VELVET VICE — Preflight Console

A Velvet Vice LTX workflow is a Rube Goldberg machine of dependencies: a reference image, a BF16 base model, a Gemma text encoder, VAEs, an Ollama server writing the prompt, FFmpeg and NVENC for output, a watermark file, and an output directory. Any one of those being missing only announces itself after you've queued the render and waited. VelvetVicePreflightConsole is the pack's answer: a non-blocking environment check you run before queueing, so the missing piece gets found in seconds instead of after a 10-minute failure.

How it works

This one is unusual in a good way: it's a UI node. On the backend its function is a literal no-op - it has no outputs and computes nothing. All the work happens in the node's console panel, drawn by the pack's web JavaScript. When you click RUN PREFLIGHT, the frontend walks the checklist against your local ComfyUI install and workflow state: the reference image is present, the core model files and VAE files resolve to real paths, Ollama is reachable, FFmpeg is installed, NVENC works, the watermark file exists, and the output directory is writable. Results print to the node's console in a readable pass/fail report.

Because it's non-blocking and has no outputs, it can sit anywhere in the graph without becoming part of the data flow - it won't interfere with execution, and ComfyUI won't try to route data through it.

The few inputs

Four toggles, and they're all about scope, not behavior:

• auto_check_on_load (true) - run the check automatically whenever the workflow loads, so problems surface before you even think about queueing.
• check_ollama (true) - include the Ollama server check. Turn this off if you're running a workflow variant that doesn't use the LLM prompt path (there are Director modes that make no Ollama calls, so this is a legitimate toggle, not a footgun).
• check_nvenc (true) - include the NVENC probe. Disable if you're deliberately using CPU-only output.
• strict_report (false) - when false, problems are reported as warnings and you can queue anyway; when true, the console hard-flags failures. This is a report, not a gate - even in strict mode the node won't physically block a queue, it just makes the problem unmissable.

Installing it

Part of the Velvet Vice LTX pack (registry velvet-vice-ltx). ComfyUI Manager → search "VELVET VICE - LTX", or:

cd ComfyUI/custom_nodes
git clone https://github.com/Velvet-Vice/velvet-vice-ltx

Restart ComfyUI, then hard-refresh the browser with Ctrl+F5 - this node is frontend-driven, and a stale browser cache is exactly how you end up with a preflight console that shows nothing. The pack has no Python dependencies.

Common issues

• The console is blank / the RUN PREFLIGHT button does nothing - the web files didn't load. Hard-refresh first (Ctrl+F5), then check the browser console for a failed script load from the pack's web directory. If you upgraded an older Velvet Vice install, remember the README's warning: never merge an old node directory into a new release - reinstall fresh.
• It reports a missing model file you know you downloaded - the preflight checks exact filenames and paths from the pack's model profile (things like 10Eros_v1_bf16.safetensors and the LTX 2.5/gemma4_e2b_it_bf16.safetensors text encoder). Wrong subfolder or renamed file = "missing." This is a genuinely useful catch, not a false alarm.
• Ollama shows as failing but you don't use it - either start ollama serve, or flip check_ollama off for a non-LLM workflow variant.

There's a whole genre of "health check" nodes in ComfyUI that are more theater than tool. This one is worth the wire because the pack's dependencies are specific - exact filenames, exact folders - and that's precisely what a machine can check faster and more honestly than you can by eye.