Blog

n8n 3.0 breaking changes that kill a self-hosted FFmpeg workflow: removed binary file nodes, a 60-second runner timeout, a 256 MiB cap, and what to change.

RunPod FFmpeg on a rented RTX 4090 lists at $0.33/hr. The real arithmetic nobody quotes: idle time, egress, NVENC vs CRF, and where a per-minute API wins.

n8n 3.0 breaking changes kill the npx + host ffmpeg video step. Here's the Docker Compose and distroless Dockerfile that work, and the HTTP alternative.

The Squarespace video size limit is 500 MB per file and 30 minutes site-wide. Get the bitrate math, an H.264 preset, and a batch encode for client libraries.

"No usable ffmpeg was found" in n8n means the server has no FFmpeg binary, and n8n's installer makes it impossible for a community node to ship one. Three real fixes.

Going from a transcription API to SRT is where caption pipelines break. Fix timestamps, line wraps, and escaping, then burn them on with FFmpeg or one call.

YouTube now demotes template-based bulk edits. How to repurpose long form video to shorts with per-clip cuts, real 9:16 crops, and captions from each segment.

FFmpeg impossible to convert between the formats supported by the filter? Two branches of your graph disagree on pixel format. Fix it before the overlay.

The gpt-4o-transcribe retirement hits October 15, 2026 with no replacement listed. Two migration paths, plus the API chain that burns captions into video.

Make's ChatGPT plugin is still in limited beta. Set up Make MCP in ChatGPT instead, and send the real video render to a job API that won't time out on you.