Coming soon

About This Software

Find that one image in your head — with a single sentence

Tens of thousands of images sit on your drive with meaningless filenames and no tags. But you remember that one picture — "sunset at the beach."

ShiGuang is a fully local AI semantic search engine. Type what you're thinking, and it understands the visual content directly, instantly locating the exact image from your massive library. And it doesn't stop at images — your videos are searchable too, taking you straight to the exact moment.

Why ShiGuang?

• 100% Local, Zero Privacy Concerns — All AI models are built in. Your images never leave your computer. Works offline. No account, no cloud, no telemetry.

• Ready Out of the Box — No dependencies to install. Set it up like a game and start searching. Interface available in 11 languages.

• AI Re-Ranking — Two-stage retrieval: a fast shortlist across the whole library first, then a reranker model studies each candidate up close. Far more accurate than ordinary solutions.

• Every Format Welcome — iPhone HEIC, camera RAW files (CR3 / NEF / ARW and more), AVIF / JXL — all natively supported, no conversion needed.

• Search Video Content Too — Videos are automatically frame-indexed. Type "a cat rolling over", find the video, and playback starts right at that moment.

Six Ways to Search

• Text to Image — "Girl in a red dress", "cyberpunk street" — search exactly how you think

• Image to Image — Drop in a reference image to find every similar scene in your library

• Text + Image Combined — Reference image and text description work together in a single multimodal inference

• Negative Prompts — Unique "-" exclusion syntax: "landscape -night" for ever-sharper results

• Capture Text — Box any text on your screen; OCR reads it and instantly finds related images

• Screenshot Search — Grab your current screen and find similar pictures in your library

Deep Video Analysis: from "finding the video" to "finding the moment"

Start a deep analysis on any video, and ShiGuang understands it frame by frame:

• Relevance Curve — Type a description and the curve shows how well every moment of the video matches; scrub along it to preview

• Smart Frame Extraction — Scene detection picks the frames automatically; switch between Fast 2s and Fine 0.5s as needed

• Frame Search — Box any region inside the frame and search the whole library with that one detail

• Text Matches — Built-in OCR recognizes text frame by frame, highlighting every moment it appears

• Clip Export — Mark the start and end, and export the clip in one click

Explore Your Library

• Star Map — Your whole library becomes a 3D starfield: drag to rotate, scroll to zoom, click a star for similar images, double-click to jump. Semantically similar pictures naturally cluster into constellations

• Image Time Machine — Your library's story: time span, content profile, format breakdown — the whole evolution at a glance

• Style Mixer — Six style tracks (composition / tone / art style / light / texture / color): dial toward cool cyber or warm film looks to steer the mood of your search

More Capabilities

• Smart Quality Filtering — Automatically filters out blurry, pixelated, or solid-color junk images during indexing; strength adjustable

• Auto Long-Image Splitting — Long screenshots and webtoons are automatically split and indexed, so individual sections are searchable

• Smart GIF Frame Selection — Automatically picks the most representative frame

• Personalized Boost — Your browsing, similar-search and copy habits fine-tune ranking locally — the more you use it, the better it knows you

• Asset Management — Favorites, custom color tags, excluded folders

• MCP Protocol Support — Plug it into your AI assistant as a tool and let AI search images for you

• 12 Carefully Crafted Themes — From minimal solid colors to ambient animated light; one of them matches your desktop's mood

Built for Performance

• GPU Acceleration — Supports NVIDIA / AMD / Intel (Vulkan). Turbo-mode searches finish in under a second on modern GPUs

• Runs Without a Dedicated GPU — Automatically falls back to multi-core CPU mode

• HNSW Index — Millisecond response even on 100,000+ image libraries

• Virtual Scrolling — Buttery-smooth browsing through massive result sets

Who Is It For?

• Designers, illustrators, and photographers hoarding terabytes of assets

• Gamers and content creators buried under mountains of screenshots

• Video editors who need to locate a specific shot — and export it — in seconds

• iPhone users facing thousands of HEIC photos with no way to search them

• Anyone who wants image search but never wants to upload to the cloud

ShiGuang — Every image deserves to be found.

AI Generated Content Disclosure

The developers describe how their game uses AI Generated Content like this:

The application icon and the promotional artwork on this store page were created with the assistance of generative AI tools, then selected and reviewed by a human. The application itself does not generate any content during use: it bundles open-source, pre-trained AI models (Qwen3-VL-Embedding-2B and Qwen3-VL-Reranker-2B, Apache-2.0) that run entirely locally and offline on your PC. These models are used solely to analyze your own photos and videos — converting them into numeric feature vectors — to enable private, offline semantic search. No images, text, audio or other content is generated, and no data ever leaves your device.