 16–16384
record_lock_aspect = true
record_mouse = true
record_click_highlight = true
[asr]
asr_voice_mode = "stream" # stream | offline
asr_voice_polish = true
asr_voice_split = true
asr_voice_split_sec = 5
asr_live_mode = "stream"
asr_live_polish = false
asr_live_split = true
asr_llm = "gpt-4o-mini"
# asr_llm_prompt = "..."
# chat_llm = ""
# chat_llm_prompt = "..."
# chat_agent = true
# chat_auto_tts = true
[[llm]]
name = "gpt-4o-mini"
url = "https://api.openai.com/v1"
# key = ""
model = "gpt-4o-mini"
think = "low" # preset or typed token (minimal, …)
[translate]
translate_compute = "Auto" # Auto | Gpu | Cpu | Igpu
# translate_llm = ""
# translate_llm_prompt = "Translate the user's text to {dst}. Output only the translation."
translate_llm_batch = 8 # items per LLM call, 1–64
[gif_record]
gif_fps = 8
gif_max_size = true
gif_max_w = 1280
gif_max_h = 720
gif_colors = 128
gif_scale = 100
gif_mouse = true
gif_click_highlight = truethink: off sends thinking.type=disabled. Any other token (low / medium / high / max, or a typed value such as minimal) sends thinking.type=enabled plus reasoning_effort. If off is rejected, retry with low. Access to opencode.ai adds x-opencode-session / x-opencode-client. Old keys asr_llm_url / asr_llm_token / asr_llm_model are ignored. translate_llm_batch is how many lines one LLM translate call takes (default 8).
Set capture_log = true for log/capture.log (DPI / save timings; SLOW if ≥500ms). Set llm_log = true for log/llm.log (API keys are not written). CLI ScreenKit --snap dumps full-monitor bitmaps under log/snap/; --test-overlay-layout shows the screenshot overlay and logs per-monitor HWND/DPI.
When enabled, the OCR API listens on http_port (default 1224). http_lan = true (default) accepts this PC and the LAN; false listens on 127.0.0.1 only. File transfer, the web manager, and Wi-Fi cast share this port and need LAN access left on. There is no authentication — do not expose it on an untrusted network.
LAN PC file transfer shares this HTTP port (1224) plus UDP 17531 (pairing required): HTTP-API.md · android/README.md.
GET /api·/api/statusGET/POST /api/cast/stop— close screencast viewer now- WebSocket
/cast— screencast media (same HTTP port) POST /api/ocr—boxis original-image pixelsPOST /api/qr— barcode / QR only (/api/barcode)GET /api/ocr/get_optionsGET /api/asr/models·POST /api/asrGET /api/tts/models·POST /api/tts— Sherpa, SAPI, Windows (engine=winrt), Edge (engine=edge)POST /api/itnPOST /api/translate— LLM batch (items[])GET /api/face/models·POST /api/face
Full field reference: HTTP-API.md · Chinese HTTP API.
ScreenKit --image <path> [options]
ScreenKit --snap [--out <dir>]
ScreenKit --test-overlay-layout # screenshot overlay HWND/DPI per monitor
ScreenKit --test-overlay-span-adj # cross-monitor region handles on guest screens
ScreenKit --test-clipboard-path # path copy after delayed image; 4K timing
ScreenKit --test-apk-qr # encode/decode LAN APK QR; HTTP GET /apk
ScreenKit --test-img-convert # png→jpg rotate 90 + box 100×100 + shorter side 40
ScreenKit --test-qr-make
ScreenKit --test-rename
ScreenKit --test-hash
ScreenKit --test-texttool
ScreenKit --test-pwgen
ScreenKit --test-nettool
ScreenKit --test-wintop # list windows; pin/unpin a probe HWND
ScreenKit --test-dict-search # read-only lookup in dict.db beside the exe
ScreenKit --test-dict-sel # copy a selected word with Ctrl+C and read it back
ScreenKit --test-dict-word # clipboard text is one dictionary word?
ScreenKit --test-dict-tts # dictionary voice cache kept 1 day; rate math
ScreenKit --test-sendfile # sendfile sandbox + web login / public download
ScreenKit --test-face-overlay
ScreenKit --list-models
ScreenKit --list-face
ScreenKit --list-sapi # local SAPI + (x64) x86host voices
ScreenKit --test-tts-sherpa <model> # `-d auto|gpu|cpu`
ScreenKit --list-edge-tts
ScreenKit --test-edge-tts ko-KR-SunHiNeural
ScreenKit --test-http-tts
ScreenKit --test-llm-chat
ScreenKit --test-llm-agent
ScreenKit --test-http-chat
ScreenKit --probe-cuda
ScreenKit --help
Useful options: -d gpu|cpu, -p rapid-ch, -v <variant>, -m <models-dir>, --det-limit, --no-cls.
Some classic SAPI voices register only for 32-bit processes. Ship x86host.exe next to ScreenKit.exe (built from x86host/; a Release build of ScreenKit also builds and copies it).
| Item | Detail |
|---|---|
| Role | HTTP helper: list SAPI voices + synth WAV; no GUI |
| Start | On demand by the x64 app, or run x86host.exe |
| Bind | 127.0.0.1 only, default port 17886 |
| Idle | Exit after 60s (--idle-ms to override) |
| API | GET /api/sapi/status · GET /api/sapi/voices · POST /api/sapi/synth · POST /api/sapi/shutdown |
dotnet build x86host/x86host.csproj -c Release
x86host.exe --port 17886 --idle-ms 60000
x86host.exe --list-sapi
In the UI, engine SAPI lists local voices plus x86-only entries (name ends with · x86).
OCR/
├── ScreenKit/ # WPF app (net48, x64)
│ └── bin/Release/
│ ├── net48/ # Dev output (models / runtimes live here)
│ └── ScreenKit/ # Slim package: exe + x86host.exe + managed deps
├── x86host/ # 32-bit SAPI HTTP helper
├── android/ # Companion app (com.whj.screenkit)
├── docs/ # README screenshots
├── scripts/publish-release.mjs
├── README.md · README.zh.md · CHANGELOG.md
└── HTTP-API.md · Chinese HTTP API doc
Model packs and large native runtimes are not in git. Place them next to the exe (or install in-app):
ScreenKit/bin/Release/net48/
├── ScreenKit.exe
├── config.toml
├── ocrmodels/ asrmodels/ ttsmodels/ translatemodels/ facemodels/
├── onnxcpu64/ onnxgpu64/ onnxdml64/
└── ffmpeg64/
A Release build does not copy those folders. Each OCR pack needs ONNX + configs.txt (and dict/keys). Optional pack.json (name / nameEn / variants) supplies English UI labels; defaults live in ocr-display.json.
cd ScreenKit
dotnet build -c Release
./ScreenKit/bin/Release/net48/ScreenKit.exe- Includes:
ScreenKit.exe,x86host.exe, managed deps,wetext/(ITN), Assets, LICENSE. - Does not include: the Android APK, OCR/ASR/TTS/face models, ORT, OpenCV/Skia/PDFium/Sherpa natives,
opencv_videoio_ffmpeg4110_64.dll,ffmpeg64,dict.db,ZXing.dll,SharpCompress.dll,SharpSevenZip.dll,7za.dll, ore_sqlite3.dll. Install the dictionary from Help → Install features, or placedict.dbbeside the executable yourself. See Dictionary for where that data comes from. Barcode, archive extract, and the dictionary each ask to install the missing library when you use them. - End users install those via Install features. Local Opus-MT ONNX is placed under
translatemodels/by hand if needed.
For local development with models already present, run bin\Release\net48\.
node scripts/publish-release.mjsRelease-builds, then packs ScreenKit/bin/Release/ScreenKit/ into release/screenkit_<version>.7z (needs 7-Zip on PATH). release/ is gitignored.
Release documentation:
- Every
CHANGELOG.mdversion has matching English and Chinese sections. - Every GitHub Release description is bilingual (English first, Chinese second).
- Checksums and links are listed once after both summaries.
ScreenKit application source is MIT. See LICENSE.
Copyright (c) 2026 ScreenKit Contributors
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND.
Bundled or optional dependencies are not all MIT (models, FFmpeg, CUDA/cuDNN, some natives). See THIRD_PARTY_NOTICES.md. Do not commit large ONNX weights, CUDA redistributables, or FFmpeg shared binaries without a redistribution plan.
- CHANGELOG.md
- HTTP-API.md · Chinese HTTP API
- android/README.md — companion app
- LICENSE · THIRD_PARTY_NOTICES.md
- README.zh.md