Image Generation and Document Processing

Relevant source files

• Lovable/Agent Tools.json

This document covers Lovable's specialized capabilities for AI-powered image generation and document content extraction. These tools enable web applications to create visual assets dynamically and process user-uploaded documents without external service dependencies. For general file operations, see File Operations and Line-Replace Editing. For external resource handling, see Dependency Management and Search Tools.

Purpose and Scope

Lovable provides two distinct content processing subsystems:

1. Image Generation: AI-powered image creation and editing using FLUX models (imagegen--generate_image, imagegen--edit_image)
2. Document Processing: Content extraction from multiple document formats (document--parse_document)

Both subsystems operate as integrated tools within Lovable's agent architecture, enabling automated asset creation and document parsing workflows.

Image Generation Architecture

Model Selection Strategy

Lovable provides two FLUX models with distinct performance characteristics:

Model	Use Case	Performance	Max Resolution	Selection Criteria
flux.schnell	Small images, icons, thumbnails	Fast	1920x1920	Default; images < 1000px
flux.dev	Hero images, fullscreen banners	Slower, higher quality	1920x1920	Large images, explicit quality requests

Dimension Constraints:

• Minimum: 512px per dimension
• Maximum: 1920px per dimension
• Step: 32px multiples (512, 544, 576, ..., 1920)

The tool automatically selects flux.schnell unless the image size, placement context, or user requirements indicate the need for flux.dev.

Sources: Lovable/Agent Tools.json299-327

Image Generation Workflow

Prompt Engineering Guidelines Lovable/Agent Tools.json300-327:

• Aspect Ratio Mention: Include aspect ratio in prompt (e.g., "A 16:9 aspect ratio image of...")
• Context Specification: Mention image purpose ("hero image", "icon", "banner")
• Quality Suffix: Append "Ultra high resolution" for maximum quality
• Location Awareness: Consider aspect ratio based on page placement

Example Prompt Structure:

A 16:9 aspect ratio hero image of a sunset over a calm ocean. Ultra high resolution.

Sources: Lovable/Agent Tools.json299-327

Image Editing and Merging

The imagegen--edit_image tool provides two operational modes:

Single Image Edit Examples Lovable/Agent Tools.json328-351:

• "make it rainy" - Add weather effects
• "change to sunset lighting" - Modify lighting
• "add snow" - Environmental changes
• "make it more colorful" - Color enhancement

Multiple Image Merge Examples:

• "blend these two landscapes seamlessly"
• "combine the foreground of the first image with the background of the second"
• "merge these portraits into a group photo"
• "create a collage from these images"

Use Case: Object Consistency: The tool excels at maintaining character or object consistency across scenes by reusing the same base image in different contexts.

Sources: Lovable/Agent Tools.json328-351

Tool Parameter Reference

imagegen--generate_image

Parameters Lovable/Agent Tools.json301-327:

• prompt (required): Text description with aspect ratio and quality hints
• target_path (required): Save location, prefer src/assets/ folder
• width (optional): Image width, default determined by model
• height (optional): Image height, default determined by model
• model (optional): "flux.schnell" (default) or "flux.dev"

Critical Constraint: Generated images must be imported as ES6 modules:

Important Note Lovable/Agent Tools.json300-301: Do not replace user-uploaded images with generated images unless explicitly requested.

Sources: Lovable/Agent Tools.json299-327

imagegen--edit_image

Parameters Lovable/Agent Tools.json330-351:

• image_paths (required): Array of source image paths
  • Single path: Edit mode
  • Multiple paths: Merge mode
• prompt (required): Edit/merge instructions
• target_path (required): Output file path

Preferred Usage: When users request image tweaks, use imagegen--edit_image rather than generating a new image from scratch.

Sources: Lovable/Agent Tools.json328-351

Document Processing

Supported Formats and Capabilities

The document--parse_document tool provides comprehensive document content extraction:

Processing Capabilities Lovable/Agent Tools.json286-298:

1. Structure Preservation: Maintains document hierarchy and formatting
2. Table Extraction: Converts tables to structured data
3. Image Extraction: Retrieves embedded images
4. OCR Processing: Extracts text from image content within documents

Page Limit: First 50 pages are processed per document.

Sources: Lovable/Agent Tools.json286-298

Document Parsing Tool

Parameter Lovable/Agent Tools.json289-298:

• file_path (required): Path to document file, typically from user-uploads:// virtual filesystem

Integration Pattern: Document parsing typically follows this workflow:

1. User uploads document via Lovable interface (saved to user-uploads://)
2. Agent invokes document--parse_document with uploaded file path
3. Extracted content used for analysis, summarization, or data extraction
4. Results displayed to user or stored in application database

Sources: Lovable/Agent Tools.json286-298

File System Integration Patterns

Virtual Filesystem Interaction

Both image and document tools interact with Lovable's virtual filesystem structure:

Key Integration Points:

1. Image Generation Lovable/Agent Tools.json316-317:

   • Saves directly to project root, preferably src/assets/ folder
   • Must be imported as ES6 modules in React components
   • Becomes part of application bundle

2. Image Editing:

   • Can read from user-uploads:// or project root
   • Saves edited results to project root
   • Supports reusing uploaded images in different contexts

3. Document Processing Lovable/Agent Tools.json286-298:

   • Reads from user-uploads:// virtual filesystem
   • Returns structured content (not saved to project)
   • Content used for runtime processing only

Filesystem Tool Relationship: For copying files between virtual filesystems, see lov-copy tool in Dependency Management and Search Tools.

Sources: Lovable/Agent Tools.json127-143 Lovable/Agent Tools.json286-298 Lovable/Agent Tools.json299-351

Tool Selection Decision Matrix

Requirement	Tool	Key Parameters
Generate new image	imagegen--generate_image	prompt, target_path, model
Edit existing image	imagegen--edit_image	image_paths (1), prompt, target_path
Merge multiple images	imagegen--edit_image	image_paths (2+), prompt, target_path
Extract PDF content	document--parse_document	file_path
Process Office documents	document--parse_document	file_path
Extract audio transcription	document--parse_document	file_path

Model Selection for Image Generation:

if image_size < 1000px or standard_quality:
    use flux.schnell  # Default, fast
elif image_size >= 1000px or hero_image or user_requests_quality:
    use flux.dev      # Slower, higher quality

Sources: Lovable/Agent Tools.json286-351

Operational Constraints and Best Practices

Image Generation Constraints

Dimension Rules Lovable/Agent Tools.json304-322:

• All dimensions must be multiples of 32
• Minimum dimension: 512px
• Maximum dimension: 1920px
• Invalid examples: 500px (not multiple of 32), 2048px (exceeds max)
• Valid examples: 512px, 768px, 1024px, 1920px

Import Requirement Lovable/Agent Tools.json300-301:

User Upload Protection Lovable/Agent Tools.json300-301: Never replace user-uploaded images with AI-generated images unless explicitly requested by the user.

Document Processing Constraints

Page Limit Lovable/Agent Tools.json287: Only the first 50 pages of documents are processed.

Format Support: While the tool supports "many other formats" beyond PDF, Word, PowerPoint, Excel, and MP3, the exact list is not enumerated in the tool description.

Parallel Execution Recommendation

Both image generation and document parsing are I/O-bound operations. When multiple images need generation or multiple documents require parsing, invoke tools in parallel rather than sequentially for performance optimization. This pattern aligns with Lovable's general parallel execution strategy (see File Operations and Line-Replace Editing).

Sources: Lovable/Agent Tools.json286-351