Google Stitch 2.0: A Fully Free AI Coding Agent from Google Just Dropped

Hello everyone. Google has released a major update 

to Stitch. It's AI powered tool for UI design and front-end code generation. Stitch lets you create 

complete interface designs and productionready code using simple text prompts, sketches, 

or screenshots. And it's currently available for free. You describe what you want and Stitch 

generates responsive layouts that can be exported directly to Figma, sent to Google's AI builder 

or downloaded locally as code. In this video, I'll break down what's new in this update and how 

it impacts realworld development workflows. First, choose whether you're building a mobile app or a 

web UI. As you can see, Stitch now includes two new options. The first two options were covered in 

the earlier video. With fast mode, you can export designs directly to Figma. Thinking with Gemini 

3.0 Pro, which focuses on reasoning and quality, and redesign powered by Nano Banana Pro. Select 

Gemini 3.0 Pro. Enter your prompt and submit it. Generation starts immediately, and you can switch 

between modes within the same project. something that wasn't possible before. Here you can see 

the SAS landing page we generated and the quality comes from using Gemini 3 Pro. You can also view 

the generated code. By default, Stitch outputs HTML with Tailwind CSS and you can download the 

full source folder including the UI screenshot and HTML files. This is the generated HTML running 

in the browser which highlights how well Gemini 3 performs for front-end code inside Stitch. Stitch 

now includes a new variation generation panel. You can start by choosing how many design options you 

want to generate. Next, set the creative range, refine, medium, or yolo, which controls how 

conservative or experimental the output is. You can also add custom instructions like 

enforcing a specific style or layout direction. Below that, you can decide which aspects to 

vary such as layout, color scheme, images, fonts, or text content. Once everything is set, 

clicking generate variations produces multiple design alternatives within the same project. Once 

the design is ready, you can export it in multiple ways. You can send it everything directly to 

Google AI Studio, export the files as a zip, copy the code to your clipboard, or hand it off 

to Jules. When you choose AI Studio, Stitch opens it automatically and downloads the HTML and 

image assets in the background. This makes it easy to move from design to actual implementation 

without any manual setup. Another export option here is Jewels. Jewels is Google's autonomous 

coding assistant that takes these generated screens and helps turn them into a working 

project. You can connect it to a repository, add a short description of what you want to build 

and let Jules handle the implementation steps. This is useful if you want to move beyond static 

UI and start wiring things into a real codebase. When you choose build with AI Studio, you can 

add an additional description and continue from there. Stitch automatically passes the source 

context, including the HTML and UI screenshots and redirects everything into AI Studio. From 

there, you can extend the project further. AI Studio converts the HTML into a structured 

React plus TypeScript codebase instead of a single flat HTML file. This results in a more 

professional folder structure and in some cases it even enhances the UI, for example, automatically 

adding a yearly pricing option. Once the landing page is ready, you can continue building out the 

rest of the application from that foundation. With the new prototypes feature, you can 

stitch multiple screens together into an interactive flow. You can click 

through screens, test transitions, and validate the overall user journey before 

moving to development. This makes it easier to catch UX issues early without exporting 

anything. If you notice any UI issues, you can keep editing directly in Stitch. 

Just describe what you want to change and submit it and the tool updates the design 

automatically. For example, a responsiveness issue with a button was fixed immediately. By 

uploading images or entering a website URL, you can use existing designs as a reference and 

regenerate the UI inside Stitch. As you can see, the generated layout closely matches the reference 

website. Using the nano banana powered annotate feature, you can mark areas of the design to 

add, remove, or adjust specific layout elements. This update also adds a command 

pallet aimed at power users. Instead of navigating through menus, you 

can press command or control K and trigger actions directly by typing. This makes it much 

faster to jump between screens, run commands, and adjust the canvas without breaking your 

flow. If you're working on large or multi-creen projects, this significantly speeds things up. 

Stitch now introduces predictive heat maps. You can instantly run an attention audit on any 

screen to see where users are most likely to focus. This helps you validate visual hierarchy 

before writing any code without waiting for real user data. The feature is available now in the 

generate menu. With Gemini 3.0 Pro now powering Stitch by default, the platform removes many of 

the previous limits on what can be generated. Designs are now reasoned through before any 

code is written, resulting in stronger layouts, cleaner structure, and more reliable front-end 

output. This is especially noticeable in complex interfaces like dashboards and multi-creen flows 

where visual reasoning and DOM structure really matter. And because every design is backed by 

real exportable code, Stitch is no longer just a design tool. It's a practical starting point for 

building real applications. Thanks for watching.