Nodes/ComfyUI-EACloudNodes/Groq Chat
ComfyUI Node

Groq Chat

Interact with Groq's API for ultra-fast inference. Model list dynamically fetched from Groq API (5-min cache). Supports text generation, JSON output, and vision analysis with compatible models.

By EnragedAntelope·Created 2 years ago·Updated 3 months ago· 9
Groq Chat
  • image_input
  • response
  • status
  • help
◄api_key►
◄modelllama-3.3-70b-versatile►
◄manual_model►
◄system_promptYou are a helpful AI assistant. Please provide clear, accurate, and ethical responses.►
◄user_prompt►
◄send_systemyes►
◄temperature0.70►
◄top_p0.70►
◄max_completion_tokens1000►
◄frequency_penalty0.00►
◄presence_penalty0.00►
◄response_formattext►
◄seed_moderandom►
◄seed_value0►
◄max_retries3►
◄debug_modeoff►
◄additional_params►
CategoryGroq

Inputs (18)

NameTypeDefaultDescription
api_keySTRING⚠️ Your Groq API key from https://console.groq.com/keys (Note: key will be visible - take care when sharing workflows)
modelCOMBOllama-3.3-70b-versatileSelect a Groq model or choose 'Manual Input'. Categories: Featured, Production (stable), Preview (evaluation). Use ComfyUI Refresh to update model list from Groq API.
manual_modelSTRINGEnter a custom model identifier (only used when 'Manual Input' is selected above). Leave empty if using dropdown selection.
system_promptSTRINGYou are a helpful AI assistant. Please provide clear, accurate, and ethical responses.Optional system prompt to set the AI's behavior and context. Note: Vision models may not support system prompts - toggle 'send_system' to 'no' if needed.
user_promptSTRINGMain prompt or question for the model. For vision tasks, describe what you want to know about the image.
send_systemCOMBOyesToggle system prompt sending. Set to 'no' for vision models that don't accept system prompts (e.g., Llama-4 vision models).
temperatureFLOAT0.700–2Controls response randomness and creativity. Lower values (0.0-0.3) = more focused and deterministic. Higher values (0.7-2.0) = more creative and varied.
top_pFLOAT0.700–1Nucleus sampling threshold. Controls diversity of word choices. Lower values (0.0-0.3) = more focused vocabulary. Higher values (0.7-1.0) = more diverse word selection.
max_completion_tokensINT10001–131072Maximum number of tokens to generate in the response. Note: actual limit varies by model (check model documentation). Range: 1-131,072.
frequency_penaltyFLOAT0.00-2–2Penalizes tokens based on their frequency in the output. Positive values reduce repetition. Range: -2.0 to 2.0. Note: not all models support this parameter.
presence_penaltyFLOAT0.00-2–2Penalizes tokens that have already appeared in the output. Positive values encourage topic diversity. Range: -2.0 to 2.0. Note: not all models support this parameter.
response_formatCOMBOtextResponse format: 'text' for natural language, 'json_object' for structured JSON output. When using JSON, instruct the model in your prompt to output JSON.
seed_modeCOMBOrandomSeed behavior control: 'fixed' uses the seed_value below, 'random' generates new seed each time, 'increment' increases by 1, 'decrement' decreases by 1.
seed_valueINT00–9007199254740991Seed value for reproducibility when seed_mode is 'fixed'. Use same seed + parameters for identical outputs. Valid range: 0-9007199254740991 (JavaScript safe integer limit).
max_retriesINT30–5Maximum number of automatic retry attempts for recoverable errors (rate limits, temporary server issues). 0 disables retries. Range: 0-5.
debug_modeCOMBOoffEnable detailed error messages and request debugging information. Useful for troubleshooting API issues or parameter problems.
image_inputoptIMAGEOptional image input for vision-capable models. Currently supported: meta-llama/llama-4-scout-17b-16e-instruct. Maximum size: 2048x2048.
additional_paramsoptSTRINGAdditional Groq API parameters in JSON format. Example: {"stop": ["\n"], "min_p": 0.1}. Use for advanced model-specific parameters not exposed in the UI.

Outputs (3)

NameTypeDescription
responseSTRING—
statusSTRING—
helpSTRING—