infinite-image-browsing

Infinite Image Browsing (IIB) is a media browser and file manager aimed especially at images and videos generated by AI image tools. It indexes generation metadata from Stable Diffusion web UI, including stealth metadata support, Fooocus, NovelAI, StableSwarmUI, Invoke.AI, and partially ComfyUI; Pixiv support is available through a separate plugin. The indexed prompt, model, LoRA, and similar metadata are converted into searchable tags with autocomplete, frequency sorting, translation support, fuzzy filename/information search, advanced search syntax, media filters, video tag search, random sorting, and custom auto-tagging rules. The interface provides thumbnail grids with configurable thumbnail size and grid width, full-screen previews, nested JSON/EXIF metadata inspection, keyboard navigation, remote video playback, video covers, a TikTok-style vertical browsing mode, side-by-side image comparison with generation-info comparison, and a walk mode that flattens folders for continuous browsing. Files can be selected with Ctrl/Shift/Cmd for deletion, moving, copying, drag-and-drop, folder creation, tag changes, packaged download, and batch processing. Users can add custom search paths and mark favorites through custom tags. IIB includes usage statistics such as a daily contribution-style heatmap, monthly generation trends, top models, source distribution, total image count, and disk usage. It also provides a tag relationship graph and multilingual support for Simplified Chinese, Traditional Chinese, English, and German. AI features include Smart Organize and experimental natural-language categorization/search. Smart Organize uses embeddings to cluster images by prompt semantics, asks an LLM to generate folder names, and presents a preview before files are moved or copied; options cover minimum cluster size, recursive folders, and background processing. Semantic search stores embeddings incrementally in SQLite and ranks results by cosine similarity. These features require an OpenAI-compatible embedding/chat provider through variables such as OPENAI_BASE_URL, OPENAI_API_KEY, EMBEDDING_MODEL, AI_MODEL, and TOPIC_TITLE_MODEL, plus Python dependencies including numpy and hnswlib. TwelveLabs Marengo can be selected as an optional embedding backend. The README notes that semantic results depend on model and prompt-metadata quality, and failed AI calls return errors rather than using mock results. Deployment options include installation from a URL as an SD-webui extension, standalone Python execution, a Windows desktop release, and iframe/library embedding. It can also be used as a skill with AI agents such as Claude Code, Cursor, and OpenClaw. Security and administration options include a custom authentication secret, allowed-path configuration, access permissions, public-access defaults when used with SD-webui, and read-only mode. Thumbnails, image caches, and video covers can be generated in advance, and the cache directory can be set with IIB_CACHE_DIR.