Google Image Loader

The name says "Image Loader," but don't let that fool you - searchType gives you a straight choice between web and image search, so this is really a general Google-search loader that happens to keep an old, narrower name from earlier in the pack's life. Worth knowing going in so you don't skip past it looking for a plain web-search node and miss that it's the same one.

It sits in this pack's knowbase category, which is the tell for what it's for: pulling fresh, live information into an LLM's context instead of relying only on what the model already knows. The README calls out exactly this use case - building RAG pipelines where an LLM's knowledge base gets supplemented with retrieved, up-to-date content rather than frozen training data.

How it works

You give it search terms, it queries Google's Custom Search API, and it hands back the results as plain text - ready to drop into an LLM's context alongside the user's question, the standard retrieval-augmented pattern.

The inputs and outputs that matter

• searchType - web or image. Pick web for text results, image for an image search.
• keywords - your search query.
• paper_num (default 1) - how many results to pull back. Higher isn't automatically better here - more results means more tokens shoved into your LLM's context, so scale it to how much the model can actually use.
• google_api_key and google_CSE_ID (optional, but required in practice) - credentials for Google's Custom Search API. The README's own config section names this exact pair: google_api_key and cse_id, set either directly on the node or once in config.ini.
• is_enable - standard bypass toggle.

Output: text (STRING) - the retrieved results, ready to wire into an LLM node's context.

How to install it

Via ComfyUI Manager: search comfyui_LLM_party, install, restart. Manually:

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

Then pip install -r requirements.txt from inside the pack folder using ComfyUI's own Python, and restart. Beyond the base install, you need two things from Google's own console: a Custom Search JSON API key, and a Programmable Search Engine (CSE) ID - both are free to create at Google's developer console and programmablesearchengine.google.com respectively. Fill them into config.ini once and every node in the pack that needs them (this one, plus any Bing/search-adjacent tools) can just pick them up.

Common issues & troubleshooting

Requests fail or return nothing. Missing or wrong google_api_key/google_CSE_ID is the usual cause - double-check both are actually filled in, either on the node or in config.ini.

Works fine, then suddenly stops. Google's free Custom Search tier has a small daily query cap (historically around 100 free queries a day) - if you're calling this inside a loop or an agent that searches often, you can burn through it faster than expected. Google bills per query past that.

Results feel irrelevant to the question. keywords is a literal search query, not a natural-language question - if you're feeding it raw user text, consider having an LLM step distill that into good search terms first, rather than passing the user's exact phrasing straight through.