Start here — Describer Workbench 0.8.1
For Windows, macOS and Linux · documentation updated November 9, 2026.
Make one short description first. You do not need an AI account, programming, or a local AI model for this walkthrough.
1. Download for your computer
| Computer | Download | Install and start | Publisher signing |
|---|---|---|---|
| Windows x64 (Windows 11 tested) | Portable ZIP | Extract everything; open DescriberWorkbench/DescriberWorkbench.exe. Keep _internal beside it. |
Scott Converse, Azure Trusted Signing with Microsoft timestamps |
| Mac, Apple Silicon, macOS 15+ | arm64 DMG | Open DMG; drag the app to Applications. | No Apple publisher certificate or notarization; ad-hoc integrity only |
| Mac, Intel, macOS 15+ | No Intel DMG (dropped in 0.7.0) | Install from the repository; see the platform guide. Not tested on Intel. | None (source install) |
| Ubuntu 24.04 / compatible Debian desktop, x86-64 | DEB installer | sudo apt install ./DescriberWorkbench-0.8.1-linux-x64.deb; open Describer Workbench from the applications menu. |
Unsigned |
| Other Linux x86-64 desktops | Portable tar | Extract; run DescriberWorkbench/DescriberWorkbench after installing native dependencies. |
Unsigned; other distributions not acceptance-tested |
Mac: choose Apple Silicon for an M-series Mac (Apple menu → About This Mac). There is no Intel Mac download. Use the normal Privacy & Security approval if macOS blocks opening the downloaded app; no Apple publisher certificate or notarization is available. Do not disable Gatekeeper. Linux: the DEB installs required media and speech dependencies; the tar requires you to install them yourself. See platform installation.
Each release provides SHA256SUMS.txt for checking download integrity. Windows is the only publisher-signed distribution. Ollama, large AI models, subscription tools and videos are installed separately. Repository setup for Windows is in the manual; Mac/Linux have a per-user installer.
2. Open a video and save a project
- Start Describer Workbench. Choose Open video and select an actual MP4, MKV, MOV, AVI or WebM file. Extract downloaded archives first.
- Press the Play/Pause icon. Confirm that the picture and original sound play.
- Pause where you want to describe something. Choose Add at playhead in Descriptions.
- Type one short sentence about what is visible. The new cue starts at the playhead and lasts three seconds, bounded by the video end. Edit Start/End if needed.
- Choose File → Save project, using a name ending in
.adproject. The project refers to the original video; keep both.
3. Listen and export
Enable Speak descriptions and play the passage. Choose an installed narration voice in Settings → Export voice or in Export: Windows voices on Windows, macOS voices on Mac, eSpeak NG on Linux. OpenAI speech is optional and sends text to a paid remote service after confirmation.
Use Export → Subtitles · .srt for a simple text output. For narrated MP3/WAV or described MP4, the engine measures narration and checks independent speech protection. If asked, explicitly install the small speech detector; no large vision model is needed. Missing detector or unresolved cues block complete narrated output. A sentence that cannot fit safely pauses the original soundtrack; MP4 also freezes the picture. Listen to the exported file before sharing it.
4. Try an AI draft when ready
Since 0.7.0: your existing descriptions stay by default. After drafting, review the proposals and explicitly approve any replacements. Cancel keeps your project unchanged; Ctrl+Z undoes the complete accepted draft.
Use Save project as to keep a separate comparison copy. Choose Draft with AI, select your connection and exact model ID, then try Next 20 seconds. Local Ollama needs a separately installed vision model. Claude, Codex and Antigravity use their installed CLI sign-in or the existing subscription gateway. API credits are separate. Connection setup explains each route and its limits.
City and council meetings (0.8.0 and later): Public meeting mode describes only what changed, reads slides and finds votes. It has no switch in the window yet. See Public meeting mode in the manual.
Review facts, names, omitted actions, printed text and timing. Completion and a green placement status do not prove accuracy. Sampled-frame routes cannot hear dialogue. The engine protects sound independently, but that detector can also make mistakes.
5. Keep your work and update safely
Save the project, source video and engine narration workspace together. Reopen the project to continue editing. When moving computers, locate the video again, re-enter credentials and repair narration if the referenced audio is missing. Installed voices and Windows-protected keys do not migrate.
Windows uses the signed updater or a separately extracted portable ZIP. Mac/Linux updates are manual: quit the app, download the matching architecture and replace/install it while retaining the previous copy. Settings and projects are separate from the application. The full manual covers recovery, shortcuts, privacy and troubleshooting; the technical guide covers CLI, API and MCP.
Native package checks cover startup, storage, measured voice, exports and developer transports. Active screen-reader usability and human listening quality have not been certified. See the platform evidence and limitations.