Voice-powered development, prompt engineering, and building with Invoke.
Where to download WhisperDesktop.zip, which model files it needs, what it still does well on AMD and Intel GPUs, and what to use instead now that the project has gone quiet.
Half the tools in old voice coding articles are dead. A maintained map of the landscape: what still works, what stalled, what shut down, and what to use instead.
Gemini CLI can paste clipboard images with Ctrl+V and reference files with @ syntax, but the behavior differs by platform and WSL breaks it entirely. The full map.
Cursor's terminal agent accepts images, but Ctrl+V hits the same walls as every terminal AI tool. Here is what works: paths, paste, drag, and the one-hotkey version.
Codex CLI takes images three ways: the -i flag, Ctrl+V in the TUI, or a file path in the prompt. Here is each one, plus the Mac paste quirk and what to do when paste fails.
claude update runs, says it worked, and the version number doesn't move. Or it fails with a permission error. Here is the diagnosis order that untangles it.
Your screenshot is on your laptop's clipboard. Claude Code is running on a server. Ctrl+V will never bridge that gap, but these four workflows do.
Claude Code reads screenshots, mockups, diagrams, and photos, and it's one of the most useful things about it. Here is what it can do with an image, what it can't, and how to get one in front of it.
You copied a screenshot, hit paste in Cursor, and nothing attached. Here is why it happens and every fix that works, from the chat pane to the integrated terminal.
WSL has its own clipboard, separate from Windows, which is why screenshots won't paste into Claude Code. Here is what's happening and the fixes, from Alt+V to file paths.
Claude Code rejects images over 5 MB, and a full-resolution screenshot blows past that easily. Here is how to shrink it fast without losing the detail that matters.
You press Ctrl+V in Claude Code and nothing happens, or you get plain text instead of your screenshot. Here is what's actually eating the paste and how to get around it.
You do not need a monster GPU for local voice dictation. Almost any NVIDIA card from the last eight years does it in real time. Here is what actually matters: VRAM, model choice, and headroom.
MacWhisper is Mac-only and there is no Windows version. Depending on which half of MacWhisper you wanted, the dictation half or the transcription half, here is what fills the gap on Windows.
Claude Code can look at a screenshot of a UI and build it, or look at a screenshot of a bug and fix it. Here is the workflow I actually use, including the prompts that work.
You copied a screenshot, pressed Ctrl+V in Claude Code, and got no image found in clipboard. The image is right there. Here is why Claude Code can't see it and the fixes, ranked.
There are four ways to get an image into Claude Code, and which one works depends entirely on your terminal. Here is the full map: paste, drag, file paths, and the one-hotkey option.
Both run the same Whisper weights and produce the same text. The choice comes down to your hardware and how you plan to ship it. Here's the honest breakdown, including where each one wins.
Whisper runs entirely on your own machine, no API key and no internet. Here's how to get it running on Windows, which runtime to pick, and how to fix the CUDA errors everyone hits.
Most Otter.ai alternatives are meeting notetakers, because that's what Otter is. If you're a developer who actually wants to write by voice, you need a different category of tool. Here's both.
faster-whisper runs the same Whisper weights about four times faster on less memory. Here's the actual reason why, which settings matter, and what I learned shipping it in a production app.
There are genuinely good free dictation tools, and for most people one of them is enough. Here's an honest ranking of what's free, what each one costs you in other ways, and when paying starts to m...
Willow is cloud-based voice dictation on a monthly subscription. If you want the same push-to-talk feel but local, private, and paid once, here's the Windows alternative I built.
Whisper ships as code, not an app, so 'Whisper for Windows' means seven different tools depending on what you're doing. Here is every real option: const-me's WhisperDesktop, Whisper-Standalone-Win,...
Claude Code takes natural-language prompts all day, and typing them is the bottleneck. Here's how I dictate prompts straight into Claude Code with local, GPU-fast voice-to-text.
I write most of my prompts by voice now, into Claude Code, Cursor, the terminal. Here's the actual setup, the hotkeys, and why it turned out faster than typing.
Wispr Flow is $144/year. Voicy is $102/year. Here's what you can use instead for a one-time payment, or free.
Everyone talks about 150 WPM vs 40 WPM. Speed isn't the real win. The real win is that spoken prompts are actually better than typed ones.
I tested every voice dictation tool I could find on my Windows dev machine. Most of them are built for Mac users writing emails. Here's what actually works for coding.
Talon is a full voice-controlled coding environment. Invoke is push-to-talk dictation. They solve different problems and you might want both.
SuperWhisper is Mac-first with a limited Windows version. If you want local Whisper on Windows with push-to-talk and project-aware reformatting, here's what I use.
Terminal emulators like Hyper, Wezterm, and Windows Terminal can't paste images into Claude Code. Invoke's screenshot-to-path feature fixes that with one hotkey.
MacWhisper is a clean Mac transcription app. Invoke is a developer-focused dictation tool with push-to-talk and AI reformatting. Here's how they compare.
Wispr Flow is great, if you're on Mac, want a subscription, and don't mind cloud transcription. Here's what I built instead.
I dictate thousands of prompts a month. Every transcription tool I tried was either too slow, too expensive, or too dumb. So I built my own.
You can dictate prompts directly into Cursor's chat and composer. Here's how to set it up with local Whisper transcription.
Cloud transcription adds latency, costs money monthly, and sends your audio to someone else's servers. Here's why local GPU transcription is better for developers.
Download Invoke and start dictating in under a minute. Free for 7 days.