Superside Any LLM Text

Want a real chat LLM in your graph but don't want to budget VRAM for one next to your diffusion model? This node is the shortcut: it routes your prompt to OpenRouter through fal.ai, so the model runs on someone else's GPU and the only thing your machine does is wait for the answer to come back. Type "Superside" in the node search and you'll find it under the Superside category.

It's a plain text-only chat node. You give it a prompt, it returns text. The model dropdown lists 16 options covering the usual suspects - Gemini 2.5 Flash/Pro and Lite, Claude 4.6 Sonnet/Opus, GPT-4o and GPT-5 chat, Llama 4, Grok 4 Fast, even Kimi - so you're not locked to one vendor. system_prompt sets the persona or instruction context, temperature goes 0–2 (lower = more predictable, and honestly for prompt-rewriting you usually want it on the low side), and max_tokens caps the reply. Flip reasoning on and, for models that support it, you get the model's chain-of-thought on the second output instead of just the answer.

The two outputs are output (the text) and reasoning (the scratch thinking, when enabled). Both are STRING, so wire output into anything that eats text - a prompt combine node, a text preview, a Save Text. That's the whole point of a text LLM in a graph: it's upstream of your sampler, writing the prompt or dialogue that generation then uses.

The part to internalize: this is an API node, so the economics and privacy rules are different from local. Every call costs fal credits, and fal locks you out the moment your balance hits zero - there's no free tier to fall back on. Your prompt also leaves the machine. If you're only enhancing prompts or captioning, the KB's verdict applies here too: a local 8B model is free, offline, and uncensored, and for short structured rewriting it's usually plenty. You reach for this node when you want genuine frontier-model quality - long-form reasoning, a better writer than a small Llama - and you're fine with metered, non-private calls.

The security frame matters because the "LLM node with your key" shape has literally been weaponized once in this ecosystem (ComfyUI_LLMVISION shipped credential-stealing malware). This pack is open source and its whole job is wrapping fal endpoints - there's no config file, no hidden model downloads, and the key is never stored in the repo. Still, install it from the official GitHub, and know that the key you paste into the widget is what gets sent with each call.

Install - ComfyUI Manager (search "comfyui-superside-nodes"), or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/Superside/comfyui-superside-nodes
pip install -r requirements.txt

Then restart ComfyUI. The requirements.txt is just fal-client, pillow, numpy, torch, requests - no model files to download. Paste your fal API key into the api_key widget on the node. If you leave it blank the node falls back to the FAL_KEY environment variable (handy for headless pipelines), and if both are empty it fails immediately with a clear error.

One more thing worth knowing: the pack ships from a design agency (Superside) and has basically zero community footprint - don't go hunting for tutorial threads, there aren't any. The README is thorough and current, which is more than you get from most packs with 100x the installs.