Suvorin.Mebel: AI Configurator

Custom furniture has a visualization problem. A client describes what they want — a sectional sofa in a specific fabric, with particular dimensions, for a room they can picture but cannot easily share — and the seller is left translating words into a mental image, then hoping the final product matches. Suvorin.Mebel, a custom furniture studio, came to us in 2023 because that translation step was costing them deals.

Context

Suvorin.Mebel operates a Tilda-based storefront with Notion as the content management layer. Product catalogs, fabric options, and pricing lived in Notion and synced to the frontend. Payments ran through Stripe. The business model is made-to-order: every piece is built to a client's specification. That means every sale requires the client to confidently imagine the result before committing.

Problem

The bottleneck was the mockup. When a client submitted a brief — dimensions, fabric, style references — a designer had to manually produce a visualization. That took hours, sometimes days. In that window, leads went cold. Clients who could not see the product hesitated, second-guessed, or went to a competitor who could show them something faster.

The team had tried stock photos and fabric swatches. Neither solved the core issue: the client wanted to see their specific configuration, not a generic example. Manual rendering did not scale, and the cost of hiring more designers was not justified by the conversion lift.

Approach

We built a text-to-render pipeline using Stable Diffusion. When a client fills out a configuration brief on the Tilda site, the brief is sent to a generation service that produces a rendered preview of the described furniture in a neutral interior setting. The model is conditioned on furniture-specific prompts — sofa types, wood finishes, fabric textures — so the output reads as a product visualization, not a generic AI image.

The generated mockup is returned to the client within minutes, attached to their configuration, and dropped into a Stripe checkout flow. If the client wants changes, they adjust the brief and regenerate. Notion CMS tracks the configuration and the associated mockup, so the production team has a visual reference alongside the spec when the order comes through.

The key design choice was speed over perfection. The mockups are not photorealistic product photography. They are close enough to let the client say "yes, that is what I want" or "no, make it darker." That decision — not "is this render beautiful" — is what the mockup exists to support.

What was built

• A Tilda storefront with a configuration form capturing dimensions, fabric, and style preferences.
• A Stable Diffusion generation service conditioned on furniture-specific prompts, producing rendered previews from text briefs.
• A Notion CMS integration that links each configuration to its generated mockup and routes the spec to the production team.
• A Stripe checkout flow attached directly to the approved configuration, so the client can commit without leaving the page.

Results

In the first month after launch:

• Lead-to-order conversion up 7%. Clients who saw their configuration rendered were more likely to commit.
• 120+ mockups generated in the first month alone — volume that would have required a full-time designer to match manually.
• Average response time to a brief cut to 5 minutes. What used to take hours of designer time now takes the time it takes the model to run.

The conversion lift was real but secondary to the operational shift. The design team stopped being a bottleneck. Mockups that used to queue behind a designer's workload now generate on demand, and the team focuses on the orders that actually close.

Takeaway

Generative image models are most useful in e-commerce when they replace a manual step that does not scale. Suvorin.Mebel did not need better images — they needed faster ones, produced at the volume of their lead flow, tied directly to checkout. Stable Diffusion conditioned on product-specific prompts does that. The mockup is not the product. It is a decision tool that gets the client to yes, and getting to yes faster is what closes custom-order sales.