Nodes/ComfyUI-VisionPromptAssistant/H3 Ollama Model (Local)
ComfyUI Node

H3 Ollama Model (Local)

Connects a local Ollama vision model to H3 Story Director — LLM Model (API). It uses Ollama's native multimodal and JSON Schema API, offers optional internal thinking, and can unload the model from memory immediately after every plan.

By elgalardi·Created 21 days ago·Updated a day ago· 1
H3 Ollama Model (Local)
    • llm_model
    server_urlhttp://127.0.0.1:11434
    modelhuihui_ai/gemma-4-abliterated:12b
    api_key
    keep_alivefalse
    thinkingfalse
    context_length32768
    timeout_seconds600
    Categorytext/minimax_h3

    Inputs (7)

    NameTypeDefaultDescription
    server_urlSTRINGhttp://127.0.0.1:11434Paste a local, LAN, remote, or hosted Ollama-compatible address. Accepts a server root, an address ending in /api, or the complete /api/chat endpoint.
    modelSTRINGhuihui_ai/gemma-4-abliterated:12b
    api_keySTRINGOptional Bearer token for protected remote or hosted Ollama-compatible services. Local Ollama needs no key.
    keep_aliveBOOLEANfalseFalse unloads the model after every plan and frees VRAM for ComfyUI. True keeps it loaded for repeated planning.
    thinkingBOOLEANfalseAllows supported Ollama models to reason internally before returning the final structured plan. This may improve difficult scenes but increases generation time and token use.
    context_lengthINT327688192–65536Includes the system prompt, JSON Schema, visual tokens, and generated plan. Larger values consume more VRAM.
    timeout_secondsINT60060–3600

    Outputs (1)

    NameTypeDescription
    llm_modelLLMMODEL