Skip to content

Generation Types

This document details the different generation types available in the Sartiq platform.

Overview

Generations are categorized by two dimensions:

  1. Generation Type — How the generation is orchestrated (BASE vs AGENTIC)
  2. Generation Strategy — What kind of task is executed
Dimension Options
Type BASE, AGENTIC
Strategy GENERATE, EDITING, REFINE, VIDEO, BACKGROUND_FIXING, IMAGE_ADJUSTING, FACE_FIXING

Generation Types (Orchestration)

BASE

Standard generation without orchestration. Tasks are submitted directly to the Compute Server.

flowchart LR
    A[Create Generation] --> B[Submit Tasks]
    B --> C[Wait for Results]
    C --> D[Update Predictions]

Characteristics: - Simple, direct execution - No quality evaluation - No automatic retries - Fastest path to results

Use when: - Speed is priority - Manual review will handle quality - Iterating quickly

AGENTIC

Enhanced generation with quality evaluation and automatic retries.

flowchart TB
    A[Create Generation] --> B[Create OrchestratedShot]
    B --> C[Submit Task]
    C --> D[Get Result]
    D --> E[Evaluate Quality]
    E --> F{Score >= Threshold?}
    F -->|Yes| G[Accept]
    F -->|No| H{Attempts < Max?}
    H -->|Yes| I[Adjust & Retry]
    I --> C
    H -->|No| J[Accept Best]

Additional records: - OrchestratedShot — Stores orchestration metadata per prediction

Configuration:

Parameter Description Default
max_attempts Maximum retry count 3
min_confidence_score Quality threshold (0-1) 0.7
evaluation_provider Evaluator to use auto

Use when: - Quality is critical - Automated quality control needed - Final deliverable generation

Generation Strategies (Tasks)

GENERATE (Standard)

Creates new images from prompts with product/subject/style inputs.

Endpoint: POST /generations

Inputs:

Field Type Required Description
batch_size int No Number of predictions (default: 1)
subject_id uuid Yes* Subject to use
product_id uuid Yes* Product to feature
style_id uuid No Visual style reference
shot_type_string string Yes Shot type identifier
generated_prompt string No Custom prompt
auto_generate_prompt bool No Generate prompt via AI
width, height int No Output dimensions
seed int No Random seed

Task created: GENERATION

Flow: 

flowchart TB
    A[Validate Inputs] --> B[Load Subject/Product/Style]
    B --> C[Build or Generate Prompt]
    C --> D[Create GENERATION Task]
    D --> E[Include LoRA Weights]
    E --> F[Submit to Compute]

EDITING

Garment placement and image compositing operations.

Endpoint: POST /generations/edit

Inputs:

Field Type Required Description
base_img_path string Yes Base image to edit
reference_images array No Reference images for placement
subject_id uuid No Subject reference
product_id uuid No Product reference
tool enum Yes EDIT, SUBJECT_SWAP, or PRODUCT_ENHANCEMENT
width, height int No Output dimensions

Task created: EDITING

Tools:

Tool Purpose
EDIT General editing and compositing
SUBJECT_SWAP Replace subject in image
PRODUCT_ENHANCEMENT Enhance product appearance

Flow: 

flowchart TB
    A[Load Base Image] --> B[Extract Dimensions]
    B --> C[Process Reference Images]
    C --> D[Create EDITING Task]
    D --> E[Optional: Add Detail Enhancer]
    E --> F[Submit to Compute]

REFINE

Upscaling and detail enhancement for existing images.

Endpoint: POST /generations/refine

Inputs:

Field Type Required Description
base_image_url string Yes Image to refine
batch_size int No Refinement variations
target_width int No Target output width
target_height int No Target output height
upscale_enabled bool No Enable upscaling
upscale_factor float No Upscale multiplier
upscale_steps int No Upscale iterations
upscale_denoise float No Denoise strength
face_detail_enabled bool No Enhance faces
face_detail_steps int No Face enhancement iterations

Task created: REFINE

Creates additional record: Refine — Stores refinement parameters

Flow: 

flowchart TB
    A[Load Base Image] --> B[Create Refine Record]
    B --> C[Configure Upscale Settings]
    C --> D[Configure Face Detail Settings]
    D --> E[Create REFINE Task]
    E --> F[Submit to Compute]

VIDEO

Generate video from static images.

Endpoint: POST /generations/video

Inputs:

Field Type Required Description
base_image_url string Yes Source image
motion_bucket_id int No Motion intensity
fps int No Frames per second
duration float No Video duration

Task created: VIDEO_GENERATION

Flow: 

flowchart TB
    A[Load Source Image] --> B[Configure Motion]
    B --> C[Create VIDEO Task]
    C --> D[Submit to Compute]

BACKGROUND_FIXING

Fix or replace image backgrounds.

Endpoint: POST /generations/background-fix

Inputs:

Field Type Required Description
base_image_url string Yes Image to process
background_prompt string No New background description
background_image_url string No Background to use
mask_expansion int No Mask edge expansion

Task created: BACKGROUND_FIX

IMAGE_ADJUSTING

General image adjustments (color, exposure, etc.).

Endpoint: POST /generations/image-adjuster

Inputs:

Field Type Required Description
base_image_url string Yes Image to adjust
brightness float No Brightness adjustment
contrast float No Contrast adjustment
saturation float No Saturation adjustment
prompt string No Adjustment guidance

Task created: IMAGE_ADJUSTER

FACE_FIXING

Enhance facial details in images.

Endpoint: POST /generations/face-fixer

Inputs:

Field Type Required Description
base_image_url string Yes Image to enhance
strength float No Enhancement strength
prompt string No Enhancement guidance

Task created: FACE_ENHANCER

Comparison

Strategy Primary Use Creates Task Additional Records
GENERATE New images GENERATION —
EDITING Compositing EDITING —
REFINE Upscaling REFINE Refine
VIDEO Motion VIDEO_GENERATION —
BACKGROUND_FIXING BG replacement BACKGROUND_FIX —
IMAGE_ADJUSTING Color/exposure IMAGE_ADJUSTER —
FACE_FIXING Face enhancement FACE_ENHANCER —