ComfyUI Node

CC Llama Vision

A ComfyUI node in llama.cpp with 14 inputs and 1 output.

By cicalooo·Created 2 months ago·Updated about a month ago· 1
CC Llama Vision
  • image
  • image_batch
  • video_frames
  • advanced_options
  • caption
llama_server_pathllama-server
models_dir/root/models
model_path
mmproj_path
system_prompt
user_prompt
port8080
max_tokens2048
max_video_frames16
label_video_framestrue
Categoryllama.cpp

Inputs (14)

NameTypeDefaultDescription
llama_server_pathSTRINGllama-serverPath to the llama-server executable. Leave as 'llama-server' to resolve automatically from PATH (recommended after `winget install llama.cpp`), or provide a full path to llama-server.exe.
models_dirSTRING/root/modelsFolder to scan (recursively) for .gguf model and mmproj files. Click '🔄 Refresh Models' after changing this to repopulate the dropdowns below.
model_pathCOMBOThe GGUF vision-language model file to load (mmproj files are excluded from this list).
mmproj_pathCOMBOThe GGUF multimodal projector (mmproj) file matching the selected model — required for image/vision input.
system_promptSTRINGSystem-level instructions sent to the model before the user prompt, e.g. persona or output-format guidance.
user_promptSTRINGThe instruction/question sent to the model along with the image(s). This text is appended after any images.
portINT80801024–65535Local TCP port llama-server will listen on. Must be free unless 'keep_server_alive' is reusing an existing server on this port.
max_tokensINT204816–8192Maximum number of tokens the model may generate in its response.
max_video_framesINT161–64Maximum number of frames to sample from 'video_frames' input; frames beyond this are evenly subsampled.
label_video_framesBOOLEANtrueIf enabled, prefixes each video frame image with a text label like '[Video frame N of M]'.
imageoptIMAGEA single image to send to the model.
image_batchoptIMAGEA batch of images; each is sent as a separate image in the same message.
video_framesoptIMAGEA sequence of image frames (e.g. from a video) to sample and send to the model, subject to max_video_frames.
advanced_optionsoptLLAMA_VISION_OPTSOptional settings from a 'CC Llama Vision Advanced Options' node — sampling, performance, lifecycle, and diagnostics. If not connected, sensible defaults are used.

Outputs (1)

NameTypeDescription
captionSTRING