Celtra · AI Creative Flow

From atom to infinite canvas, reimagining a node-based creative tool to scale advertising production.

TIMELINE

Alpha → Limited Release → GA
Feb - Oct 2026

TOOLS

Figma
Claude Code

MY ROLE

Product Design
Product Research
Usability Testing
Design QA

TEAM

Sr. Product Design Director
6 Software Engineers

OVERVIEW

AI Creative Flow is a node-based creative production tool that helps advertising designers experiment with AI models, connect creative operations, and generate variations of ad content through flexible workflows.

In February 2026, I joined the Product Development team to help ship the product from an early Alpha prototype to Limited Release, then scale to General Availability (GA). I designed core node interactions, workflow connections, canvas interaction patterns, bulk workflow organization.

CONTEXT

What is a node-based creative tool, and why it fits advertising production?

A node-based creative tool is a platform where individual creative operations, such as generating a background image, editing a video, or transforming an asset, are represented as nodes connected into workflows on a canvas.

For advertising designers, this modular approach supports continuous creative exploration, and the workflow supports scalable production. They can swap assets, adjust individual operations, and generate creative variations without rebuilding the entire workflow.

PROBLEM

Designers need a more intuitive way to navigate and manage complex AI creative workflows

In AI-powered advertising production, designers move through a continuous creative journey: bringing in references and prompts as inputs, connecting and experimenting with AI operations, and generating, refining, and managing creative outputs.

As workflows grow more interconnected, designers need the right information and interactions at different levels of the canvas.

Node level: Interacting with individual AI operations

Workflow level: Connecting and refining creative steps

Canvas level: Organizing the broader workflow and reusing creative assets

So the opportunity was, how to create a flexible experience that efficiently supports designers throughout their creative production journey across different zoom levels?

OUTCOME

I brought structure and clarity to AI-powered creative workflows from Alpha to GA

I designed a consistent interaction system that helped advertising designers navigate, connect, and manage creative operations across different levels of canvas zoom.

Throughout the product’s evolution toward GA, our work contributed to a more cohesive experience as the product’s NPS increased by 4 points in 2 quarters.

Before: Alpha Version

After: GA Version

LEVEL 1: NODE ANATOMY

Problem: Node controls were difficult to access and disrupted the canvas

In the Alpha version, node controls were scattered across expandable content areas. Generating creatives and accessing settings required expanding the node, increasing its height and disrupting the surrounding canvas. These extra steps interrupted Generate and Iterate stages of creative journey.

Solution: A consistent node structure with accessible controls

I redesigned the foundational node structure across different node types, separating the header, content, and footer. Frequently used controls sit in the footer, while advanced settings open in a dedicated Property Panel. This keeps interactions predictable and node dimensions stable, allowing designers to focus on their creative content and keep the canvas organized.

LEVEL 2: WORKFLOW CONNECTIONS

Problem: Connecting nodes broke workflow continuity

In the Alpha version, generic input and output handles made it difficult to identify compatible connections. Designers were unsure what types of content each node could accept or how to connect different node types, such as prompts, images, videos, and audio.

Solution: Make connections contextual and discoverable

I introduced color-coded handles to clarify connection types and highlighted compatible handles on downstream nodes during connection. This helps designers identify valid connections at a glance, making it smoother to discover how different nodes can be linked within a workflow.

LEVEL 3: ORGANIZATION AT SCALE

Problem: Grouping lacked flexibility for organizing complex workflows

In the Alpha version, designers could group nodes and run or tidy a selection, but groups offered limited ways to distinguish different workflows. User interviews revealed that designers often created multiple sections or creative variations within a single canvas, making it important to visually identify and organize each group.

Solution: Make groups easier to identify and manage

I redesigned group interactions with customizable names and color coding, allowing designers to label and visually distinguish different workflows. I also refined group level controls, including group/ungroup, tidy, download selection, and run selection, with controls to stop running. This supports bulk generation and more flexible canvas organization as workflows grow.

KEY ITERATIONS

User Feedback

Making the next node easier to find

Based on usability testing with 4 advertising designers, it was difficult to find and add nodes while building workflows, so I introduced a contextual menu that appears when dragging from a node’s output handle. This lets designers discover and add the next node directly on the canvas, without interrupting their workflow to navigate the toolbar.

Engineering Feedback

Turning errors into next steps

In Celtra’s component library, errors appeared as generic messages at the bottom of the canvas, making it difficult to understand what went wrong at a specific node. During an engineering review, we identified new type of failures in AI Creative Flow that designers could fix directly on the node. I introduced more specific error states on node level to help users understand what went wrong and what they could do next.

Cross-team Collaboration

Bringing generated assets across products

Designers needed a way to reuse assets generated in AI Creative Flow within CA Editor, Celtra’s main platform for deep editing advertising creatives. I worked with the PM across teams to design the asset management experience between both products, so generated assets could sync into CA Editor in a read-only format while keeping the experience consistent.

Asset panel in AI Creative Flow

Asset panel in CA Editor

This project is under NDA

Thanks for stopping by!

Machi is open for massages

© 2026 Ivory Zhao

NYC

Thanks for stopping by!

Machi is open for massages

© 2026 Ivory Zhao

NYC