Back to all projects
cli tool•2024•Creator & Lead Developer

Smart Image CLI (AI-Powered Image Optimizer & SEO Renamer)

Lightning-fast image compression and local vision AI SEO renaming CLI built with Bun, TypeScript, and LM Studio.

Source Code Live Demo

Technologies & Tools

buntypescriptlm-studiolocal-aigemmaimage-optimizationcli

The Challenge

Frontend developers and content editors frequently face two time-consuming image bottlenecks:

  1. Compressing high-resolution PNGs/JPEGs into modern WebP/AVIF formats without visual degradation.
  2. Manually writing descriptive, SEO-optimized alt text and renaming non-descriptive files like IMG_9482.png into search-friendly slugs like hotel-sonmarg-snow-suite.webp.

Cloud vision APIs (OpenAI, Google Cloud Vision) solve the labeling problem but introduce recurring API costs, network latency, and privacy compliance issues when processing confidential client assets.

The Solution & Architecture

smart-img-cli is an open-source command-line tool built on the Bun runtime that integrates with offline Small Language Models (SLMs) running via LM Studio:

[Raw Photos / Screenshots] ──> [smart-img-cli @ Bun]
                                     │
                 ┌───────────────────┴───────────────────┐
                 ▼                                       ▼
    [Local Gemma 2B Vision]                 [Multi-Core Image Engine]
    (LM Studio OpenAI Endpoint)             (Sharp / WebP Compression)
                 │                                       │
                 ▼                                       ▼
    "alt: Modern dark-mode UI..."           "compressed: 4.2MB -> 180KB"
                 │                                       │
                 └───────────────────┬───────────────────┘
                                     ▼
                      [Optimized & Renamed Assets]

Key Engineering Features

  • Local SLM Vision Pipeline: Connects to local OpenAI-compatible inference endpoints (http://127.0.0.1:1234/v1) running Google’s Gemma 2B Vision model, generating accurate 15-word alt tags and slugified filenames in ~60ms per image.
  • Bun Zero-Copy File I/O: Takes advantage of Bun’s native process spawning and memory-mapped file buffers for fast batch transformations.
  • Next-Gen WebP/AVIF Encoding: Converts raw assets with configurable perceptual quality presets while preserving color profiles and aspect ratios.
  • 100% Privacy & Zero API Cost: Runs entirely on local machine hardware with no outbound internet traffic and $0.00 cloud fees.