How It Works

Sozdai isn't a "magic designer". Under the hood — a couple of neural networks, each doing a specific operation. Knowing what actually happens saves credits and helps you get the right result faster.

Two phases: analysis and generation

Analysis — the first thing the neural network does with an uploaded product. It looks at photos and text (if present) and fills in structured data: name, category, key statements, specifications. All of it is visible on the product page right after analysis and becomes the base for generation. This phase is free.

Generation — building the actual card. AI takes as input: product photo + structured data from analysis + your settings (template / reference / instructions). Outputs a finished image. This phase costs credits (2 💎 for Flash, 4 💎 for Pro).

Two modes: Flash and Pro

Different models — different quality.

• Flash — fast model, result in 15-25 seconds. Right for most photos; it is the default.
• Pro — heavier model, 30-45 seconds. Better detail, especially on small text, hands, textures.

The model is the first field under “Parameters” (generation settings); set your default in your profile.

Refine — how AI edits its own work

Once the first generation is ready, instead of starting over, ask AI to tweak it. Mechanism: your text request ("lighter background", "remove the bottom text") goes together with the current image into a new generation. Output — a new image that differs from the previous one in the direction you specified.

Cost — 2 💎 (Flash).

Why not everything works on the first try

The neural network isn't a human designer. It doesn't understand the business context "this card will be seen by a Wildberries shopper looking for kids' sneakers". So:

• Give a clear template (wizard Step 1) — AI knows the format.
• Give short, clear statements (wizard Step 2) — AI places them on the card.
• Critique precisely (refine), not descriptively. "Make it nicer" is too vague. "Move the QR code to the bottom-left corner" is clear.