Bringing AI Capabilities Directly onto the Canvas Stage
Native AI embeds AI-powered content and design capabilities directly on the canvas as part of the manual editing flow, eliminating chat context-switching and enabling frictionless editing for self-creators.
Project Timeline
4 months
Team
Sole Product Designer, Product Manager, 3 Developers, Wix Media Team, AI Platform Team, Marketing Designers, Brand Team
Role Played
Lead Product Designer — Spearheaded end-to-end UX/UI, conducted domain research and pattern classification, drove iterative concept generation, built interactive prototypes using Figma Make, defined component specifications, and contributed the Composer Bottom Sheet Widget to the core Design System platform.
Impact
Native AI increased core AI image editing adoption by +50.7%, with +343% growth in background removal and +984% in background replacement.
Premium remained stable: purchases moved from 3.42% to 3.44%, while service collections shifted by −3.5%; neither change was significant.
AI engagement rose +4.6%, and 38.8% of exposed users triggered at least one action.
How do we utilise AI for this project?
To match the speed and innovation of the product, we integrated AI directly into our design and prototyping methodology: Accelerated Interactive Prototyping (Figma Make): Instead of relying on static wireframes or slow code handoffs, we leveraged Figma Make to transform high-level concepts into interactive, high-fidelity prototypes. This gave stakeholders a realistic "product feel," enabling rapid testing and instant iteration without developer bottlenecks. Streamlined Alignment: Utilizing AI tools allowed us to rapidly map visual variations, test edge-case content, and align across marketing, brand, and AI platform teams early in the lifecycle.
AI capabilities in the editor were historically locked inside a standalone conversational chat surface (Aria), breaking focus and requiring manual prompts for routine edits. Native AI transforms this experience by embedding context-aware AI tools directly onto the editing stage across text, images, sections, and headers. By offering direct, one-click actions, guided preview suggestions, and lightweight prompt widgets directly on selected elements, Native AI brings AI directly into the manual editing flow—keeping users focused and in control.
Single click Aria Action - Improve Text
03
Problem Statement
Self-creators building sites in the Wix Harmony Editor faced significant friction when trying to leverage AI:
Context Interruption
Forcing users to leave the canvas and switch to a separate chat surface disrupts their focus and breaks their creative momentum.
Prompting Fatigue & Friction
Simple, repeatable tasks (e.g., tone adjustments, minor rewrites, background removal) required writing free-form prompts, creating high cognitive load and uncertainty.
Lack of Direct Stage Tools
Everyday micro-adjustments lacked direct-manipulation triggers on the canvas, rendering AI tools slower than manual tweaks for quick edits.
Users struggle to move quickly and make progress when relying solely on chat-based requests. Tasks that require heavy tool-orchestration calls are especially costly in time and effort
04
Project Goals
Deliver Direct Canvas Execution
Enable fast AI actions with immediate, visible visual results directly on the editor stage.
Build a Scalable Framework
Develop a generic, reusable action framework that adjacent product teams can easily plug into.
Eliminate Prompting Barriers
Replace mandatory free-form prompting with structured, one-click actions for common editing tasks.
Ensure Transparency & Control
Maintain clear labels and predictable outcomes so users feel confident before triggering actions.
Maximize Efficiency
Reduce clicks, typing, and mode switches for self-creators.
A new approach to using Aria: invoking it directly on stage through a direct API call for faster results.
05
Target Audience
Primary Users
Self-creators and DIY users building and editing their own websites in the Wix Harmony Editor.
Characteristics
Non-professional site builders who prioritize speed, clarity, and low-effort tools over conversation or manual prompt engineering.
06
Discovery Research
Understanding the Product Area & Scope
Research showed users viewed AI as "distant" and ill-suited for micro-edits because it lived in a separate panel. Native AI redefines the editing surface by making AI capabilities an extension of core element selection.
Competitive & Ecosystem Analysis
We analyzed market implementations across seven distinct domains:
Site Builders
Design Tools
Content Editing Tools
Vibe Coding / Dev Tools
LLM Canvas Tools
Browsers
Wix Internal Tools
This landscape mapping allowed us to classify recurring patterns and establish a clear baseline for stage-based interactions.
Key Insights
Direct element selection and immediate execution beat conversational interfaces for in-context edits.
Structured visual options cure "blank input syndrome" far better than empty text boxes.
Product Research Deck
07
Ideation
Concept Exploration & Iteration
We navigated three major review checkpoints (V1, V2, and V3) to refine our approach to stage-level AI actions:
Concept 1: Expose AI Action
Selecting an element reveals a spread of highlighted, relevant actions beneath it: some execute in one click, some open a panel for choices, and others transform the floating UI into an input field to refine intent—inspired by Google Slides and GPT Canvas.
Concept 2: Intent-Based Flow
A floating action bar that clearly separates interaction types.
Concept 3: Input First
Start by describing intent, then receive smart suggestions beneath the input—inspired by Relume and Studio.design.
Concept 4: Bottom AI Action Bar
A secondary bottom action bar with relevant AI actions that can switch between an options list or a mini input field with suggestions—inspired by Figma.
Selection & Validation
After cross-team alignment, we selected the final concept combining direct stage triggers with a focused prompt composer. Using our Figma Make prototype, we validated the full flow before moving into detailed production design.
08
Design
We translated our validated direction into robust UI foundations on the canvas:
Floating Action Bar Experience
An on-stage toolbar offering element-tailored, one-click triggers (e.g., Improve Text, Change Tone, Redesign Section, Auto Enhance).
Composer Bottom Sheet Widget
Designed a novel widget bridging the action bar and the AI chat. Placed directly above the input field, it presents suggested visual options to guide intent and overcome "blank input syndrome."
Cross-Team Alignment
Partnered with Brand, Marketing, and AI Platform teams to standardize visual guidance popovers and unify the overall visual language.
Inline Ask Aria Floating Input Button
A lightweight floating input that lets users ask Aria for contextual changes without leaving the canvas.
Image Actions Popover Helpers
Contextual helpers for image actions.
09
Speccing & Design System Contribution
To ensure scalability across multiple internal teams adopting this framework:
Action Bar Specification
Established strict structural rules, layout variations, accessibility (a11y) standards, and tab-order rules for seamlessly integrating new AI actions into the floating bar.
Design System Contribution
As the Composer Bottom Sheet Widget was completely new to our platform, I built and standardized this component, contributing it directly back to the global Design System for platform-wide adoption.
10
Key Impact & Metrics
Core AI Adoption (+50.7% Lift)
The core KPI—Image edited with AI—grew from 9.4% to 14.1% (2,724 → 4,082 users), proving strong and durable adoption.
Success rates across core modes increased to 90.7%–95.8%, with user retention ("stay rate") of AI edits holding between 88%–98%. The newly introduced edit_with_ai mode achieved a 95.9% success rate.
Additive Engagement (No Cannibalization)
Floating bar exposure reached 43.4% of users. Overall AI Assistant engagement grew (+4.6% messages sent), confirming Native AI generated new activity rather than cannibalizing existing tools. Furthermore, 68.8% of users clicking "Ask Aria" submitted detailed free-form prompts.
Editor & Business Stability
Core Editor KPIs (Component Selection, Add, Attach, Save, Publish) and Premium conversion (+2.0% non-significant variation) remained completely stable, with AI compute costs scaling predictably and linearly with growth.