Kinburg Live Log 📜

Running a local LLM inside ComfyUI has one brutal UX problem: it's a black box for the length of the generation. You queue, you wait, and if you want to see anything, you stare at the console. Kinburg Live Log fixes that properly - it's a live view of every LLM node in the pack as it generates, streaming text token by token into a panel that feels like a chat client, not a debug log.

The mechanism is the good kind of magic: it needs no wiring. Turn on the live_preview toggle on any supported source node - Local LLM (GGUF), Local LLM Chat, Morpheus Storyboard, or the Morpheus sampler's in-loop writer - drop a Kinburg Live Log node anywhere on the canvas, and the stream appears. One log shows every LLM node in the pack at once, each generation in its own block labelled by its source (and by the call, when the source labels them - "Morpheus Storyboard · shot 2/4"). It's a pack-wide event bus with a nice UI on top.

What it shows is genuinely diagnostic rather than decorative. Each block header counts tokens against the run's max_tokens ceiling (142/512 tok) with a live tok/s reading and a thin budget bar; on finish it adds the context fill (ctx 1780/4096 (43%)), going amber when the output hit the cap or the context is nearly full. Reasoning gets its own treatment: <think> blocks (or everything before your answer_marker line) render dim and italic in a collapsible section headed thinking… 4.2s, folding away the moment the answer starts. Vision calls show the frames they were actually given, and Morpheus posts each shot's last frame as it decodes, so the log doubles as a live storyboard of a render. There's a copy button per block (copies the answer, without the reasoning), thumbnails copy to clipboard on hover, and the view stays put when you scroll up while generation continues - a ↓ latest pill jumps you back.

No inputs, no outputs. It's purely observational, which is exactly what a log should be - nothing about your graph changes because the log exists.

Installing

Ships in Kinburg-Nodes: ComfyUI Manager → search "Kinburg-Nodes", or clone into custom_nodes and restart. It's one of the pack's browser-side web/*.js extensions, so it loads with the frontend. To see anything, you need the pack's LLM nodes working, which means llama-cpp-python (installed automatically by the pack's install.py, which Manager runs) and a GGUF model configured.

Gotchas

If the log stays empty, the live_preview toggle on the source node is off - the log only shows what sources stream to it. And a subtlety worth knowing: this is the pack's general LLM log; the Ouroboros loop keeps its own Ouroboros Live Log 🐍📜 with an iteration/score/stage layout, so don't expect self-correcting-loop telemetry to land here. For everything else - chat, storyboarding, vision calls - this is the one.