url2img tool

Same capability as this pack's URL2IMG node - fetch an image from a URL - packaged completely differently. URL2IMG is a node you wire by hand with a fixed link you already know. url2img_tool instead wraps that same fetch-an-image ability as a callable tool the LLM decides to use on its own, mid-conversation, whenever it judges it needs to: it found a link in a search result, a user pasted one, whatever the reasoning is. That's the general pattern across every tool-suffixed node in this pack - the graph doesn't run the action itself, it hands the ability to the model.

It's about as minimal as this pack's tool nodes get. There's no URL field on the node at all - nothing to configure beyond whether it's active. The URL comes from the agent's own reasoning at call time, not from anything you set up in the graph.

Inputs and outputs that matter

• is_enable - the only input, the pack-wide skip switch.

The single output, tool, is the packaged capability - wire it into tool_combine or tool_combine_plus alongside your agent's other tools, not into anything expecting an actual image.

Installing it

Search comfyui_LLM_party in ComfyUI Manager and install, or clone directly:

cd ComfyUI/custom_nodes
git clone https://github.com/heshengtao/comfyui_LLM_party

pip install -r requirements.txt inside your ComfyUI Python environment, then restart. This node itself has no configuration-time dependency beyond what the underlying fetch needs at call time (outbound network access), but it's bundled inside the full pack, which has a much larger overall install. The README's only_api branch is the leaner option if you only want API-based LLM calling.

Common issues

Because there's nothing to configure on the node, most problems with url2img_tool actually show up at call time rather than setup time - the same constraints as URL2IMG apply, they're just invisible until the agent decides to use it. If the model tries to fetch an image and the environment has no outbound internet access, or the link it picked isn't a direct image URL, the tool call will fail. Since you won't see that until it happens mid-conversation, it's worth testing the underlying fetch path (with URL2IMG directly, using a known-good link) before trusting this tool inside a live agent.

The other thing worth being deliberate about: giving an agent the ability to fetch arbitrary URLs is giving it a small amount of outbound network reach it decides how to use. That's generally fine for images from links a user or search result surfaced, but it's worth knowing this tool exists on your agent's toolbox rather than being surprised later that it can reach outside the graph on its own.