GENIMG – System & UI Overview
1. Core System Architecture & Workflow
GENIMG is a local workspace designed for the structured generation, management, and versioning of AI-generated images and videos.
How it works:
- Strict Project Structure: Work is organized into numbered folders (e.g.
PROJECT_001), initialized from a _template. Each project contains a PROJECT.md file that dictates the style, color palette, and context, serving as the absolute source of truth for generation.
- Milestone Tracking: The system relies heavily on files—both at the workspace root and within individual projects—to log major changes, new versions, and project history.
- Generation Engine: The actual asset creation is handled by Python scripts (
generate.py, generate_video.py) that communicate with Google's Vertex AI APIs (Imagen for images, Veo for videos) via OAuth2 authentication.
- Meticulous Prompt Logging: Every generated asset (in
generated/) is versioned (e.g. v1_asset.jpg) and never overwritten. A Markdown file in logs/ stores the exact prompt, parameters, timestamp, and link to the output file.
2. Graphical User Interface (GENIMG UI)
The genimg-ui is a local Next.js web application on localhost:3000. It is the visual frontend and control center for the GENIMG workspace.
Key Features & Functions:
- Project Dashboard: Overview of all
PROJECT_XXX directories, project details, visual timeline (from CHANGELOG.md), and new project creation.
- Asset Gallery: Customizable gallery for images and videos—Grid/List views, resizing, filtering by type.
- Interactive Generation Forms: Forms for images (Imagen) and videos (Veo): prompts, target project, parameters (e.g. first frame for video). Backend spawns Python scripts via API routes.
- Real-Time Streaming (SSE): Terminal logs and script progress streamed live; inline preview when the asset is ready.
- Advanced Asset Interaction: Lightbox on click; copy prompts, view prompt log inline, compare versions side-by-side.
- Workspace Navigation: Batch View for related collections, source badges (standard vs. fine-tuned), prompt search across all log files.