How the AI works
Built AI-native, and honest about it.
1
Understand the product and the shelf
- in:
- Your one line, mood, shelf, claims, and brand kit voice
- out:
- A structured brief
The language layer reads the product like a packaging strategist would: what it is, who it is for, what the category looks like on that shelf, and which way to stand out. The brief is internal; you see its results in the rationale attached to every direction.
2
Write the copy
- in:
- The brief
- out:
- product_name, flavor_line, front_claims, back_story, palette_hint, motif, rationale
A language model writes the name lockup, flavor line, and back story, and suggests a palette and motif. Every reply is validated with a strict schema. Front claims are then replaced on our server with exactly the claims you typed, and any sentence that introduces a regulated term you did not supply is removed.
3
Generate label directions
- in:
- Copy, palette, motif, mood, the format's wrap proportion
- out:
- Flat label art, no text
The current OpenAI image model (gpt-image-2) paints background art for the wrap: shapes, illustration, texture, in your palette. We ask for art only, never lettering, logos, or trademarks. If generation is unavailable or slow, a procedural engine draws a typographic label from the same copy and palette, so a run never comes back empty.
4
Set the type and wrap it on true-size packaging
- in:
- Art plus copy plus format dieline
- out:
- A label texture on a live 3D can, carton, or pouch, and print files
Our typesetting engine composes the front panel, sides, back story, and a marked regulated area at the format's exact proportions, then wraps the result onto procedurally modeled packaging built from real millimeters. The same engine renders the 300 DPI print file with bleed, so what you approve in 3D is what you export.
Model-agnostic
The copy layer routes to OpenAI or Anthropic language models in a configured order and accepts whichever answers first with valid output. Swapping models changes nothing about your results format. Agency projects run on the priority tier.
Typography is ours
Image models are good at shapes and texture and unreliable at spelling. So the art never contains text: we set every word from your copy, in real typefaces, at print resolution. That is also what lets you edit copy without regenerating art.
What we never do
Invent claims, write nutrition facts or allergen statements, imitate real brands or trademarks, or train models on your brand kits and projects. Read the Care page for the full stance.