Hi all, does anyone else find editing is the actual bottleneck on Reels, not the ideas or the filming? Cutting silences and filler words out by hand was eating more of my time than either of those. Had Claude Code help me build a small local script that handles the boring part automatically. 1. Record raw footage as normal. 2. Run faster-whisper locally for a transcript (free, no API key, stays on your machine). 3. Use the transcript timestamps to auto-cut dead air and filler words. 4. Burn in captions from the same transcript instead of typing them out separately. Curious if anyone else has automated this part, or if I was just avoiding it longer than I should have.

Fits the spirit of this place, so sharing it here. Dictation apps kept wanting a monthly subscription, so we built one with Claude Code that runs on our own machine: FU Flow. Right Alt to start, right Alt to stop, and the text is typed at your cursor. Whisper runs locally, offline after setup, no account, unlimited words. Windows 10/11, and Ubuntu 22.04+ as of today. MIT, source on GitHub. fuckyouflow.app Happy to share how we handled the 1.7 GB model download and the auto-updater if anyone is building something similar.

📜 Ethan Mollick posted on LinkedIn about getting Claude to make explainer videos drawn entirely in code. No image generator, no stock footage, just code drawing every frame. I liked it enough to build it into a Claude Code skill, and the video on this post is what it makes. ⚓ What the skill does - I point /explainer at a post, a doc or a topic - Claude writes a 20 to 30 second script from it, facts only from the source - The voice gets made, then Claude uses Whisper to find the exact second each word is spoken - Every animation is tied to a word, not a clock time. When the narration says "voice", the Voice step pops in - Claude draws the frames on an HTML canvas and renders MP4s in 16:9, 1:1 and 9:16 - Cost per video: $0 ⚓ About the voice The voice is mine, cloned with Pocket TTS and running locally on my own machine. You don't need that. You can paste the script into ElevenLabs (or any TTS you like), or record it yourself, hand Claude the audio file, and have Claude match the animation to it. ⚓ Why this video jumps around My videos normally stick to one style start to finish. For Claude Code Pirates the default is a pirate ship at sea. This one hops through every world I've built so far, just as a demo, so you can see the range: - Pirates: cannon broadside on every point, a kraken for the warning - Platformer: a side-scroller that bonks a block on every point - Shooter: a starfighter blasting an alien per step - Girder: barrels, ladders and a boss robot - Pow: 60s TV superhero, KAPOW on every hit That's the part Mollick's post showed well: once the engine exists, a new look is just another drawing function. Ask Claude for "a new world" and describe it. ⚓ Build your own I'm not handing out my exact skill, since it's wired into my own setup. Instead I put a new lesson in the Classroom with the kickoff prompt and a skill outline you can copy, paste and make your own: - 🏴‍☠️ Build Your Own Explainer Video Skill (in 🗺️ Quick Wins & Recipes)

Anyone else building an AI feature for a client and immediately hit the "no budget for API credits" problem? Ran into this on a room visualiser I'm building (client uploads a photo of their room, gets an AI preview of what new curtains would look like on their windows). Didn't want to put a card on OpenAI credits for something still in testing, so spent some time finding a free way to do it. 1. Cloudflare Workers AI has a free daily allowance for image models like FLUX.2, no card needed to start. 2. you just need a Cloudflare account ID and an API token, then you're calling the model directly instead of going through a paid provider. 3. keep input images under the provider's size limit (512px on the long edge worked for me), the output still upscales fine. 4. worth checking the provider's data policy too. Cloudflare states it doesn't train on customer content, which mattered since this is client-facing. Not sponsored, just glad it exists. Happy to share the exact setup if anyone's hitting the same wall.

Two small Claude Code tricks that took me one evening to set up, plus a refinement of Matthew’s canary idea that came out of it. The problem Claude has no clock. Mid-session it genuinely does not know if it’s been ten minutes or six hours, whether it’s afternoon or nearly midnight — and in long sessions it slowly gets sloppy about early instructions without anyone noticing. Two cheap fixes and one tripwire: Trick 1 — inject the time (and turn count) every prompt Claude Code has hooks — scripts that fire on events. A UserPromptSubmit hook runs every time you send a message, and whatever it prints gets injected into Claude’s context. So: a tiny script that prints the turn number and wall-clock time. ~/.claude/hooks/turn-time.sh: #!/bin/bash # Injects "[turn N · HH:MM TZ]" into Claude's context on every prompt. STATE_DIR="${HOME}/.claude/state"; mkdir -p "$STATE_DIR" SESSION_ID=$(python3 -c "import json,sys; print(json.load(sys.stdin).get('session_id',''))" 2>/dev/null) [ -z "$SESSION_ID" ] && exit 0 F="$STATE_DIR/turns-$SESSION_ID" COUNT=$(( $(cat "$F" 2>/dev/null || echo 0) + 1 )); echo "$COUNT" > "$F" echo "[turn $COUNT · $(date '+%H:%M %Z')]" Wire it up in ~/.claude/settings.json (and chmod +x the script): { "hooks": { "UserPromptSubmit": [ { "hooks": [ { "type": "command", "command": "~/.claude/hooks/turn-time.sh" } ] } ] } } Cost: roughly 10 tokens per turn — noise. Benefit: Claude knows what time it is, how long the session has run, and can pace itself (“it’s 11pm and turn 80 — want to wrap?”). Session notes get real timestamps too. Trick 2 — timestamps in the UI One setting, no script: { "showMessageTimestamps": true } Every Claude response gets stamped with its arrival time in the terminal — like a shell prompt for your conversation. (Display-only; it doesn’t enter the transcript content. That distinction matters for trick 3.) Trick 3 — the canary, with an answer key Matthew’s original idea: put an instruction early in your setup like “always begin every response with the word X.” When Claude stops doing it, your context is getting muddy — the canary died, get out of the mine.