Fal.ai Integration¶

Sartiq uses Fal.ai as the primary ML inference platform for image generation, LoRA training, and specialized vision tasks.

Overview¶

Property	Value
Provider	Fal.ai
Primary Use	Image generation, LoRA training, video generation
SDK	`fal-client` Python SDK
Architecture	Serverless GPU inference

Use Cases¶

1. Image Generation¶

Generate images from text prompts with optional LoRA models for subject/style consistency.

flowchart LR
    PROMPT[Prompt + LoRAs] --> FAL[Fal.ai API]
    REF[Reference Images] --> FAL
    FAL --> IMG[Generated Images]

2. LoRA Training¶

Train custom LoRA models for subjects, products, and styles.

flowchart LR
    DATASET[Training Dataset] --> TRAIN[flux-lora-fast-training]
    TRAIN --> LORA[LoRA Weights]
    LORA --> GEN[Use in Generation]

3. Image Editing¶

Edit existing images with inpainting, virtual try-on, and garment fixing.

4. Video Generation¶

Generate videos from images with motion and timing control.

5. Vision Tasks¶

Face Detection - Moondream3 for face bounding boxes
Segmentation - SAM-3 for person/garment segmentation
Image Expansion - Bria for outpainting

Architecture¶

flowchart TB
    subgraph ComputeServer["Compute Server"]
        FC[FalClient]
        FSP[FalStandardProcessor]
        FGF[FalGarmentFixerProcessor]
        FTS[FalTrainingService]
    end

    subgraph FalPlatform["Fal.ai Platform"]
        FLUX[Flux Models]
        SAM[SAM-3]
        MOON[Moondream3]
        TRAIN[Training API]
        WF[Custom Workflows]
    end

    FC --> FLUX
    FC --> SAM
    FC --> MOON
    FSP --> WF
    FTS --> TRAIN

Key Features¶

Multiple LoRA Support¶

Generate images with up to 4 simultaneous LoRAs:

LoRA Type	Purpose	Default Weight
Subject	AI model identity	0.8
Product	Product appearance	0.8
Style	Visual aesthetic	0.8
Product2	Secondary product	0.8

Custom Workflows¶

Sartiq uses custom Fal.ai workflows for specialized tasks:

Workflow	Purpose
`workflows/Shootify/flux-krea-lora-i2i`	Image generation with LoRAs
`workflows/Shootify/editing-workflow`	Image editing pipeline

Task Status Tracking¶

Real-time status updates during processing:

Queued - Waiting for worker
InProgress - Processing
Completed - Results ready

Documentation¶

Section	Description
Setup	Configuration and API keys
Models	Available models and endpoints
Examples	Code examples and workflows