wanx-troopers

LLM Nodes

Various nodes interacting with or running LLM models are used by AI artists to enhance prompts during image and video generation. This page collections information on some of them.

• prompt-enhancement-qwenvl-advanced - example of how to enhance prompt for LTX using QwenVL, using LTX 2.3 system prompt; same using GH:richservo/rs-nodes: prompt-enhancement-qwenvl-advanced-rs-nodes
• GH:vrgamegirl19/comfyui-vrgamedevgirl now contains a Gemma 4 node, workflow in GH:vrgamegirl19/comfyui-vrgamedevgirl:Workflows/betatesting
• GH:WASasquatch/ComfyUI_LMStudio_EasyQuery MIT licensed, reasonably up to date as of March 2026
• GH:BennyDaBall930/ComfyUI-Z-Engineer seems to be from the same source as the “Z-Enginner” flavor of Qwen3, popular with ZIT users
• GH:WASasquatch/WAS_Extras “I like the WAS LM Studio nodes, work great”
• GH:SeargeDP/ComfyUI_Searge_LLM node to enhance text to image prompts via LLM-s
• GH:mattjohnpowell/comfyui-lmstudio-image-to-text-node node pack to generate advanced prompts using LLM-s
• GH:fblissjr/ComfyUI-QwenImageWanBridge LLM tool to run z-image and other image generation models
• GH:dagthomas/comfyui_dagthomas
• GH:gabe-init/ComfyUI-Openrouter_node

btw, I fired Josie as my prompt editor, and hired Mistral 7b Q4 running on ComfyUI_Searge_LLM

Apparently unlike LMStudio, VLLM can do parallel queries.

Notes On LLM Models

• gemma-4-26b-a4b-it-heretic.q4_k_m.gguf
• supergemma4-26b-uncensored-fast-v2-Q4_K_M.gguf
• gemma-4-26B-A4B-it-UD-IQ2_M.gguf
• gemma-4-...-mmproj.. to enable Gemma to “see” images

Stef:

I used both Qwen3.6 and Gemma4. Both are equally good at describing the reference images, I had the impression Gemma4 folllows specific instructions a bit better (as “mention the reference images always as ‘from image0’, ‘from image1’ etc. Qwen 3.6 sometimes forgot that)

This site is open source. Improve this page.