Comflowy LLM

The Comflowy LLM node is the pack's answer to the "prompt generator" genre, and its pitch is refreshingly concrete: it calls a hosted LLM through an API, so you don't install Ollama, you don't download a 4GB model, and you don't need a GPU that can hold one. The README says it plainly - "you don't need to install Ollama to call LLM models. No need to worry about whether your computer configuration is sufficient." And the kicker: it's free. Comflowy runs the inference on SiliconFlow's infrastructure and eats the cost, at least for now.

What is it actually good for? Prompt remixing, mostly. The 2026 reality is that your prompt is fed to an LLM anyway on the modern model families, and this node gives you a lightweight LLM inside the graph to do the "expand this into a detailed scene description" or "translate this idea into a rich prompt" job before it reaches your sampler. It's not a frontier model - think of it as a hosted mini-LLM for pipeline glue, not a reasoning engine.

How it works

Under the hood it's a single requests.post to https://app.comflowy.com/api/open/v0/prompt with your prompt, system prompt, chosen model, and max_tokens=4000, authenticated with your Comflowy API key. The text that comes back is returned as a STRING you can pipe into any prompt input in ComfyUI. Simple, and it's the same endpoint every LLM-ish node in the pack (including the Omost one) uses.

Inputs and output

• prompt (required, multiline) - your actual request.
• system_prompt (required, multiline) - the role framing. Leave empty if you want, but a line like "You are a prompt engineer who writes detailed Stable Diffusion prompts" changes the output character a lot.
• llm_model (required) - seven hosted options: Qwen/Qwen2-7B-Instruct, Qwen/Qwen2-1.5B-Instruct, THUDM/glm-4-9b-chat, THUDM/chatglm3-6b, 01-ai/Yi-1.5-9B-Chat-16K, 01-ai/Yi-1.5-6B-Chat, internlm/internlm2_5-7b-chat. The 7B models give the best quality-to-latency balance; the 1.5B is for when you want it snappy.
• seed - 0 to a 64-bit max; the node truncates oversized seeds to 32 bits internally, so don't stress about the exact value.

Output: a single STRING. This node is marked as an output node, so it renders its result as a text box in the UI - handy for checking what the LLM said before you wire it onward.

Installing and the gotchas

Install the pack via ComfyUI Manager (search "Comflowy") or git clone https://github.com/6174/comflowy-nodes into custom_nodes, restart, and set your key with the Comflowy Set API Key node. One dependency: requests.

The one failure mode you'll actually hit is spelled out in the README itself: network problems. If you see Failed to get response from LLM model with https://app.comflowy.com/api/open/v0/prompt, that's the API call failing - check your connection, proxy, and whether the API host is reachable from your region. The key being unset is the other classic: you'll get "API Key is not set. Please use the 'Comflowy Set API Key' node first."

Also worth knowing: the node never crashes the graph on an error - it returns the error string as its output. So if your prompt suddenly contains the text of a Python exception, the LLM call failed and you need to read the console, not the prompt box.

One honest caveat: "free" here means free-at-the-moment, backed by a commercial partner. If this becomes your workflow's backbone, keep a local Ollama setup as the fallback - but for a first project it's a genuinely zero-setup way to add an LLM to your graph.