Ollama Option — Temperature

Temperature is the single knob most people reach for when a model feels "off" - too robotic at one extreme, too unhinged at the other. Ollama Option - Temperature is how you grab it in this pack, and honestly it's the option node you'll use the most, because it controls the personality of every answer your chat node produces.

Like the rest of the option nodes, it's one value in, one merged OLLAMA_OPTIONS dict out. It sets Ollama's temperature, which is how freely the model samples: low values keep it hugging the most-likely next token, high values let it wander. The mechanism underneath is bog-standard LLM sampling - this node just translates your slider into the parameter the local model actually consumes.

The inputs and outputs

• temperature (required) - a FLOAT from 0 to 2, default 0.8, stepping by 0.01. The map: 0 is near-deterministic (the model will still wander a little, but it's as "same answer every time" as local models get), 0.8 is the balanced default, 1.5+ is where it starts getting weird and creative, and 2.0 is the maximum chaos allowed.
• options (optional input) - the merged dict from the previous option node, so temperature stacks with seed, max tokens, and the rest of the chain.
• options (output) - the merged dict, wired into Chat Completion.

Installing it

Part of the ComfyUI Ollama Model Manager pack: Manager → search "Ollama Manager" → Install → restart, or clone https://github.com/darth-veitcher/comfyui-ollama-model-manager into ComfyUI/custom_nodes and run python install.py. Python 3.12+, deps auto-installed.

Where people get burned

The classic mistake is cranking temperature up to "fix" bland output. It'll make the output more varied, sure - and less coherent. For prompt-enrichment pipelines where the LLM feeds the image prompt, people get better results at 0.7–0.8 than at 1.5, because you want a focused rewrite, not a fever dream. And if you're chasing reproducibility, remember temperature works with the Seed option: a seed pins the random state, so the same seed + same temperature + same prompt gives you the same answer. No seed, and the chat node re-rolls every run regardless of what you set here.