🚀 Loop Trigger (Auto-Queue)

You know the feeling. You've got a 900-frame video, your GPU can render maybe 80 at a time, and ComfyUI is staring at you waiting for the next click. So you sit there like a vending machine attendant, bumping an index, re-queuing, waiting, bumping again. SequentialLoopTrigger is the node that fires the vending machine for you. Hit "Queue Prompt" once, and it keeps re-queuing itself until the whole timeline is done.

This is the heart of the pack's whole pitch - the "recursive auto-queue" that turns ComfyUI into a continuous rendering engine. The pack's own wiring guide puts it at the very end of the graph, and honestly that's the right place: it's the exit door that decides when to send the whole workflow around again.

How it works

It's an output node with a single input, trigger_dependency (type *). You wire whatever your pipeline ultimately produces into it - usually the ALL_IMAGES output of the pack's Incremental Auto-Stitcher. The value itself is ignored; the wire exists so the trigger only runs after everything before it has finished.

When it does run, it checks the pack's global progress counters (frames accumulated so far vs. the total the calculators registered), then:

• bumps loop_idx on the SequentialLoopStart node,
• randomizes every seed/noise_seed it finds in the workflow,
• flips any MasterSwitch inputs to the right is_final_cycle value,
• and POSTs the entire prompt back to ComfyUI's own HTTP API at 127.0.0.1:8188.

No API key, no external service - it's talking to the same ComfyUI you're using. That's also why the whole pack cares about your port: if you run ComfyUI on something other than 8188, the Loop Start node detects it and forces the real system port for safety.

Wiring it up

The README's recipe, compressed:

1. VideoAnalyzer at the front feeds total_frames into an Auto Loop Calculator.
2. The calculator hands chunk_frames / skip_frames to the sampler side (usually through VHS_LoadVideo).
3. The Stitcher accumulates frames and feeds the Trigger.
4. Press Queue Prompt once.

It stops when the accumulated frames reach the source count - or when a batch loader reports "no more batches." One thing that surprises people: each cycle's seeds are randomized, so every chunk is a fresh sample. That's deliberate, not a bug. Continuity between chunks is the job of the keyframe/continuity nodes elsewhere in the pack, not this one.

The real gotchas

The big one is documented in the README and worth repeating: don't start ComfyUI with --highvram. It pins models in VRAM and blocks the memory cleanup the loop relies on, which is exactly how you get an OOM three cycles in. Standard or --normalvram is the supported mode.

Install

Via ComfyUI Manager, search "comfyui-sequential-batcher". Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/Meisoftcoltd/comfyui-sequential-batcher

Restart ComfyUI. The pack pulls in torchaudio, opencv-python, and safetensors, and FFmpeg is mandatory for the analyzer nodes to pull audio. The install script tries sudo apt install ffmpeg on Linux; on Windows you add FFmpeg's bin folder to your PATH. A detail nobody warns you about: the pack's log output is in Spanish, because that's the author's language. Don't panic when 🚀 Ciclo 5: Solicitando 81 frames efectivos shows up in your console - it's working.