Search 🔍 (Pollinations)

Search 🔍 (Pollinations) looks like it should be the pack's utility belt, and it mostly is - it just isn't a search engine. It's the pack's Text Gen node pointed at a model that can actually browse the web. You type a query, and a hosted model goes out, reads what it finds, and comes back with a paragraph. Think of it as "ask a search-equipped assistant," not "run a query and get links."

How it works

Mechanically it's the same trick as Text Gen: a GET against text.pollinations.ai/{query} with model, seed, and private as URL params, and the raw text response handed back as a string. The difference is the model under the hood, which does realtime web search. Two choices in the dropdown:

• searchgpt - OpenAI's search-capable model with realtime news and web results. The default, and the one you'll use.
• elixposearch - the alternative search model. It's in the code and the dropdown but has next to no documentation anywhere, so treat it as "the other one."

No API key, no token - this is the free tier. Same deal as Text Gen: the service is free and keyless, and your generation stays private by default.

Inputs that matter

• query - your question, multiline. "What's the current stable version of..." works far better than a bare keyword; it's an LLM, talk to it like one.
• model - searchgpt unless you have a reason.
• seed - reproducibility, same as the rest of the pack.
• private (default on) - keeps the search out of Pollinations' public feed.

Output is a single search_results string. And like Text Gen, this is not an output node - wire the string into a ShowText node or it'll generate into the void and you'll wonder where your answer went.

Where it fits

Honestly? This is the pack's most awkward node, and it's worth saying plainly. A workflow that needs live facts - "what's the latest version of X," "what's today's news about Y" - can pull them in and hand them to Text Gen to summarize or reformat. That's a genuinely neat trick for automations. But if you wanted to search from a ComfyUI graph, you probably actually wanted a browser tab. It's a fun addition, not a replacement.

Install & issues

Installs with the pack (ComfyUI Manager, or clone + pip install -r requirements.txt - just requests and pydub). No models, no setup beyond restart.

• "Error: 5xx" in the output - the free API is a shared, no-keys service and it stumbles sometimes. Re-run.
• Stale answers - bump the seed. The node marks itself changed on every execution so it won't cache, but the answer quality varies run to run.
• Output is a wall of text - that's by design. It's prose, not structured results; you get one answer blob, not a ranked list of links.

If you squint, Search is really just the pack demonstrating that Text Gen with the right model does everything. Which is its own kind of advertisement for the Text Gen node - and a reminder that the whole pack is one free API wearing several hats.