All Projects

Albin

Written by .

An arts client came to us wanting an AI tool that could discuss art with real critical depth instead of labeling the subject and moving on. We built Albin, a platform that uses GPT-4 Vision to analyze a piece of artwork across composition, color, technique, and historical context, and return feedback organized the way a critic or curator actually thinks about a piece. It's live today at albin.ai, serving artists, curators, teachers, and students.

Background

Most AI image tools are built for classification: tag the subject, guess the style, done. That's fine for a stock photo library. It doesn't help a curator writing catalog copy that explains why two pieces belong in the same exhibition, or an artist trying to understand why a composition is or isn't working. Our client had already looked at what existed and found two camps: consumer apps built for style transfer and image generation, and technical pipelines that needed an ML engineer just to configure. Neither one was usable by someone sitting down to actually think about art.

The problem: AI that can't tell you why a piece works

The gap was for something in between: a tool accessible enough for a working artist or curator to use directly, with enough analytical depth to be worth their time. Raw calls to a multimodal API return generic descriptions. Getting from there to structured, art-literate output meant building a layered analysis architecture and doing substantial prompt engineering so the system reliably evaluated compositional structure, color harmony, brushwork and texture, historical context, and technical execution, instead of producing a wall of generic text.

What we built

The core of Albin is that analysis pipeline. Upload or send an image, and the platform returns structured feedback by category rather than one block of prose. An artist gets specific notes they can act on. A curator gets language they can adapt directly for catalog copy. A student gets a framework for what they're actually looking at.

We added a browser extension so users can right-click any image on the web and send it straight to Albin for analysis, which matters more than it sounds: a curator browsing an online gallery, or a student researching a movement, can analyze a piece in context instead of switching tabs and losing their place. We'd built browser extensions before, including Classway, so the integration pattern wasn't unfamiliar territory. Beyond single-image analysis, the platform includes a prompt library and multiple chat modes for comparing a user's own work against reference pieces or building out exhibition narratives through guided conversation.

The platform runs on Next.js, with Supabase handling authentication and data, Stripe managing subscription billing, and Vercel handling deployment. Access is tiered by usage, from a free plan for individual artists up to a plan built for institutional users running high query volumes.

The result

Albin is live and in use. The part that surprised us during testing: art professionals used it less for formal critique and more for generating language they could adapt into catalog copy and grant applications. That's a different use case than what we designed around, and it was the clearest sign the tool was solving a real problem instead of just being an interesting demo.

Who did what

Burhan Pasha was the client lead and owned delivery from scoping through launch. Saif Pasha led product design and the GPT-4 Vision integration architecture. The Silverthread Labs team built the analysis pipeline, the browser extension, and the subscription platform.

If you're building an AI product for a specialized audience, I'm happy to talk through what that build could look like for you. Reach out or see how we scope full builds.

// NEXT STEP

Building something like this?

Tell me what you're trying to build. The first call ends with what to build first, whether or not you hire us.