Integrations¶

Sartiq integrates with external AI services for image generation, text processing, and multimodal tasks.

Overview¶

Integration	Purpose	Primary Use
Anthropic	Claude API	Image captioning, vision analysis, outfit generation
Fal.ai	ML inference platform	Image generation, LoRA training, video
Google AI	Gemini + Vertex AI	Styling, captioning, image generation

Anthropic (Claude)¶

Claude's vision and text capabilities power image analysis and structured content generation.

Feature	Model	Use Case
Image Captioning	Claude Sonnet 4	Product descriptions for ML training
Outfit Generation	Claude 3.5 Sonnet	Styling recommendations
Vision Analysis	Claude Sonnet 4	Multi-image comparison

Key capabilities: Vision, structured JSON output, multi-turn conversations

Anthropic Integration →

Fal.ai¶

Serverless GPU inference for image generation, training, and processing.

Feature	Model/Endpoint	Use Case
Image Generation	Flux + Custom LoRAs	Fashion photography
LoRA Training	flux-lora-fast-training	Subject/style models
Face Detection	Moondream3	Face bounding boxes
Segmentation	SAM-3	Person/garment masks
Video Generation	video-gen	Motion from stills

Key capabilities: Multiple LoRAs, custom workflows, real-time status

Fal.ai Integration →

Google AI (Gemini + Vertex AI)¶

Google provides two integration pathways:

Gemini API (LLM)¶

Text generation and multimodal analysis via LiteLLM abstraction.

Feature	Model	Use Case
Styling	gemini-2.0-flash	Outfit recommendations
Captioning	gemini-2.5-flash	Product descriptions
Visibility	gemini-flash	Product filtering

Vertex AI (Image)¶

Native image generation and editing with Gemini models.

Feature	Model	Use Case
Generation	gemini-3-pro-image	Text-to-image
Editing	gemini-3-pro-image	Virtual try-on
Vision	gemini-3-pro	Fashion analysis

Key difference: Vertex AI supports image generation with permissive safety settings for fashion content.

Google AI Integration →

Architecture¶

flowchart TB
    subgraph Backend["Backend API"]
        CAP[Captioning]
        STYLE[Styling]
        LLM[LLM Handler]
    end

    subgraph Compute["Compute Server"]
        GEN[Generation]
        EDIT[Editing]
        TRAIN[Training]
    end

    subgraph External["External Services"]
        ANTH[Anthropic Claude]
        FAL[Fal.ai]
        GEM[Gemini API]
        VERT[Vertex AI]
    end

    CAP --> ANTH
    STYLE --> GEM
    LLM --> ANTH
    LLM --> GEM

    GEN --> FAL
    GEN --> VERT
    EDIT --> FAL
    EDIT --> VERT
    TRAIN --> FAL

Common Patterns¶

API Key Management¶

All keys are managed via environment variables:

# Anthropic
ANTHROPIC_API_KEY="sk-ant-..."

# Fal.ai
FAL_KEY="..."

# Google
GEMINI_API_KEY="..."
GOOGLE_API_KEY="..."
GOOGLE_CLOUD_PROJECT="..."

Error Handling¶

Standard error wrapper across integrations:

from sartiq.integrations.errors import (
    ExternalServiceError,
    RateLimitError,
    AuthenticationError,
)

Retry Logic¶

All integrations use exponential backoff:

@retry(
    stop=stop_after_attempt(3),
    wait=wait_exponential(multiplier=1, min=1, max=10),
    retry=retry_if_exception_type(RateLimitError)
)
async def call_external_service():
    ...

Provider Selection¶

Task Type	Primary Provider	Fallback
Captioning	Anthropic	Gemini
Styling	Gemini	Anthropic
Image Generation	Fal.ai	Vertex AI
LoRA Training	Fal.ai	-
Vision Analysis	Anthropic	Gemini