Cloud KSampler (Quick)

This is the node that makes the whole pack click for most people: Cloud KSampler, displayed as "Cloud KSampler (Quick)", is a monolithic text-to-image / image-to-image node. Pick a checkpoint, type your prompts straight into the widget, set size, seed, steps, cfg, sampler and scheduler - then hit run and get an actual IMAGE tensor back. No checkpoint loader, no separate CLIP encode, no VAE decode, no terminal node to wire. It builds the entire checkpoint→encode→sample→decode→save workflow internally, submits it to Comfy Cloud, polls it, and downloads the result. It is the closest thing this pack has to a one-click cloud generator.

The pack's own README frames it as the all-in-one for people who just want a single-node txt2img/img2img call, and points out the trade: if you need LoRAs, separate CLIP loaders, two-stage sampling, or any real graph structure, drop down to the graph-style nodes (Cloud Checkpoint Loader, Cloud CLIP Text Encode, Cloud Empty Latent, Cloud KSampler Graph, Cloud VAE Decode). The Quick node gives you convenience and the graph nodes give you power; they're the same engine underneath.

The inputs that matter

• ckpt_name - the checkpoint dropdown. Same cache gotcha as everywhere: if it shows the placeholder, run Cloud List Models (folder=checkpoints) and restart ComfyUI.
• positive / negative - multiline prompt strings, typed right here. No conditioning nodes involved.
• width / height / batch_size - defaults 1024², batch 1; multiples of 8.
• seed / steps / cfg / sampler_name / scheduler / denoise - the usual sampler controls. 22 samplers and 7 schedulers are offered, the same familiar list (euler, dpmpp_2m, karras, and friends).
• image (optional) - a local IMAGE tensor. Wire one in and the node switches to img2img: it uploads your image to the cloud (/api/upload/image), builds a LoadImage → VAEEncode → sample → decode chain around it, and uses denoise to control how much of your image survives.

Outputs: images (IMAGE) - the stacked result batch; prompt_id (STRING) - the cloud job ID.

How it works

It's a terminal node with its own internal builder (build_text2img / build_img2img in the pack's source). For img2img it literally uploads your local image to the cloud, which is why the optional image input is a real tensor rather than a cloud handle. The default timeout here is 600s (10 min), shorter than the fetch nodes' 1800s - raise it for slow models.

How to install

cd ComfyUI/custom_nodes
git clone https://github.com/Dobidop/ComfyUI-CloudAPI-worker

Copy config.json.example to config.json, paste an API key from https://platform.comfy.org/profile/api-keys, restart. ComfyUI Manager works too. Dependencies: requests, Pillow, safetensors.

Common issues

• "No checkpoint selected" - run Cloud List Models and restart; the dropdown needs the cache.
• img2img ignores your image - make sure the image input is actually connected; a disconnected optional input silently means txt2img.
• Timeouts - default 600s burns fast on video-adjacent workloads; raise timeout.
• Cost - every run is a fresh cloud submission on metered GPU time. It's the convenience node, so it's also the one that makes iteration expensive. Iterate cheap on quick models, spend on the jobs that matter.