TL;DR
Showing a designer 10 written packaging directions beats handing them a blank brief. Feed a text model your product, three brand adjectives, one reference brand, and your real per-unit budget, then ask for 10 distinct concepts with color palettes, structures, and trade-offs. Use a text model (Claude Sonnet 4.6, GPT-5.5, or Gemini 3.1 Pro) for the thinking, and only after the direction is locked move to an image model for mood boards. Keep your budget honest: a printed folding carton runs about $1.00 to $3.00 per unit at 1,000 to 3,000 units, while a magnetic rigid box is $12 to $25+, so a concept that ignores cost is a concept you cannot ship.
The task
You’re shipping a physical product (or a digital “box” like an app onboarding) and need a packaging direction before talking to a designer. Ten written concepts cost you ten minutes and a clear head; a blank-page kickoff costs the designer a week of guessing and you a stack of revisions.
This use case covers consumer packaged goods, indie product launches, gift kits, Etsy bundles, subscription-box inserts, and even SaaS onboarding “boxes” where the metaphor carries the first impression.
Where AI actually helps (and where it does not)
Use a text model for the divergent phase: getting from zero to ten distinct directions in one sitting. Models are strong at remixing a brand vibe with structural archetypes (tube, sleeve, magnetic flap, kraft pouch) and proposing named color palettes with hex codes. They also work as a translator. Describe your brand in plain language and the model returns references a designer recognizes (“Aesop-style apothecary,” “Tony’s Chocolonely loud color blocks”).
What AI cannot judge: real-world print viability, shelf contrast against your actual competitors, and unit economics. A model does not know your category’s current shelf, and it will happily suggest a six-color soft-touch box that blows your budget. Treat every cost and material claim as a hypothesis you verify with a printer quote, and treat AI-generated renders as mood references, never production art.
Here is the honest division of labor:
| Stage | Best tool type | Example (as of June 2026) | What it produces |
|---|---|---|---|
| 10 written directions | Reasoning text model | Claude Sonnet 4.6, GPT-5.5, Gemini 3.1 Pro | Concepts, palettes, structures, trade-offs |
| Mood-board visuals with on-box text | Text-accurate image model | GPT Image 2 in ChatGPT, Gemini 3.1 Pro image (Nano Banana Pro) | Rough labels, headlines, color comps |
| Stylized hero shots | Aesthetic image model | Midjourney v7 | Vibe references (poor at legible text) |
| Dieline + print files | Structural/vector tool | Pacdora, Adobe Illustrator | CMYK production art and cut lines |
For mockups where on-box text must be legible, image models that render type cleanly win. As of June 2026, GPT Image 2 (inside ChatGPT) and Gemini 3.1 Pro’s image mode hit roughly 85 to 90 percent first-try accuracy on short label strings, while Midjourney v7 still lands closer to 40 to 60 percent. Use Midjourney for the feel, not the words.
What to feed the AI
- Product: what it is, dimensions, weight, fragility
- Brand vibe: 3 adjectives plus one reference brand you admire
- Buyer and context: who picks it up, and where (a Whole Foods endcap, an online unboxing, a gift table)
- Constraints: budget per unit, sustainability requirements, regulatory text you must carry
- What you already considered and rejected
The “rejected” list is the highest-leverage field. Without it the model circles back to the obvious flat box.
Copy-ready prompt
Paste this into Claude, ChatGPT, or Gemini and fill the bracketed fields. The bracket placeholders are for you to replace, not literal text to send.
You are a senior packaging strategist. Generate 10 distinct packaging
directions for the product below.
Product: [product_description]
Brand vibe: [three_adjectives] (reference: [reference_brand])
Buyer + context: [buyer_and_where_they_see_it]
Constraints: budget [budget_per_unit], [sustainability], [regulatory]
Already rejected: [list]
For each direction provide:
- One-line concept name
- 2-sentence story (what the buyer feels picking it up)
- Color palette (3-5 hex codes with names)
- Structural idea (form factor + material)
- Hero typography style (serif / sans / display + a reference typeface)
- One risk or trade-off, including a cost note if it pushes past my budget
Make the 10 directions genuinely different in structure and story, not the
same box in different colors. End with a 2-sentence "which 3 I'd test first"
recommendation, weighted toward shelf differentiation and my stated budget.The closing “which 3 to test first” line forces the model to prioritize instead of dumping ten equal options.
How to check the output
Score each concept on three axes and drop anything weak on two of three:
- Shelf differentiation against your top 3 named competitors.
- Manufacturability at your stated budget. Sanity-check against real ranges below.
- Emotional fit with your brand and buyer.
Use these June 2026 per-unit benchmarks as a reality filter (always confirm with a printer quote; materials alone are typically 45 to 60 percent of the cost):
| Format | Typical volume | Per-unit cost |
|---|---|---|
| Corrugated mailer | low volume | $0.40 to $1.20 |
| Printed folding carton | 1,000 to 3,000 | $1.00 to $3.00 |
| Custom rigid box | 500 to 2,000 | $4 to $12 |
| Magnetic / two-piece rigid box | 500 to 2,000 | $12 to $25+ |
Watch minimum order quantity too: most manufacturers set a floor of 500 to 1,000 units, and true wholesale pricing usually starts at 2,500 to 5,000. Tooling spreads out fast: an $800 die at 1,000 units is $0.80 each; at 10,000 units it is $0.08. If a concept only works at 10,000 units and you are launching 800, it is not a real option yet.
Then print or assemble the top 3 mood boards (manually, using the AI palettes and references) and walk them past 5 target buyers before committing budget.
Common mistakes
- Ten superficially different concepts that are the same idea in different colors. Fix with the “genuinely different in structure” instruction and the rejected list.
- No shelf or unboxing context, so the model defaults to flat boxes.
- Skipping budget and sustainability constraints, then getting concepts you cannot manufacture.
- Treating AI palettes or renders as final. Always print a physical sample and check it in CMYK, not on a backlit screen.
- Asking for “creative” directions without naming a reference brand, which makes the model average everything into beige.
Next steps
Save the top 3 directions and feed each back to the model for a deeper pass: a typography study, the actual copy on the box, and 3 cap or sleeve variants. Generate mood-board visuals with an image model that renders text well, then convert to a real dieline in a structural tool like Pacdora or Adobe Illustrator before handing the designer a concrete brief. For the visual prompts themselves, our product image prompt guide and product visual direction guide carry the templates.
FAQ
Can AI design the production artwork? No. Treat AI visuals as inspiration only. A human designer working in a vector tool produces the CMYK print files and dieline. AI renders are RGB and structurally inaccurate.
Which model should I use for the 10 directions? Any current reasoning model handles this well. As of June 2026, Claude Sonnet 4.6, GPT-5.5, and Gemini 3.1 Pro all carry roughly a 1M-token context, so you can paste a full brand brief and competitor notes without trimming.
How specific should the brand vibe be? Three adjectives plus one reference brand is the sweet spot. More than that and the model starts averaging instead of committing to a direction.
What about regulatory text? Tell the model every region you sell in. It will flag where nutrition panels, allergen statements, or recycling marks must appear, but confirm the exact rules with your local authority before print.
What budget number should I give it? Your real per-unit ceiling. If you are not sure, anchor to the table above: a printed folding carton at $1 to $3 per unit covers most indie launches, and rigid magnetic boxes start around $12.
Related
Pair the concept work with a sharp product description prompt, align the brand layer using brand positioning statement prompts, and clarify the strategic angle with product positioning prompts.
Tags: #AI writing #Workflow