Create your own custom LLM
The article explains how to “create your own custom LLM” by tweaking an existing model in Ollama rather than training a new one. It uses a ComfyUI workflow that needs detailed photographic prompts, so it shows how to define a Modelfile (based on gemma3:4b
) that sets temperature, context size, and a system prompt that expands simple scene descriptions into full camera‑style prompts. With Docker and Open WebUI (which acts like a local ChatGPT), you can run the modified model and then generate tailored prompts for ComfyUI. The result is a lightweight, reusable setup that delivers consistent, high‑quality prompts for realistic image generation without heavy fine‑tuning.