Parsons School of Design, 2022—2023

Raphael: Generate designs using AI

Raphael is my thesis project developed during my graduate degree at Parsons School of Design in New York. I crafted this project under the guidance of Professor Morry Galonoy, an Assistant Professor and Design Strategist.





Project Supervisor

UX Designer (me)

What I did

UX/UI design

Design Validation

UX Research

Conducted Surveys & User Interviews

A design assistant that uses language models

to generate digital images from natural language descriptions and lets you edit them on the go

Raphael aims to combine current design software interfaces such as Illustrator, Photoshop, Figma, etc., with AI tools such as Dall.e. Simplifying and cutting down training time to produce customized graphic design and giving the power to tweak the output to its users.

Raphael is great for people who don’t consider themselves as an artist

but more of a designer and curator. The end goal was to:

Goals for the app

  • Highly reduce the training time of a design software
  • Provide a plethora of editing tools to give users a sense of control
  • The simplest method for those who are in constant pursuit of inspiration to start building and creating

Shortcomings of present AI Tools

To start, I examined existing AI image generation tools like DALL·E and MidJourney, noting their shortcomings. Addressing these limitations could expand the use cases, providing these AI tools with a new audience aligned with business goals and offering users a more powerful and useful tool.

  • The platforms provide no control over the result. Once the image is generated, it can't be edited.
  • A significant number of results are inaccurate, leading to low user satisfaction with the generated content.
  • At times, users may want to retain a portion of the generated image but are unable to do so. Generating a new image from scratch becomes the only option.

I interviewed 2 users

Primary Research

Interviewed two users each with different levels of design backgrounds and different professional backgrounds in order to determine unique use cases.

User with no experience with Design Tools:

Mary is a creative and social mid-50s woman with two children and an engineer husband.

She enjoys fashion and interior design and frequently uses her mobile phone for messaging, video calls, and creating digital wishing cards or video clips for special occasions.

However, Mary finds image editing apps limited and wishes for an easier and more reliable way to make personalized messages with more control over fonts and colors.

Examples of graphics Mary creates to wish on special occasions

User with prior experience with Design Tools:

George is a mid-30s corporate UX Manager with a busy schedule of meetings. He leads a team of 40 and offers daily personal sessions to improve their work.

He presents bi-weekly progress reports to company leadership and often shares design thinking methodologies. However, George finds it time-consuming to create graphics for his presentations using design software.

He usually has to get someone from his team to assist in making the presentation, slowing down the productivity of their main projects.

Examples of graphics George creates for his corporate presentations

Secondary Research

A research text by Evelina Leivada, Elliot Murphy, Gary Marcus influenced my project.

DALL.E 2 Fails to Reliably Capture Common Syntactic Processes

Analysis of the ability of Dall.e 2 to capture various grammatical phenomena pertaining to compositionality that are widely discussed in linguistics and pervasive human language.

DALL.e is unable to reliably infer meanings that are consistent with the syntax of the prompts

The paper tells that Dall.e’s accuracy rate is too low. So I aim to develop an interface around those weeknesses

Wireframing & Prototyping

To start, I borrowed elements of the interface from other AI tools like DALL·E and contemplated how I could expand upon them. I created wireframes and iterated over them through constant user testing.

My initial goal was to create an interface for people who were inexperienced with design tools as well as people who were great at use design software. My initial precedents were programs like Dall.e, Midjourney, Canva, Illustrator, Instagram Image editing etc.

After constant testing with users

of different backgrounds and experienced I decided to set my audience to people who have less design experience and market the app as mass consumer app.

Final Design

The experience starts with a simple home page featuring a text input panel for prompts. Users can also attach media to obtain a customized result.

The homepage also includes additional controls for enhanced personalization.

After the user inputs their prompt and taps the Generate button, the app displays four generated designs. Users can download the design immediately or choose to further edit them.

Users can use the image editor to move certain elements in the image or generate alternates in their place and create their desired design.


After completion, I presented my project in front of a jury of designers from major tech companies. I received a positive response and an A+ from my supervisor.

I had the pleasure of having Archit as a student at Parsons School of Design where he excelled in the class. Archit is creative, intellectually curious, disciplined, and self-motivated. He is both an analytical and creative thinker. In addition he was collaborated with other students and provided feedback to his classmates in formal critical feedback sessions as well as working sessions in a manner consistent with our best faculty.

Archit Saxena is in my Creative Practice Seminar at Parsons MFA Design & Tech program -- he is an innovative thinker and an excellent student. His projects are well-executed and grounded in research. Archit is a terrific writer and an asset to our program.

Other Cases

Optum Virtual Care: designing a Telehealth app for 28M+ users

B2C · web & mobile(ios & android) · end-to-end design


Clinical Transformation Tool: increasing clinicians’ efficiency by 90%

SaaS enterprise tool · UX research · user testing