Slimy_CueTimer

There's a moment in every long render where you stop doing anything useful and just stare at the KSampler's progress bar. Slimy_CueTimer is the node for exactly that situation. It's a progress monitor you drop anywhere on the canvas - no wires in, no wires out - that shows a big stopwatch, separate steps and whole-queue progress bars, speed in it/s, an estimated time remaining, and a live preview, then pings you when it's done.

The Python class is almost embarrassingly small. INPUT_TYPES is empty (it only grabs the hidden prompt/unique_id), it outputs nothing, and execute() just returns {}. All the real work happens in the browser: the pack's frontend extension subscribes to ComfyUI's execution events - execution_start, executing, progress, executed - and paints everything onto the node's canvas. The clock is global, not per-node: when the queue starts, every CueTimer node on the graph shows the same elapsed time.

PeepPreview is the part that's actually clever. The pack also ships a backend patch (video_preview_patch.py) that wraps latent_preview at the module level, so it picks up KSampler previews even when the sampler is buried inside a subgraph - the usual node-to-node wiring can't see those. For video jobs it decodes up to 16 frames per step as a strip (cheaply, via latent2rgb; it deliberately avoids loading TAESD so you don't leave extra VRAM pinned on the GPU) and slides the window backward along the video timeline as steps advance. A 20-step, 100-frame render shows frames 0–7 at step one, 50–57 at step ten, and lands at the end when it finishes. For still images it just falls back to the standard single-frame preview.

The "estimated remaining" bit is a small history model: it averages your last five completed runs (ignoring anything under 15 seconds and anything that errored), then caps the prediction at 98% so it never looks done early. History sticks around after a run and lives in the node's properties, so it's saved along with your workflow JSON.

What you actually touch

The entire control surface is six checkboxes painted across the node: systemNotify (browser notification on finish), peepSound (a chime - rising notes for done, a low two-note for errors/interrupts), Timer, Peep, VRAM Clear, and AutoPlay. That's it. No inputs, no outputs, nothing to miswire - which is the whole point of the node.

VRAM Clear is worth knowing about. On queue completion - and on error or interrupt too - it calls ComfyUI's /free endpoint with unload_models and free_memory, the same thing ComfyUI-Manager's "Free model and node cache" button does. If you're chaining jobs in a queue, each run ends with models unloaded, so the next job pays a small reload tax. Usually that's the trade you want if you keep hitting OOM between runs.

Installing it

Nothing to it - no requirements.txt, no pip dependencies, no model downloads. Either use ComfyUI Manager (search "Slimy_CueTimer") or:

cd ComfyUI/custom_nodes
git clone https://github.com/Slimy-Comfy/Slimy_CueTimer

Then restart ComfyUI. The README lists "Slimy Custom Nodes" as a requirement, but that's the author's umbrella branding - this pack runs standalone on stock ComfyUI.

Gotchas

• The README advertises an AutoQueue feature that auto-starts the next queued job after the current one. The shipped code doesn't implement it - the six checkboxes above are the full feature set. Trust the source over the README here.
• Notifications need browser permission, which the node requests on first run. If alerts never appear, check the site's permission settings - and the tab has to stay open, since everything is frontend-side.
• The fancy timer font (Orbitron) loads from Google Fonts, so on an offline machine you get a plain monospace fallback. Cosmetic only.
• This is a very new pack (last commits August 2026), so expect rough edges, feature drift like the AutoQueue above, and zero community write-ups yet.

Verdict

If you're happy with ComfyUI's status-bar progress, you don't need this. But if you run long batches or overnight video renders, the notification-on-finish alone earns its keep, and the sliding-window preview for video is something no other pack does. It's free, dependency-free, and hard to break. Drop one next to your KSampler and let it stare at the render so you don't have to.