If you installed n8n with npm install -g n8n, that install route goes away in 3.0. Your install keeps running today. It just can't upgrade past that point. - Copy the .n8n folder into a Docker volume. The encryption key rides along inside it, so every credential still opens. - Skip export and import. It reports success, and then every credential is dead because the new instance encrypts with a different key. - Run the last version 1 release once before going to version 2. Straight to 2 and the workflow list comes up empty with no error. - Pin the image tag and set your timezone in the compose file. Schedules run on the container's clock. - The old folder is never touched, so starting the npm install again puts you right back where you began. Docker Desktop on a Mac or Windows machine you sit at, Docker Engine and Compose on a headless server. Same container either way. The compose file is in the video description. Any questions? Let me know. Cheers, Kawika

Studio 1.10.0 is live. The headline is workers: you can now give each machine running a worker its own credential, and a worker you did not expect cannot quietly join. Worker keys you can revoke one at a time Mint a one-time join token under Infrastructure > Workers, pick which queues it covers, and hand it to the machine you are setting up. That machine gets its own credential, limited to those queues. Retiring a machine is one click, and it does not touch any other worker. New workers wait for your approval If a worker outside your install connects with the shared secret, it now shows up as a pending request instead of joining. You approve or reject it, and you can narrow which queues it gets before you approve. Also in this release - Wait for Webhook steps pause the run and pick back up when the callback arrives, with no worker tied up while it waits. - Uploads have a size limit, 1024 MB by default, set with SHS_MAX_UPLOAD_SIZE_MB. - Two files with the same name keep their own contents, and replacing a file no longer changes copies that other steps are using. - Storage totals count every file in your organization. - The Infrastructure page refresh picks up new worker requests without reloading the page. To upgrade Run studio-console upgrade and pick 1.10.0. If you run workers outside Docker with the studio-workers package, upgrade them to 1.11.0 as well (uv tool upgrade studio-workers), since Studio checks that worker and API versions match. Docker workers update with the rest of the install. Nothing else changes: existing workers keep running on the shared secret, and the new enrollment is there when you want it. Any questions? Let me know. Cheers, Kawika

Studio 1.8.0 is out. The video worker got a lot lighter and a lot quicker. ✅ WHAT'S NEW - Video worker, 1.26 GB instead of 2.92 GB. Less than half the download, and it no longer requests a GPU, so it runs on any box and leaves your GPU free for image generation. - Subtitles about 2.3x quicker on CPU: 6.2 seconds against 14.4 on the same 64 seconds of narration. Same Whisper model, same transcript. - Whisper models stay put. They live on your workspace volume, so they survive a container recreate and every video worker on the volume shares one copy. - Workers tell you at startup. A worker checks its own requirements before it registers, so a missing engine or an unwritable cache path shows up with a reason instead of a run of failed jobs. - Gated model files are visible up front. Provider docs show the file set and the access column per package, so you know which files need a Hugging Face terms acceptance before you start the install. Flux.1 Schnell 1.2.2 and Schnell GGUF 1.2.3. - Patched base packages across every image in the release. 📋 UPGRADE IMPACT - subtitles_backend is gone from the shs/video provider. Faster-Whisper is the engine the image ships. - Stored workflows are migrated on upgrade, so nothing you have built needs editing. - Requires shs/video 1.1.0 or later. Questions? Drop them in the comments. Kawika

ComfyUI installs with an empty models folder. Nothing runs until you fill it, and nothing tells you what to put in there. Here is the whole file set for Flux.1 Schnell GGUF: which four files, where each one goes, the custom node Manager does not install for you, and the one file that is gated and fails until you accept the terms. https://www.youtube.com/watch?v=40HC-M4QeuY The four files, about 12 GB total: - flux1-schnell-Q4_K_S.gguf, 6.8 GB, into models/diffusion_models/FLUX1 - t5xxl_fp8_e4m3fn.safetensors, 4.9 GB, into models/text_encoders/t5 - clip_l.safetensors, 0.2 GB, into models/text_encoders - ae.safetensors, 0.3 GB, into models/vae/FLUX1 8 GB of VRAM runs this. The full precision version of the same model is 23.8 GB and makes pictures you cannot tell apart. Three things that will cost you an hour if nobody tells you: 1. Manager will not install the VAE. Gated repository, so the install just errors. Accept the terms on Hugging Face once, grab the file yourself, and create models/vae/FLUX1, which does not exist yet. 2. Restart ComfyUI after every model install, and do a real browser refresh. A file on disk that has not been imported does not exist as far as the loader is concerned. 3. Manager's Save Path column says "default" for clip_l and never tells you where that resolves to. Only the install log says. Check Filter: Installed afterwards, every time. This is one of four ComfyUI workflow packages in Studio, and they overlap. clip_l and the VAE are shared across all three Schnell builds, so adding a package later is usually one new file, not a fresh start. All four installed comes to roughly 92 GB of weights. At the end the video runs a faceless video workflow on these files. Fifteen images, generated locally, no API call and no per image charge. Commands and the full checklist are in the video description. Questions, drop them below. Cheers

If you have been meaning to run Studio and stalled at setup, start here. Windows 11 with WSL2, start to finish: clean machine to the full platform running. https://www.youtube.com/watch?v=zJgbnvfbWQI What you have by the end: - Studio running under Ubuntu on WSL2, API and web UI on localhost - Docker Engine inside WSL, no Docker Desktop needed (though it works too if you already run it) - All seven workers online - The community catalog installed: providers, prompts, and workflows - If you have an NVIDIA card, the GPU wired through to Docker. No GPU? Skip that chapter, everything runs on CPU - A clean full stop and restart, so you know how to bring it back This is the foundation video for Windows. Every walkthrough after this one starts from a working install and links back here instead of repeating setup, so you install once and never sit through it again. Hit a step that did not go the way it went on screen? Drop it in the comments. Cheers