Vibe Coded
News Image Editor
Autonomous system for editorial image tasks that replaced round-the-clock designer support.
$40k/yr saved
<$0.10 per task
Overview
Full image automation system for Cointelegraph’s editorial workflow. Editors interact with an AI agent through Slack to generate, edit, and compose images. The system handles everything from article illustrations to press release compositions with brand-consistent output at 1450x966px.
Problems Solved
- 24/7 designer availability needed for a global newsroom
- Brand consistency across thousands of images
- Press release images required manual logo composition
- Format conversion bottleneck (PSD, TIFF, HEIC to web)
Architecture
Slack Bot agent with Google ADK framework. Observer bot monitors channels for design tasks and forwards them. Web admin panel for prompt management, reference images, and model selection. All image outputs stored with versioning.
Key Features
- Generate images from text prompts, with optional Cointelegraph style matching
- Edit existing images via text instructions with multi-image support
- Press release generator: up to 5 variations with pixel-perfect logo protection (logos never sent to AI)
- Multi-logo composition: horizontal, partnership (X separator), press wall layouts
- AI background removal (~$0.002/image)
- Auto-conversion: SVG, PSD, TIFF, BMP, HEIC, AVIF to web formats
- Task review system with Accept/Reject buttons and quality metrics
- Dynamic model selection (Gemini, FLUX) via Slack Home Tab
- Rate limiting (10 req/min), event deduplication
- Web admin panel: prompt editor with version history, reference image management, model settings
Results
- Saves ~$40,000/year in designer costs
- Average cost under $0.10 per task (generation ~$0.02-0.05, editing ~$0.02-0.05)
- Deployed via Docker with GitLab CI/CD
- Full cost tracking and quality metrics dashboard
Stack