nodes/slahiri-ComfyUI-AI-Photography-Toolkit
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
Files
dev/docs
slahiri-ComfyUI-AI-Photogra…/llm_providers/sid_llm_api.py
T
slahiriandClaude Opus 4.5 97002ca490 Disable dynamic Ollama/LM Studio model detection 
Removed automatic model querying for local providers.
Users should now use the custom_model field to manually
specify their model name. This avoids issues with:
- Slow startup when Ollama/LM Studio isn't running
- Listing non-vision models that won't work for image analysis
- Users can now specify exactly which model they want

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2025-12-11 15:55:19 +05:30

749 lines
32 KiB
Python
Raw Permalink Blame History

"""
SID_LLM_API Node - Unified Cloud LLM Provider
Single node for all cloud-based and local LLM providers.
Only vision-capable models are listed (required for image analysis).
Cloud Providers (API key required):
- Anthropic (Claude - all models support vision)
- OpenAI (GPT-4o, GPT-4.1 - vision models)
- Google Gemini (all models support vision)
- xAI Grok (grok-2-vision)
- Mistral AI (Pixtral vision models)
Free/Freemium Cloud Providers:
- Groq (free tier - Llama 4, Llama 3.2 vision)
- Together AI (free tier - Llama Vision, Qwen2.5-VL)
- OpenRouter (aggregator - free vision models)
- Fireworks AI (Llama 4, Llama 3.2 vision, Qwen VL)
- HuggingFace Inference (Llama 3.2 vision, Qwen VL)
Local Providers (no API key):
- Ollama (dynamic detection)
- LM Studio (dynamic detection)
- Custom OpenAI-compatible endpoints
For local vision models with VRAM management, use SID_LLM_Local instead.
"""
from typing import List, Dict, Any, Tuple
import requests
from comfy_api.latest import io as comfy_io
from .llm_model_type import LLMModelConfig
# Create custom LLM_MODEL type for ComfyUI
LLM_MODEL_Type = comfy_io.Custom("LLM_MODEL")
# =============================================================================
# Dynamic Model Detection
# =============================================================================
# Dynamic model detection disabled - users should use custom_model field
# def get_ollama_models(timeout: float = 2.0) -> List[str]:
# """Query Ollama API for installed models."""
# ...
# def get_lmstudio_models(timeout: float = 2.0) -> List[str]:
# """Query LM Studio API for loaded models."""
# ...
def is_likely_vision_model(model_name: str) -> bool:
"""
Check if a model name suggests vision capability.
Used to warn users when selecting non-vision models for image analysis.
"""
model_lower = model_name.lower()
vision_keywords = [
"vision", "vlm", "vl", "visual",
"llava", "bakllava", "pixtral", "qwenvl", "qwen-vl", "qwen2-vl", "qwen2.5-vl",
"moondream", "florence", "phi-3-vision", "phi3-vision", "smolvlm",
"llama-3.2-11b", "llama-3.2-90b", # Llama 3.2 vision models
"llama3.2-vision", "llama3.2:11b", "llama3.2:90b",
"llama-4", "llama4", # Llama 4 models are multimodal
"gpt-4o", "gpt-4-turbo", "gpt-4.1", # OpenAI vision models
"claude", # All Claude models support vision
"gemini", # Gemini supports vision
"grok-vision", "grok-2-vision", "grok-4", # Grok vision models
]
return any(kw in model_lower for kw in vision_keywords)
# =============================================================================
# Provider Configurations
# =============================================================================
PROVIDERS = {
# =========================================================================
# Major Cloud Providers (API key required)
# =========================================================================
"Anthropic": {
"api_url": "https://api.anthropic.com",
"api_key_url": "https://console.anthropic.com/",
"provider_name": "anthropic",
"requires_key": True,
"is_local": False,
"models": [
# All Claude models support vision
"claude-sonnet-4-5-20250929",
"claude-haiku-4-5-20251001",
"claude-opus-4-5-20251101",
"claude-sonnet-4-20250514",
"claude-3-5-sonnet-20241022",
],
"default_model": "claude-sonnet-4-5-20250929",
},
"OpenAI": {
"api_url": "https://api.openai.com/v1",
"api_key_url": "https://platform.openai.com/api-keys",
"provider_name": "openai",
"requires_key": True,
"is_local": False,
"models": [
# Vision-capable models only
"gpt-4.1",
"gpt-4.1-mini",
"gpt-4o",
"gpt-4o-mini",
],
"default_model": "gpt-4o",
},
"Google Gemini": {
"api_url": "https://generativelanguage.googleapis.com/v1beta",
"api_key_url": "https://aistudio.google.com/app/apikey",
"provider_name": "gemini",
"requires_key": True,
"is_local": False,
"models": [
# All Gemini models support vision
"gemini-2.5-flash",
"gemini-2.5-pro",
"gemini-2.0-flash",
],
"default_model": "gemini-2.5-flash",
},
"xAI Grok": {
"api_url": "https://api.x.ai/v1",
"api_key_url": "https://console.x.ai/",
"provider_name": "grok",
"requires_key": True,
"is_local": False,
"models": [
# Vision-capable models only
"grok-2-vision-1212",
],
"default_model": "grok-2-vision-1212",
},
"Mistral AI": {
"api_url": "https://api.mistral.ai/v1",
"api_key_url": "https://console.mistral.ai/api-keys/",
"provider_name": "mistral",
"requires_key": True,
"is_local": False,
"models": [
# Pixtral (vision models only)
"pixtral-large-latest",
"pixtral-12b-2409",
],
"default_model": "pixtral-large-latest",
},
# DeepSeek removed - no vision models available
# =========================================================================
# Free/Freemium Cloud Providers
# =========================================================================
"Groq (Free Tier)": {
"api_url": "https://api.groq.com/openai/v1",
"api_key_url": "https://console.groq.com/keys",
"provider_name": "groq",
"requires_key": True,
"is_local": False,
"models": [
# Vision-capable models only (Llama 4 multimodal)
# Note: Llama 3.2 vision preview models have been decommissioned
"meta-llama/llama-4-scout-17b-16e-instruct",
"meta-llama/llama-4-maverick-17b-128e-instruct",
],
"default_model": "meta-llama/llama-4-scout-17b-16e-instruct",
},
"Together AI (Free Tier)": {
"api_url": "https://api.together.xyz/v1",
"api_key_url": "https://api.together.xyz/settings/api-keys",
"provider_name": "together",
"requires_key": True,
"is_local": False,
"models": [
# Vision-capable models only
"meta-llama/Llama-Vision-Free",
"meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
"meta-llama/Llama-4-Scout-17B-16E-Instruct",
"meta-llama/Llama-3.2-11B-Vision-Instruct-Turbo",
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo",
"Qwen/Qwen2.5-VL-72B-Instruct",
],
"default_model": "meta-llama/Llama-Vision-Free",
},
"OpenRouter": {
"api_url": "https://openrouter.ai/api/v1",
"api_key_url": "https://openrouter.ai/keys",
"provider_name": "openrouter",
"requires_key": True,
"is_local": False,
"models": [
# Vision-capable free models (from /v1/models API Dec 2025)
"nvidia/nemotron-nano-12b-v2-vl:free",
"google/gemini-2.0-flash-exp:free",
"google/gemma-3-27b-it:free",
"google/gemma-3-12b-it:free",
"amazon/nova-2-lite-v1:free",
],
"default_model": "nvidia/nemotron-nano-12b-v2-vl:free",
},
"Fireworks AI": {
"api_url": "https://api.fireworks.ai/inference/v1",
"api_key_url": "https://fireworks.ai/api-keys",
"provider_name": "fireworks",
"requires_key": True,
"is_local": False,
"models": [
# Vision-capable models only
# Llama 4 multimodal (latest)
"accounts/fireworks/models/llama4-maverick-instruct-basic",
"accounts/fireworks/models/llama4-scout-instruct-basic",
# Llama 3.2 vision
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct",
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct",
# Qwen vision
"accounts/fireworks/models/qwen2p5-vl-32b-instruct",
# Phi vision
"accounts/fireworks/models/phi-3-vision-128k-instruct",
],
"default_model": "accounts/fireworks/models/llama4-maverick-instruct-basic",
},
# Cerebras removed - no vision models available (text-only: llama3.1-70b, llama3.1-8b, llama-3.3-70b)
"HuggingFace Inference": {
"api_url": "https://api-inference.huggingface.co/v1",
"api_key_url": "https://huggingface.co/settings/tokens",
"provider_name": "huggingface",
"requires_key": True,
"is_local": False,
"models": [
# Vision-capable models only
"meta-llama/Llama-3.2-11B-Vision-Instruct",
"meta-llama/Llama-3.2-90B-Vision-Instruct",
"Qwen/Qwen2-VL-7B-Instruct",
],
"default_model": "meta-llama/Llama-3.2-11B-Vision-Instruct",
},
# =========================================================================
# Local Providers (no API key required)
# =========================================================================
"Ollama (Local)": {
"api_url": "http://localhost:11434/v1",
"api_key_url": "",
"provider_name": "ollama",
"requires_key": False,
"is_local": True,
"models": "dynamic", # Auto-detected from Ollama API, use custom_model for manual entry
"fallback_models": [], # No fallback - use custom_model field instead
"default_model": "use custom_model field",
},
"LM Studio (Local)": {
"api_url": "http://localhost:1234/v1",
"api_key_url": "",
"provider_name": "lmstudio",
"requires_key": False,
"is_local": True,
"models": "dynamic", # Auto-detected from LM Studio API, use custom_model for manual entry
"fallback_models": [], # No fallback - use custom_model field instead
"default_model": "use custom_model field",
},
"Custom OpenAI-Compatible": {
"api_url": "http://localhost:8080/v1",
"api_key_url": "",
"provider_name": "openai_compatible",
"requires_key": False,
"is_local": True,
"models": [
"custom (use custom_model field)",
],
"default_model": "custom (use custom_model field)",
},
}
# Model metadata for all providers
MODEL_METADATA = {
# =========================================================================
# Anthropic models
# =========================================================================
"claude-sonnet-4-5-20250929": {"supports_reasoning": True, "max_output_tokens": 64000},
"claude-haiku-4-5-20251001": {"supports_reasoning": False, "max_output_tokens": 64000},
"claude-opus-4-1-20250805": {"supports_reasoning": True, "max_output_tokens": 64000},
"claude-3-5-sonnet-20241022": {"supports_reasoning": True, "max_output_tokens": 8192},
"claude-3-5-haiku-20241022": {"supports_reasoning": False, "max_output_tokens": 8192},
# =========================================================================
# OpenAI models
# =========================================================================
"gpt-4o": {"supports_reasoning": False, "max_output_tokens": 16384},
"gpt-4o-mini": {"supports_reasoning": False, "max_output_tokens": 16384},
"gpt-4.1": {"supports_reasoning": False, "max_output_tokens": 32768},
"gpt-4.1-mini": {"supports_reasoning": False, "max_output_tokens": 32768},
"gpt-4.1-nano": {"supports_reasoning": False, "max_output_tokens": 16384},
"gpt-4-turbo": {"supports_reasoning": False, "max_output_tokens": 4096},
"o1": {"supports_reasoning": True, "max_output_tokens": 100000},
"o1-mini": {"supports_reasoning": True, "max_output_tokens": 65536},
"o1-preview": {"supports_reasoning": True, "max_output_tokens": 32768},
"o3": {"supports_reasoning": True, "max_output_tokens": 100000},
"o3-mini": {"supports_reasoning": True, "max_output_tokens": 65536},
"o4-mini": {"supports_reasoning": True, "max_output_tokens": 100000},
# =========================================================================
# Gemini models
# =========================================================================
"gemini-2.0-flash": {"supports_reasoning": False, "max_output_tokens": 8192},
"gemini-2.0-flash-lite": {"supports_reasoning": False, "max_output_tokens": 8192},
"gemini-1.5-pro": {"supports_reasoning": False, "max_output_tokens": 8192},
"gemini-1.5-flash": {"supports_reasoning": False, "max_output_tokens": 8192},
"gemini-1.5-flash-8b": {"supports_reasoning": False, "max_output_tokens": 8192},
"gemini-2.5-pro-preview-06-05": {"supports_reasoning": True, "max_output_tokens": 65536},
"gemini-2.5-flash-preview-05-20": {"supports_reasoning": True, "max_output_tokens": 65536},
# =========================================================================
# Grok models
# =========================================================================
"grok-2-vision-1212": {"supports_reasoning": False, "max_output_tokens": 32768},
"grok-vision-beta": {"supports_reasoning": False, "max_output_tokens": 8192},
"grok-3": {"supports_reasoning": True, "max_output_tokens": 131072},
"grok-3-mini": {"supports_reasoning": True, "max_output_tokens": 131072},
# =========================================================================
# Mistral models
# =========================================================================
"pixtral-large-latest": {"supports_reasoning": False, "max_output_tokens": 16384},
"pixtral-12b-2409": {"supports_reasoning": False, "max_output_tokens": 8192},
"mistral-large-latest": {"supports_reasoning": False, "max_output_tokens": 16384},
"mistral-medium-latest": {"supports_reasoning": False, "max_output_tokens": 8192},
"mistral-small-latest": {"supports_reasoning": False, "max_output_tokens": 8192},
"codestral-latest": {"supports_reasoning": False, "max_output_tokens": 16384},
"open-mistral-nemo": {"supports_reasoning": False, "max_output_tokens": 8192},
# =========================================================================
# DeepSeek models
# =========================================================================
"deepseek-chat": {"supports_reasoning": False, "max_output_tokens": 8192},
"deepseek-reasoner": {"supports_reasoning": True, "max_output_tokens": 16384},
# =========================================================================
# Groq models (free tier, very fast)
# =========================================================================
"llama-3.3-70b-versatile": {"supports_reasoning": False, "max_output_tokens": 32768},
"llama-3.1-70b-versatile": {"supports_reasoning": False, "max_output_tokens": 32768},
"llama-3.1-8b-instant": {"supports_reasoning": False, "max_output_tokens": 8192},
"llama-3.2-90b-vision-preview": {"supports_reasoning": False, "max_output_tokens": 8192},
"llama-3.2-11b-vision-preview": {"supports_reasoning": False, "max_output_tokens": 8192},
"llama-3.2-3b-preview": {"supports_reasoning": False, "max_output_tokens": 8192},
"llama-3.2-1b-preview": {"supports_reasoning": False, "max_output_tokens": 8192},
"mixtral-8x7b-32768": {"supports_reasoning": False, "max_output_tokens": 32768},
"gemma2-9b-it": {"supports_reasoning": False, "max_output_tokens": 8192},
# =========================================================================
# Together AI models
# =========================================================================
"meta-llama/Llama-Vision-Free": {"supports_reasoning": False, "max_output_tokens": 4096},
"meta-llama/Llama-3.2-11B-Vision-Instruct-Turbo": {"supports_reasoning": False, "max_output_tokens": 4096},
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {"supports_reasoning": False, "max_output_tokens": 4096},
"meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo": {"supports_reasoning": False, "max_output_tokens": 4096},
"meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo": {"supports_reasoning": False, "max_output_tokens": 4096},
"Qwen/Qwen2.5-72B-Instruct-Turbo": {"supports_reasoning": False, "max_output_tokens": 4096},
"Qwen/QwQ-32B-Preview": {"supports_reasoning": True, "max_output_tokens": 16384},
"mistralai/Mixtral-8x7B-Instruct-v0.1": {"supports_reasoning": False, "max_output_tokens": 4096},
"deepseek-ai/DeepSeek-R1-Distill-Llama-70B-free": {"supports_reasoning": True, "max_output_tokens": 8192},
# =========================================================================
# OpenRouter models (free tier)
# =========================================================================
"meta-llama/llama-3.2-90b-vision-instruct:free": {"supports_reasoning": False, "max_output_tokens": 4096},
"meta-llama/llama-3.2-11b-vision-instruct:free": {"supports_reasoning": False, "max_output_tokens": 4096},
"qwen/qwen-2-vl-7b-instruct:free": {"supports_reasoning": False, "max_output_tokens": 4096},
"google/gemini-2.0-flash-exp:free": {"supports_reasoning": False, "max_output_tokens": 8192},
"deepseek/deepseek-r1:free": {"supports_reasoning": True, "max_output_tokens": 8192},
"deepseek/deepseek-chat:free": {"supports_reasoning": False, "max_output_tokens": 8192},
"meta-llama/llama-3.3-70b-instruct:free": {"supports_reasoning": False, "max_output_tokens": 4096},
"microsoft/phi-3-medium-128k-instruct:free": {"supports_reasoning": False, "max_output_tokens": 4096},
"mistralai/mistral-7b-instruct:free": {"supports_reasoning": False, "max_output_tokens": 4096},
# =========================================================================
# Cerebras models (free tier, very fast)
# =========================================================================
"llama3.1-70b": {"supports_reasoning": False, "max_output_tokens": 8192},
"llama3.1-8b": {"supports_reasoning": False, "max_output_tokens": 8192},
"llama-3.3-70b": {"supports_reasoning": False, "max_output_tokens": 8192},
# =========================================================================
# Default for unknown/local models
# =========================================================================
"_default": {"supports_reasoning": False, "max_output_tokens": 4096},
}
def get_provider_models(provider_name: str, config: Dict) -> List[str]:
"""Get models for a specific provider."""
models = config.get("models", [])
# For local providers with "dynamic", return empty list
# Users should use the custom_model field to specify their model
if models == "dynamic":
return []
return models if isinstance(models, list) else []
def get_all_models() -> List[str]:
"""
Get all models from all providers with provider prefix.
Models are organized with folder-like grouping:
- Models with "/" in their name are grouped by prefix (e.g., meta-llama/, Qwen/)
- Within each provider, models are sorted with grouped models together
- This creates a visual hierarchy in the dropdown
"""
all_models = []
for provider_name, config in PROVIDERS.items():
models = get_provider_models(provider_name, config)
# Separate models into groups based on their prefix
# Models with "/" get grouped, others stay as-is
grouped = {} # prefix -> list of models
ungrouped = [] # models without "/"
for model in models:
if "/" in model:
# Extract prefix (everything before the first /)
prefix = model.split("/")[0]
if prefix not in grouped:
grouped[prefix] = []
grouped[prefix].append(model)
else:
ungrouped.append(model)
# Build the sorted list: ungrouped first, then grouped by prefix
provider_models = []
# Add ungrouped models first (they don't have subfolders)
provider_models.extend(ungrouped)
# Add grouped models, sorted by prefix
for prefix in sorted(grouped.keys()):
prefix_models = grouped[prefix]
# Sort models within each group
prefix_models.sort()
provider_models.extend(prefix_models)
# Add provider prefix to all models
for model in provider_models:
all_models.append(f"[{provider_name}] {model}")
return all_models
def parse_model_selection(model_str: str) -> Tuple[str, str]:
"""Parse model string to get provider and model name."""
if model_str.startswith("["):
# Format: [Provider] model_name
end_bracket = model_str.find("]")
if end_bracket > 0:
provider = model_str[1:end_bracket]
model = model_str[end_bracket + 2:] # Skip "] "
return provider, model
return "OpenAI", model_str
def get_model_metadata(model: str) -> Dict[str, Any]:
"""Get metadata for a model."""
return MODEL_METADATA.get(model, MODEL_METADATA["_default"])
class SID_LLM_API(comfy_io.ComfyNode):
"""
Unified LLM Provider for Cloud and Local APIs.
Only vision-capable models are listed (required for image analysis).
Cloud Providers (API key required):
- Anthropic (Claude - all models support vision)
- OpenAI (GPT-4o, GPT-4.1 - vision models)
- Google Gemini (all models support vision)
- xAI Grok (grok-2-vision)
- Mistral AI (Pixtral vision models)
Free/Freemium Cloud Providers:
- Groq (free tier - Llama 4, Llama 3.2 vision)
- Together AI (free tier - Llama Vision, Qwen2.5-VL)
- OpenRouter (aggregator - free vision models)
- Fireworks AI (Llama 4, Llama 3.2 vision, Qwen VL)
- HuggingFace Inference (Llama 3.2 vision, Qwen VL)
Local Providers (no API key):
- Ollama (localhost:11434 - dynamic detection)
- LM Studio (localhost:1234 - dynamic detection)
- Custom OpenAI-compatible endpoints
For local vision models with VRAM management, use SID_LLM_Local instead.
"""
@classmethod
def define_schema(cls) -> comfy_io.Schema:
"""Define the node schema."""
all_models = get_all_models()
provider_list = list(PROVIDERS.keys())
# Max tokens presets
max_tokens_options = [
"Low (512)",
"Medium (2048)",
"High (8192)",
"Very High (Model Max)",
"Custom",
]
return comfy_io.Schema(
node_id="SID_LLM_API",
display_name="SID LLM API",
category="SID Photography Toolkit/LLM Providers",
description="Unified LLM provider: Cloud (Anthropic, OpenAI, Gemini, Groq, etc.) + Local (Ollama, LM Studio)",
inputs=[
# Provider selection
comfy_io.Combo.Input(
"provider",
options=provider_list,
default="Anthropic",
tooltip="Select LLM provider"
),
# API Key
comfy_io.String.Input(
"api_key",
default="",
multiline=False,
tooltip="API key for your provider (not needed for local providers)"
),
# Model selection - all models from all providers
comfy_io.Combo.Input(
"model",
options=all_models,
default="[Anthropic] claude-sonnet-4-5-20250929",
tooltip="Select model (provider prefix shows which API will be used)"
),
# Custom model override
comfy_io.String.Input(
"custom_model",
default="",
multiline=False,
tooltip="Override model name (for custom/unlisted models)"
),
# Custom API URL
comfy_io.String.Input(
"api_url",
default="",
multiline=False,
tooltip="Custom API URL (leave empty to use provider default)"
),
# Temperature
comfy_io.Float.Input(
"temperature",
default=0.3,
min=0.0,
max=2.0,
step=0.1,
round=0.1,
display_mode=comfy_io.NumberDisplay.slider,
tooltip="Creativity level (0=deterministic, 0.3=balanced, 1+=creative)"
),
# Max tokens preset
comfy_io.Combo.Input(
"max_tokens_preset",
options=max_tokens_options,
default="Medium (2048)",
tooltip="Output length: Low=512, Medium=2048, High=8192, Very High=Model Max"
),
# Custom max tokens (only used when preset is "Custom")
comfy_io.Int.Input(
"custom_max_tokens",
default=4096,
min=128,
max=200000,
tooltip="Custom max tokens (only used when preset is 'Custom')"
),
# Reasoning toggle
comfy_io.Boolean.Input(
"enable_reasoning",
default=True,
display_name="Enable Reasoning",
tooltip="Enable extended thinking for supported models (Claude 4.5, o1, o3, DeepSeek R1, etc.)"
),
],
outputs=[
LLM_MODEL_Type.Output(
"llm_model",
display_name="LLM_MODEL",
tooltip="LLM configuration to connect to SID_ZImagePromptGenerator"
),
],
)
@classmethod
def execute(
cls,
provider: str,
api_key: str,
model: str,
custom_model: str,
api_url: str,
temperature: float,
max_tokens_preset: str,
custom_max_tokens: int,
enable_reasoning: bool,
) -> comfy_io.NodeOutput:
"""Create and return the LLM model configuration."""
try:
# Parse model selection
model_provider, model_name = parse_model_selection(model)
# Use custom model if provided
if custom_model.strip():
model_name = custom_model.strip()
# Get provider config
provider_config = PROVIDERS.get(provider)
if not provider_config:
raise ValueError(f"Unknown provider: {provider}")
provider_name = provider_config["provider_name"]
# Determine API URL
if api_url.strip():
actual_url = api_url.strip()
else:
actual_url = provider_config.get("api_url", "")
# Use default URLs for local providers if not specified
if not actual_url:
if provider_name == "lmstudio":
actual_url = "http://localhost:1234/v1"
print(f"[SID_LLM_API] Using default LM Studio URL: {actual_url}")
elif provider_name == "ollama":
actual_url = "http://localhost:11434/v1"
print(f"[SID_LLM_API] Using default Ollama URL: {actual_url}")
else:
raise ValueError(f"API URL is required for {provider}")
# Auto-fix common URL mistakes for local providers
# LM Studio and Ollama require /v1 suffix for OpenAI-compatible API
if provider_name in ("lmstudio", "ollama", "openai_compatible"):
# Remove trailing slash first
actual_url = actual_url.rstrip("/")
# Add /v1 if missing
if not actual_url.endswith("/v1"):
actual_url = f"{actual_url}/v1"
print(f"[SID_LLM_API] Auto-corrected URL to: {actual_url}")
# Check if provider requires API key
requires_key = provider_config.get("requires_key", True)
is_local = provider_config.get("is_local", False)
# Validate API key - RAISE ERROR if required but missing
if requires_key and not api_key.strip():
api_key_url = provider_config.get("api_key_url", "")
error_msg = f"API key is required for {provider}."
if api_key_url:
error_msg += f"\n\nGet your API key at:\n{api_key_url}"
raise ValueError(error_msg)
# Get model metadata
metadata = get_model_metadata(model_name)
model_supports_reasoning = metadata["supports_reasoning"]
model_max_tokens = metadata["max_output_tokens"]
# Resolve max_tokens from preset
max_tokens_map = {
"Low (512)": 512,
"Medium (2048)": 2048,
"High (8192)": 8192,
"Very High (Model Max)": model_max_tokens,
"Custom": custom_max_tokens,
}
max_tokens = max_tokens_map.get(max_tokens_preset, 2048)
# Cap max_tokens to model's maximum
if max_tokens > model_max_tokens:
print(f"[SID_LLM_API] Note: Capping max_tokens from {max_tokens} to model max {model_max_tokens}")
max_tokens = model_max_tokens
# Validate max_tokens
if max_tokens < 1:
raise ValueError(f"max_tokens must be at least 1, got {max_tokens}")
# Validate temperature
if temperature < 0 or temperature > 2:
raise ValueError(f"Temperature must be between 0 and 2, got {temperature}")
# Determine if reasoning should be enabled
# Auto-disable reasoning for models that don't support it
actual_reasoning = model_supports_reasoning and enable_reasoning
# Inform user if reasoning was requested but not supported
if enable_reasoning and not model_supports_reasoning:
print(f"[SID_LLM_API] Note: Reasoning disabled - {model_name} does not support it")
# Create configuration
config = LLMModelConfig(
provider=provider_name,
model=model_name,
api_key=api_key.strip(),
api_url=actual_url,
max_tokens=max_tokens,
temperature=temperature,
supports_vision=True,
supports_system_prompt=True,
supports_reasoning=actual_reasoning,
extra_params={
"model_max_output_tokens": model_max_tokens,
"original_provider": provider,
"is_local": is_local,
"requires_key": requires_key,
"reasoning_supported": model_supports_reasoning,
},
)
provider_type = "Local" if is_local else "Cloud"
reasoning_status = "ON" if actual_reasoning else ("OFF (not supported)" if not model_supports_reasoning else "OFF")
print(f"[SID_LLM_API] Configured: {provider} ({provider_type})")
print(f" Model: {model_name}")
print(f" URL: {actual_url}")
print(f" max_tokens={max_tokens} ({max_tokens_preset}), temp={temperature}")
print(f" reasoning={reasoning_status}")
# Warn if local provider model may not support vision (but allow it)
if is_local and not is_likely_vision_model(model_name):
print(f" ⚠️ WARNING: '{model_name}' may not support vision/image input.")
print(f" For image analysis, consider using a vision model like 'llava', 'llama3.2-vision', etc.")
print(f" Proceeding anyway - if this model does support vision, you can ignore this warning.")
return comfy_io.NodeOutput(config)
except ValueError:
# Re-raise validation errors as-is
raise
except Exception as e:
# Catch any unknown errors and raise with context
raise RuntimeError(f"[SID_LLM_API] Unexpected error: {type(e).__name__}: {str(e)}") from e
Reference in New Issue View Git Blame Copy Permalink