sd1 / sdcpp / createImage

SD 1.5 is the old warhorse that refuses to die - it still has more full checkpoints on Civitai than any other base (over 10,000), and it's still the cheapest, most forgiving draft model around. This node lets you run any of them without downloading a single safetensors file. Pick a checkpoint via a Model Selector, type a prompt, and Civitai's sdcpp cloud engine renders it on their fleet, billed in Buzz. If you've ever thought "I'd try that weird 1.5 merge but I'm not downloading another 2GB file," this is the node for you.

The honest framing: 1.5's native 512×512 resolution and its chaos is a feature for ideation - people still use it as a prompt-rebellion sketch stage before re-rendering on a modern model. On the cloud, that same "spit out a dozen weird drafts" workflow costs you per image rather than per watt, so it makes sense for occasional experimentation, not bulk. A local card makes 1.5 near-instant and free; this node is for the times you're nowhere near one.

How it works

A thin client over Civitai's Orchestration API. It builds the payload (engine: sdcpp, ecosystem: sd1, operation: createImage), submits to orchestration.civitai.com, long-polls (30-minute default timeout), and returns the image as a ComfyUI IMAGE tensor. The node displays the workflow id and Buzz cost when it finishes.

The inputs that matter

• prompt (required, multiline) - and 1.5's 77-token CLIP limit applies on the cloud just like it does locally. Long prose gets truncated; keep it tag-style.
• model (required, CIVITAI_AIR) - wire a Civitai Model Selector and browse any SD 1.5 checkpoint.
• negative_prompt - 1.5 leans on this hard; a good negative is half the prompt.
• width / height - default 512×512, its native resolution. Push to 768 or 1024 without an upscaler and anatomy starts melting, same as local.
• steps / cfg_scale - 20 / 7 defaults are the classic 1.5 values; 1.5 likes CFG around 6–8.
• clip_skip - default 2, which matches what most good 1.5 checkpoints are tuned around. Leave it unless the checkpoint page says otherwise.
• embeddings - the one sdcpp input you won't see on the SDXL node in quite the same way: wire a Civitai Embedding Selector. Embeddings apply automatically to the positive prompt; to put one on the negative side, reference it by name in negative_prompt instead.
• u_cache - off/normal, Civitai's UNet cache. normal speeds up jobs you're paying for.
• vae_model, loras - via the selector loaders.

Install

Pack install, same as every node here: ComfyUI Manager → search "Civitai Comfy Nodes" (publisher civitai) → Install → restart. Or git clone https://github.com/civitai/civitai-comfy-nodes.git into custom_nodes, pip install -r civitai-comfy-nodes/requirements.txt (just requests), restart.

Common issues

Missing credentials is the usual wall: wire a Civitai Auth node into api_config, or set CIVITAI_API_TOKEN (from civitai.com/user/account) - required on headless boxes where browser OAuth can't run.

Second is resolution expectations. People feed it a 1024×1024 request out of habit and get soft, doubled faces - that's the model, not the node. Render at 512–640 and upscale downstream if you need more.

And remember the billing reality: this node has a quantity input that goes to 12, and every one of those images is a separate billable job. One click at quantity 12 is twelve Buzz charges. Keep it at 1 while you're tuning.