VisRecall is a modern reference hub for artists, powered by text-to-image retrieval.
It lets you search your saved images using natural language, so you can quickly find visual references — poses, compositions, lighting setups, outfits, and more — without digging through folders.
Think of it as your personal visual memory, organized by meaning instead of file names.
Artists collect thousands of reference images, but they’re hard to organize and even harder to find later.
VisRecall fixes that by using CLIP-based image embeddings to understand what’s in each image — so you can search by concept, not keyword.
“Show me a low-angle shot of a samurai in rain.”
“Find all references with backlighting.”
“Pose similar to a leaping figure.”
VisRecall retrieves the closest matches instantly, helping artists rediscover and reuse their references intuitively.
-
🔍 Text-to-Image Search:
Find references by describing what you remember — not by browsing folders. -
🖼️ Smart Image Embeddings:
Every image is encoded using CLIP models (LAION or Danbooru variants) for semantic understanding. -
📂 Reference Hub:
Organize, tag, and preview your visual library in one clean dashboard. -
🧩 Multi-Model Support:
Switch between general-purpose and art-specific embeddings for better accuracy. -
⚡ Instant Retrieval:
Searches return semantically relevant results in milliseconds. -
☁️ Supabase Integration (Optional):
Store embeddings and metadata in the cloud for access anywhere.
| Layer | Technology |
|---|---|
| Frontend | React + TypeScript + Vite |
| Backend | Supabase (Auth + Database) |
| Embedding Engine | LAION & Danbooru CLIP Models |
| Vector Search | Supabase pgvector / local FAISS index |
| Styling | Tailwind + shadcn/ui |
| State | React Query / Zustand |
- An artist searches “over-the-shoulder camera angles” and finds all relevant photo studies.
- A character designer types “baggy streetwear, male character” to pull up similar fashion references.
- A background painter recalls “sunset cityscapes with rim light” from months ago in seconds.
VisRecall is built for artists who collect and curate, not generate.
It’s meant to make your existing visual library searchable, meaningful, and fast — turning chaos into creativity.
“You don’t need AI to make art — just a better way to remember what inspires it.”
| Feature | Status |
|---|---|
| Local text-to-image retrieval | 🟢 Working |
| Supabase cloud sync | 🟡 Planned |
| Multi-user reference sharing | 🟡 Planned |
| Embedding visualization (t-SNE map) | ⚪ Future update |
| Mobile-friendly dashboard | ⚪ Future update |
VisRecall is developed by KMI RVNNT, a character artist and software engineer passionate about building tools that support real artists instead of replacing them.
MIT License — open source, for artists and researchers.