ToriiGate Llama.cpp Vision Generate

A ComfyUI node in ToriiGate/API with 11 inputs and 1 output.

By litch230·Created 3 months ago·Updated 3 months ago· 20

◄server_urlhttp://127.0.0.1:8080►

◄model_nameDraconicDragon/ToriiGate-0.5-GGUF:Q4_K_M►

◄timeout120►

◄custom_model_name►

◄prompt►

◄max_pixels_mp1.0►

◄max_new_tokens512►

◄temperature0.50►

◄decodingsample►

◄seed0►

CategoryToriiGate/API

Inputs (11)

Name	Type	Default	Description
image	IMAGE		ComfyUI image tensor (B, H, W, C float32). Only the first image in the batch is sent.
server_url	STRING	http://127.0.0.1:8080	Base URL of the llama-server instance. Example: http://127.0.0.1:8080
model_name	COMBO	DraconicDragon/ToriiGate-0.5-GGUF:Q4_K_M	GGUF quantization to use. The identifier must match what the llama-server router registered (shown at startup as 'operator(): ...'). Q4_K_M is the recommended balance of quality vs size (3.07 GB). Use custom_model_name below to override with any arbitrary string.
timeout	FLOAT	1205–600	HTTP request timeout in seconds. Increase for slow hardware or very long generations.
custom_model_nameopt	STRING		Override the model identifier with any custom string. Useful when running a non-GGUF backend or a locally renamed model. Leave blank to use the dropdown above.
promptopt	STRING		Optional prompt. You can connect the text output from the ToriiGate Grounding Builder here, or type your own.
max_pixels_mpopt	FLOAT	1.00.1–8	Resolution limit sent to the model, in megapixels. Lower values drastically reduce prompt evaluation time (Time To First Token) in llama.cpp.
max_new_tokensopt	INT	51264–8192	Maximum generated tokens.
temperatureopt	FLOAT	0.500–2	Generation randomness. 0 is deterministic.
decodingopt	COMBO	sample	sample uses temperature-based sampling; greedy_fast sets temperature to 0.0.
seedopt	INT	00–18446744073709550000	Seed for reproducibility. Use 0 for a random seed.

Name	Type	Description
caption	STRING	—