Settings Launcher (Mobile)

If you've ever tried to drive ComfyUI from a phone browser, you know the pain: the graph is cramped and the KSampler's widgets are tiny targets. Settings Launcher (Mobile) is this pack's answer - a single node that holds every sampling knob in one place and, when you want it to, pauses the run so you can check what's about to happen before committing.

Two jobs in one

First, it's a settings hub. Instead of reaching into the KSampler and EmptyLatentImage, you set seed, steps, cfg, width, height, batch_count, sampler_name, scheduler and denoise here, once, and wire them over. The outputs are typed to match exactly what those nodes accept - sampler_name comes out as a real ComfyUI sampler enum rather than a string, so it plugs straight into a KSampler input with no conversion node in between.

Second, it's a flow-control gate. The only required input is resume, and it defaults to "Wait to Proceed". In that mode, when the queue reaches this node, the backend blocks until you tap the Proceed button the pack's extension adds to the node. Cancel kills the run (the code raises an interrupt exception). Flip resume to "Always Proceed" and it becomes a straight passthrough of the values.

How the pause works

The mechanism is worth knowing because it shapes how you use it. With "Wait to Proceed", the node's Python thread sits in a busy-wait loop, sleeping a fraction of a second at a time, until the frontend JavaScript POSTs to the pack's /settings_launcher_flow_control_proceed endpoint. The node's IS_CHANGED is randomized, so it re-evaluates on every run. The practical consequence: the values it outputs are whatever the widgets hold when execution reaches it. Think of the pause as a review-and-confirm gate - edit the knobs between runs, not while it's frozen mid-execution.

What a beginner actually sets

• resume: leave it at "Wait to Proceed" for the confirmation gate, or switch to "Always Proceed" to use the node purely as a settings source.
• sampler_name / scheduler: the full stock ComfyUI lists. If you don't know your preference, DPM++ 2M with the Karras scheduler is the safe default for SD 1.5 and SDXL.
• steps / cfg / denoise: the usual suspects. Denoise is the img2img lever - 1.0 is full noise, lower values reuse more of the source image.

Install

Standard custom-node install - ComfyUI Manager, search "ComfyUI-Mobile", or:

cd ComfyUI/custom_nodes
git clone https://github.com/SozeInc/ComfyUI-Mobile

Then restart. The only Python requirement is requests. No models to download.

Honest caveats

This is from the same early-stage pack whose README says it's "not functional as a stand alone install" - the nodes load and work, but there's no polish. And don't confuse it with the popular mobile frontend projects (like comfyui-mobile-frontend) that let you edit the whole graph from a phone. This node doesn't make ComfyUI usable on mobile by itself; it's the settings hub and confirmation gate you'd pair with a mobile-friendly way to run and check your queue.