Rare Wing

Most designers use AI passively — generating an image here, fixing some code there. Very few can articulate a complete workflow. This case study addresses that gap: every stage of a real design production, from brief to deployed page, with AI embedded at each step.

The fictional client is Rare Wing, a premium small-group birding expedition company. The brief: design and build their new season launch campaign — brand identity, landing page, email, and social. The point is not the output. It's demonstrating a systematic understanding of where AI fits, where it falls short, and where human judgment remains irreplaceable.

The brief was developed collaboratively with Claude, starting from a loose concept: a premium travel brand built around birding expeditions to extraordinary destinations. Claude was used to develop the full creative brief — brand positioning, audience segmentation, destination rationale, add-on package logic, and campaign objectives.

What would typically require a briefing meeting, a strategist, and a few rounds of revision was done in a single session. Critically, the interesting decisions — destination selection, brand name, tone direction — were made by me. AI structured and expanded them.

Time saved: approximately 3–4 hours of strategy and briefing work.

The brand needed to sit between adventure and luxury — more rigorous than a resort holiday, more refined than a trekking expedition. The target audience spends serious money on Swarovski optics and will also sleep in a field hide for three days to see one species. Condé Nast luxury was the wrong register. National Geographic authority was closer.

The palette — moss green, off-white, warm gold — was chosen to function as a near-neutral that doesn't compete with the destination imagery. Hermann (W Type Foundry) was selected as the display serif for its literary character; Suprema (Artegra) provides clean geometric contrast for UI and body copy.

The brand name Rare Wing was validated against existing trademarks and checked for negative cultural associations. No conflicts found. The Polynesian association of terns with safe voyaging is essentially perfect for the brand.

Time saved: approximately 2–3 hours of naming research, trademark checking, and brand rationale work.

The logo mark was developed through iterative prompt refinement in Adobe Firefly across four visual directions and multiple bird species. The key insight: species selection determines mark quality. Generic "bird in flight" prompts produce generic marks. Specifying ecologically meaningful species — Arctic Tern, Bar-tailed Godwit, Albatross — produced marks with built-in narrative.

The Arctic Tern was selected for its distinctive silhouette, its status as the world's longest-distance migrant (a natural metaphor for a travel brand), and its natural two-tone colouring — dark cap, pale body — which maps directly to the gold and moss green palette. The brand safety check confirmed the tern carries universally positive symbolism across cultures: endurance, long journeys, safe return.

Firefly provided rapid direction exploration. The production-ready vector was completed in Illustrator. A detail that emerged through the designer's eye: the gold dot replacing the tittle on the 'i' in Wing, connecting the letterform to the bird's gold cap. AI got us to the right direction. The detail that makes it memorable came from craft.

Time saved: direction exploration across four concepts and multiple species compressed from 1–2 days of sketching to an afternoon of prompting.

Five destination hero images were generated across three AI models — Gemini (via Firefly partner models), GPT Image, and Firefly native — with honest evaluation of each output.

The findings were nuanced. Wide landscape and atmospheric shots: AI excels. The Wandering Albatross skimming Southern Ocean wave troughs (Gemini) is indistinguishable from commissioned wildlife photography — an image that would have cost thousands to commission five years ago. Species-specific shots in open environments: strong. Dense forest environments: all models struggled with habitat realism, though species anatomy remained accurate enough for use.

Model selection mattered. GPT Image was strongest for scenes with figures (birder observing a Jabiru Stork — importantly, looking at the bird). Gemini was strongest for atmospheric landscape shots. Firefly native models consistently failed on bird anatomy. One Gemini image was technically beautiful but showed the birder looking away from the subject — correctly rejected. Evaluating whether an image makes narrative sense for the brief required human judgment no model could provide.

For niche wildlife subjects like the Andean Cock-of-the-rock or Raggiana Bird-of-paradise, AI generation outperformed Adobe Stock's AI-generated stock library, which had limited and lower-quality coverage for specialist wildlife. The practical conclusion: use AI generation for hero images where AI stock falls short; source traditional photography only where neither delivers the required quality.

Time saved: five production-ready hero images across an afternoon. The equivalent commissioned wildlife photography shoot would require weeks of planning, travel, and significant budget.

The landing page was built in HTML and CSS with a custom font stack loaded via @font-face. The hero section randomly serves one of three background images on each page load — the same mechanic used in production for segment-driven personalisation, where audience data determines which destination is featured. For the portfolio it demonstrates the thinking without requiring a back-end.

South Georgia is featured full-width at the top of the destination grid as the new flagship offering, with the other four destinations below in a structured 2+1 / 1+2 layout. Each card carries destination imagery, species highlights, pricing, and a CTA.

Production checks: SEO meta tags, Open Graph, canonical URL, accessibility (skip nav, ARIA landmarks, role attributes), image optimisation (all assets converted from PNG to WebP — 97% file size reduction), and a CSS audit confirming no unused classes or font weights.

Time saved: a production-ready page built in a single session. Traditional front-end workflow: 2–3 days minimum.

The new season launch email was designed and built in HTML and CSS, with personalisation logic applied per audience segment. Subject lines, body copy, and destination hero images vary by segment — past tropical guests see different content to past expedition guests or first-time enquirers. Claude drafted the copy variants; each was edited and approved before use.

A/B variants for subject lines and hero images are a single prompt away — the same workflow that produced the master campaign generates as many test variants as needed, with no meaningful added time.

One honest finding from this stage: Claude cannot write production email HTML reliably from scratch. The structure of email HTML — with its table-based layout, inline styles, and client-specific hacks — is too far outside what AI models handle consistently. The approach that worked was providing a fully built, tested template as a reference, then having Claude rebuild and adapt it to the new design. The output was not tested in Litmus or Email on Acid — that step would be required before any live send.

Time saved: HTML build and copy variants completed in half a day, versus a full day of development plus briefing a copywriter for the segment variants.

A single social tile extending the brand to Instagram and LinkedIn — demonstrating how the visual system scales from a full-page layout to a constrained square format, and how the AI-generated imagery adapts to different aspect ratios. Variant testing is straightforward: swap the destination image, adjust the copy prompt, generate an alternative in seconds.

Time saved: under an hour. The system was already built — this was application, not creation.

AI compressed what would have been days of work into a single session: brand strategy, name validation, species research, logo direction exploration, image generation across multiple models, and a production-ready landing page. The workflow is faster, more iterative, and — when used well — produces better-considered output than working without it.

Copy direction runs through the whole workflow, not just the email stage. Subject lines, destination descriptions, CTAs, campaign headlines — all drafted by AI, all edited and approved before use. The collaboration makes ongoing copy work fast: refreshing a campaign, testing a new angle, optimising underperforming subject lines. What used to mean a copywriter briefing and a round of revisions is now a prompt and a quick edit.

The same applies to the broader campaign. Edits, updates, and optimisations that would previously have meant reopening a brief are now a natural part of the conversation. The AI holds the context; the designer makes the calls.

But AI didn't make the decisions that matter. It didn't choose the Arctic Tern. It didn't notice that a Gemini image showed a birder ignoring the bird. It didn't put a gold dot on the 'i'. Those were human calls — and they're the ones the brand lives or dies by.

The honest conclusion is not that AI replaces designers. It's that designers who understand how to use AI will replace designers who don't.