Skip to content

Case studies being refinedA few case studies are still being refined.

← All work

Product Design · Generative AI · UX/UI

Making creative AI easier to navigate.

Pandorra.ai brings image, video, audio and other AI-creation capabilities into one product. My contribution focuses on the interface patterns and journeys that help users find tools, make choices and create with less friction.

Client / product
Pandorra.ai
Role
Product Designer / UX/UI Designer (freelance collaboration)
Status
Ongoing
Platform
Web application
Collaboration
Product, design and development teams
Scope
Selected product and interface design work; not sole ownership of the entire product

01 / Product design

Pandorra.ai

AI creative platform

01 / The context

Powerful capabilities create complicated choices.

A multi-tool AI platform multiplies entry points, creation modes, settings and result types. People need to know what a tool does, which workflow to choose and what to expect after acting. I read the experience as a sequence of decisions:

The sequence of decisions

  1. Discover a tool
  2. Understand its requirements
  3. Create
  4. Review the output
  5. Decide what to do next

02 / The live product

One entry point for many tools.

Pandorra.ai is live and keeps evolving: the interface at pandorra.ai is the reference, and every screen on this page carries its date. In August 2026, Explore presented the creation tools side by side: a sidebar for tools and account actions, a featured banner, the tool entries and a gallery of scene presets.

Pandorra.ai Explore screen: sidebar with Explore, Profile, Affiliate and the AI resources, a presentation banner, five tool cards and a grid of scene presets
Live product · capture of August 2026Explore as it was live in August 2026 (pandorra.ai/explore, imported into the design file); the account avatar is masked. Today’s interface may differ.

03 / The design question

How might we make the next useful action obvious?

Nobody should have to guess whether a button starts a free trial, a generation, an example or a download. Clear labels, working feedback and predictable transitions matter as much as the layout.

04 / Review and decisions

Four observations, four decisions.

A review of the Explore screen as it was live in August 2026 recorded where the interface was unclear. Each observation is tied to a marker on that capture and to the decision it led to; all four are recommendations, still to be checked against today’s interface.

Pandorra.ai Explore screen: sidebar with Explore, Profile, Affiliate and the AI resources, a presentation banner, five tool cards and a grid of scene presets
  1. Navigation between toolsOn mobile, some destinations of this sidebar were unreachable without a collapsed menu.
  2. Tool entries and their actionsEach card starts a different workflow; the review asked for consistent states, hints and actions.
  3. Mixed-language labelsA French interface where the AI resources keep English names.
  4. Calls to action“Créer gratuitement” promised free access instead of naming the action; upgrade prompts sit next to the tools.
Observations from the August 2026 reviewWhere the observations sit on the Explore screen.
  • Calls to action

    Problem
    “Créer gratuitement” promised free access instead of naming the action; “Voir des exemples” did nothing.
    Design decision
    Action-led labels: “Commencer à créer”.
    Visual evidence
    Marker 4.
    Outcome · status
    Recommendation, August 2026 review · to check against today’s interface.
  • Navigation at small widths

    Problem
    Some sidebar destinations were unreachable on mobile.
    UX principle
    Recognition rather than recall  (opens in a new tab) · Nielsen Norman GroupThe recommendation follows this principle: destinations stay visible, not remembered.
    Design decision
    A compact menu with the same destinations as desktop.
    Visual evidence
    Marker 1.
    Outcome · status
    Recommendation, August 2026 review · to check against today’s interface.
  • After generation

    Problem
    No preview without download; failed generations had no removal action.
    UX principle
    Visibility of system status  (opens in a new tab) · Nielsen Norman GroupThe composer states are the feedback the generator owes: what it is doing, what happened.
    Design decision
    Preview before download, a readable state, a removal action for failures; the composer states are documented in the design system.
    Visual evidence
    Marker 2 · prompt composer, section 05.
    Outcome · status
    Recommendation, August 2026 review · states documented in the design system v1.0 · to check against today’s interface.
  • Language and consistency

    Problem
    A French interface with English names on the AI resources.
    UX principle
    Consistency and standards  (opens in a new tab) · Nielsen Norman GroupThe recommendation follows this principle: one action, one word, across tools and languages.
    Design decision
    One language per interface.
    Visual evidence
    Marker 3.
    Outcome · status
    Recommendation, August 2026 review · to check against today’s interface.
Explore redesign, final direction: sidebar with Create, Library and Resources groups, headline “Create content that sells with AI”, the prompt composer at the centre and five prompt suggestions
Earlier design exploration · Explore, final directionFinal direction of the Explore exploration, drawn after the August 2026 capture: the generator at the centre, a navigation-first sidebar, prompt presets. Design file, not the live interface.

05 / A system for a multi-tool product

One system, many tools.

The design system documents the foundations, buttons, navigation, inputs, the prompt composer and the feedback states behind the Explore work, drawn in 2026 after the August capture. Design-file documentation: not the live interface, not a measure of adoption.

Pandorra.ai design system, foundations: colour tokens from background to info, typography samples, spacing scale, radius and effects
Earlier design exploration · design system v1.0Colour tokens, typography and spacing of the dark-mode system (v1.0 · Explore).
Pandorra.ai design system, prompt composer: six states of the generator card, from empty to generating and error
Earlier design exploration · prompt composerThe generator states, reused across Image, Video, Avatar and Audio: empty, filled, focused, generating, error, disabled.

06 / Onboarding and the first creation

From welcome to the first image in three steps.

A freemium onboarding explored in 2026, in versions v1 to v1.3 (v1.3 follows the July 2026 specification): welcome, intended use, first image, result, plan. Design proposals with developer notes, not confirmed live.

Onboarding step 1, welcome: headline “Create your first image in a few seconds”, three promises and the button “Create my first image”
Earlier design exploration · Freemium v1.3Step 1 · Welcome: one sentence, no editing, three free images.
Onboarding step 2, intended use: four cards (social networks, B2B marketing, e-commerce, training) with B2B marketing selected
Earlier design exploration · Freemium v1.3Step 2 · Intended use: four contexts that personalise the presets.
Onboarding step 3, first creation: the prompt composer with an example prompt, the generate button and five preset chips
Earlier design exploration · Freemium v1.3Step 3 · First image: the prompt composer with presets as starting points.

Design proposals from the design file, 2026 · not confirmed in production · no usage data.

07

Collaboration.

I work at the intersection of product thinking, visual design and technical feasibility, within a wider team; development, product and growth results belong to that team. The changes visible in the live product are the work of the team as a whole and are not all mine.

08

Outcome and evidence.

Status

A documented review (August 2026) with recommendations for clearer copy, mobile navigation and generation workflows; the deployment status and measured impact of each recommendation remain to be confirmed against today’s interface.

09

What I learned.

Designing an AI product is not about exposing more features; it is about making the system’s behaviour understandable: which action is available, what it does, what state the product is in, what comes next.