ComfyUI Node: Multimodal — LLM (local first: GGUF + optional transformers)
Run ComfyUI workflows without the setup
No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.
Category
llm/multimodal
Inputs
use_vision BOOLEAN
load_source
- auto
- local
- huggingface
gguf_main_models_llm
- — use path fields below —
gguf_mmproj_models_llm
- — use path fields below —
gguf_model_path STRING
gguf_mmproj_path STRING
gguf_vlm_handler
- qwen3-vl
- qwen2.5-vl
- gemma3
- llava-1.5
- llava-1.6
gguf_allow_qwen25_if_qwen3_handler_missing BOOLEAN
gguf_n_ctx INT
gguf_streaming BOOLEAN
gguf_chat_template_enable_thinking BOOLEAN
include_thinking_in_output BOOLEAN
thinking_open_tag STRING
thinking_close_tag STRING
combined_model STRING
system_prompt STRING
extra_context STRING
prompt STRING
max_new_tokens INT
do_sample BOOLEAN
generation_seed_mode
- fixed
- increment_each_run
generation_seed INT
temperature FLOAT
top_p FLOAT
top_k INT
repetition_penalty FLOAT
use_repetition_penalty BOOLEAN
device
- auto
- cuda
- cpu
dtype
- auto
- bfloat16
- float16
- float32
- False
- True
trust_remote_code BOOLEAN
keep_models_loaded BOOLEAN
offload_image_mode BOOLEAN
log_load_details BOOLEAN
image IMAGE
image2 IMAGE
image3 IMAGE
Outputs
STRING
Extension: Multimodal — LLM (Local-first: GGUF + Transformers)
Local-first multimodal LLM nodes for ComfyUI (GGUF+mmproj via llama.cpp, optional Transformers fallback).
Authored by theonetwoone
Looking for a different node?
More nodes in Multimodal — LLM (Local-first: GGUF + Transformers)
- Bonus: LLM text → image intent (KSampler prompts)
- Multimodal — Context schema builder (template → prompt)
- Multimodal — Context schema parser (<context>…</context>)
- Multimodal — GGUF settings sorter (auto defaults)
- Multimodal — Context handler
- Multimodal — Context handler (debug outputs)
- Multimodal — CUDA / Python report
- Multimodal — llama-cpp wheel pick (GitHub)
- Multimodal — System prompter
- Multimodal — Unload in-ComfyUI LLM / VLM cache
Run ComfyUI workflows without the setup
No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.