Universal Metadata
Parses prompt, model, steps, CFG, sampler, seed and workflows from ComfyUI, Stable Diffusion (A1111, Forge, Fooocus), InvokeAI, Midjourney, DALL·E 3, Firefly and more — even from WebP.
11+ TOOLSSILKSTACK IMAGE BROWSER — LOCAL AI FOR YOUR AI IMAGES
Every AI image you've ever made.
local, private and searchable.
A free, local image viewer and organizer for AI-generated images — built for ComfyUI, Stable Diffusion and A1111 workflows. It reads the prompt, model, seed and workflow from your files, then adds on-device AI: semantic search, auto-tagging and similarity stacks. No accounts. No cloud. Everything on your machine.
✦ Free image viewer for AI artists
·
Windows, macOS & Linux
·
MPL-2.0 open source
·
Offline AI on your own GPU
ABOUT
Your generation history,
turned into a searchable gallery
Every AI artist accumulates thousands of images — with their secrets hidden in the metadata. SilkStack Image Browser is a local image viewer and organizer that reads those secrets and turns your output folders into a searchable, taggable, beautifully rendered library. All offline, all on your machine.
🧠 Prompt · model · seed parsed
⚡ ~98% faster indexing
🔒 100% local — zero cloud
Built for the AI image workflow
SilkStack Image Browser scans your folders and parses rich metadata from ComfyUI, Automatic1111 and a dozen other tools — reading everything from the prompt and model to the sampler, CFG, seed and even the full workflow graph. What you get back is a library you can actually live in.
- ◈ Deep metadata parsing: ComfyUI, Stable Diffusion (A1111 / Forge / Fooocus / SD.Next), InvokeAI, SwarmUI, Midjourney, DALL·E 3, Firefly, DreamStudio, Draw Things
- ◈ Automatic folder watching — new generations appear live while you wait
- ◈ Search by prompt text, model, steps, CFG, sampler, seed — then sort by date, name or relevance and filter by portrait / landscape / square
- ◈ Images, MP4 / WEBM video and animated GIFs, side by side in one thumbnail grid
- ◈ Drag an image straight back into ComfyUI to reload its prompt and workflow
0
image tools parsed
0
ComfyUI node types
0
faster indexing
FEATURES
Everything your AI image
library has been missing
Six things a local AI image viewer has to get right — hover them, they feel the way they should.
Adaptive Grid
A beautiful thumbnail grid that auto-adjusts to every image's aspect ratio, with buttery smooth scrolling — even when your folder holds 100,000 generations, videos and GIFs included.
SMOOTH-SCROLLAuto-Watch
Point it at your output folders and watch new images appear in real time, the moment your sampler finishes. No refresh, no rescan, no waiting.
LIVE-INDEXSearch & Filters
Find anything by prompt text, model, steps, CFG, sampler or seed — or switch on semantic search and find it by meaning. Filter by portrait, landscape or square; sort by date, name or relevance. Your full history, queried in a keystroke.
CTRL+KDrag & Drop to ComfyUI
Drag any image back into ComfyUI to instantly reload its prompt and workflow graph. Works with Photoshop, Discord and your file explorer too.
DND-READYTags & Collections
LLM auto-tags plus your own manual tags, similarity stacks that group near-identical and duplicate generations, collections, folders with emoji icons, and removable drives — including encrypted ones.
LIBRARY-OSAI INTELLIGENCE
Local AI that reads
what you meant
Semantic search, automatic tagging and similarity stacks — run by small AI models on your own GPU through WebGPU. No cloud service, no account, no API keys, and your images, prompts and tags are never uploaded anywhere.
Semantic Search
Keyword search matches words; semantic search matches meaning. Describe an image loosely — “a girl with a puppy in a palace” — and SilkStack finds generations whose prompts mean the same thing, even when they share almost no words, and even in another language.
MEANING, NOT WORDSAI Auto-Tagging
A small local LLM reads the prompt inside every image and writes up to 15 searchable tags — subjects, styles, lighting, concepts — including synonyms and alternate phrasings a searcher might type. Tags sit alongside your manual ones, and an interrupted run resumes where it stopped.
UP TO 15 TAGS / IMAGESimilarity Stacks
Prompts are embedded with the same local model that powers search, so reworded, translated and synonym-only variations land in the same stack. Group by Prompt, Model or Loras, and highlight exactly which words changed between two generations.
VECTOR GROUPING
✦ Differing words highlighted
◈ One stack, 40 variations
Spot what changed
between generations
A folder holding forty variations of the same idea becomes a few tidy cards. Vector similarity is what makes the grouping smart: prompts are compared by what they mean, not just how they are worded, so “a girl in a pink gown” and “a princess wearing a pink dress” end up in the same stack.
- ◈ Group by Prompt, Model or Loras — or turn grouping off entirely
- ◈ Highlight marks the exact words that differ between the prompts in a stack — spot an added colour, an age, a swapped adjective at a glance
- ◈ Vector matching can only add images to a stack — it never breaks up a group that text matching already found
- ◈ Keep working without AI: plain text matching still stacks your library, and the browser itself never needs a model
Your GPU,
your rules
Local AI is only worth it if you stay in control of it. SilkStack shows exactly which models are loaded, how much VRAM they hold, how much disk they take — and gives you a switch for all of it.
- ◈ One top-bar toggle turns off model loading, semantic search and auto-tagging together, so the app stays out of your GPU while you generate
- ◈ The footer model pill shows every model resident in GPU memory with its declared VRAM, plus an Eject button to hand the memory straight back
- ◈ Settings → AI Intelligence lists each cached model with its size and deletes them one by one to reclaim disk space
- ◈ Models are downloaded once and then run fully offline; on a dual-GPU laptop you pick which GPU they use
0
embedding models
0
tags per image, max
0
images uploaded
🔒 Eject VRAM any time
⚡ Offline after one download
Every AI feature runs on your own hardware via WebGPU — the only network access is the one-time model download. The AI pack is unlocked by a premium license — a monthly plan with a 7-day free trial, or a one-time lifetime purchase; the browser itself stays free and open source.
SHOWCASE
The library, live
Real screenshots from the actual app, running on a real machine.
Every pixel is rendered locally. Your images never leave your machine — the interface doesn't either.
SUPPORT
Bugs, ideas and questions
SilkStack is built in the open. GitHub Issues is the one place everything is tracked — bug reports, feature requests and setup or licensing questions alike — so nothing gets lost and everyone can search the answer.
Open an issue on GitHub
Tell us what happened and we'll take it from there. It's the fastest route to a fix, and there's no form to fill in — just a title and a description.
- ◈Bug? Include your OS, the SilkStack version and the file or generator that misbehaved.
- ◈Idea? Describe the workflow you want to speed up — that is what shapes the roadmap.
- ◈License or billing? Open an issue too, but never paste an activation key into a public thread.