What Is Claude Imagine | How to Use the Real-Time Generation UI and Its Limitations

What Is Claude Imagine (Imagine with Claude)?

Imagine with Claude is a limited-time research preview released by Anthropic on September 29, 2025, simultaneously with the Claude Sonnet 4.5 launch (official announcement). The official announcement describes it as follows:

Claude generates software on the fly. No functionality is predetermined; no code is prewritten.

In other words, when a user says "I want an app like this," Claude creates the screen (UI) and its behavior in real time. The defining experience is that the moment you press a button, that feature is implemented for the first time.

At launch, it was accessible via claude.ai/imagine and offered to Max plan subscribers for a limited 5-day period. It was positioned as a showcase of Claude Sonnet 4.5's reasoning ability, tool use, and multi-step planning capabilities. Note that this 5-day preview period has already ended, and as of June 2026, new access to claude.ai/imagine is not available (no plans for re-release have been announced either).

What It Can Do — How the Real-Time Generation UI Works

When you launch Claude Imagine, a screen modeled after a desktop OS opens. According to a verification article by Testing Catalog, the screen displays familiar UI elements such as icons, windows, a trash bin, and a toolbar, with an input field in the center reading "What do you want to build?"

There are three notable characteristics of how it works, all reported as internally observed behavior in the Testing Catalog verification:

1. Builds the UI by directly manipulating the DOM: Claude directly adds and updates DOM elements (screen components like buttons and lists) with each conversation turn. There is no step of writing code and then building it.
2. Automatic window system management: The verification article reported that windows of various sizes from xs to xl were dynamically laid out, with support for displaying multiple apps simultaneously.
3. Separation of UI generation and feature implementation: The process of rendering the visual appearance and the process of creating the behavior when a button is pressed were observed to work together in a layered manner.

Conventional "code-generating AI" outputs code as the deliverable, but Claude Imagine is fundamentally different in that it outputs the running app itself.

How to Use It — From Access to Generation

As mentioned above, the preview period has already ended, so the steps below cannot be followed today. Here is a record of how it was used when it was available.

1. Log in to Claude Max plan and access claude.ai/imagine (or launch from a button on the chat screen).
2. When the desktop-style screen opens, enter what you want to build in the "What do you want to build?" input field.
3. Wait as Claude opens a window and builds the app while rendering the UI.
4. Actually click the generated buttons and menus, and the corresponding functionality is implemented and executed on the spot.
5. Enter additional requests (such as "Add a search feature to this list"), and the running screen is directly rewritten.

Prompts can be entered in Japanese, but note that labels within the generated UI are often in English.

What It Can Build — Real Examples from Verification Reports

A verification article by DataCamp tested four real-world examples:

• News feed app: A feed that collects only positive news. The layout, article cards, and refresh button all worked, and it was introduced as a success story.
• Dungeon exploration game: The screen was generated, but bugs occurred in the movement processing. High-real-time-demand use cases like games are a weak point.
• Resume (CV) management app: The input form and list display functioned, but PDF output produced errors — a partial success.
• Portfolio site: Generated at a level where page structure and navigation were functional.

The general trend observed at the time of testing is that tool-type and information management apps are relatively strong, while games and processing involving external output are unstable.

Limitations and Caveats

Claude Imagine is an experimental demo, and it has clear limitations as a practical tool.

• No persistence: Refreshing the page erases all generated apps. There is no means provided to save or publish what you've built.
• Limited output variety: Nearly the same results are generated from the same prompt, making it unsuitable for brainstorming multiple design ideas.
• Bugs remain: Buttons are sometimes placed without being implemented, and output features produce errors — fine-grained behavior is incomplete.
• Limited availability and target audience: At launch, it was limited to Max plan subscribers for a 5-day period. Whether it will become a permanent feature has not been clearly stated by Anthropic.

The fact that "what you build doesn't persist" is particularly important — for creating work deliverables for business use, existing features like Artifacts or Claude Code should be used instead.

Differences from Artifacts

Claude has long had an "Artifacts" feature that generates and runs code and web apps within the chat. Here is a breakdown of the differences between the two.

• Artifacts: Claude writes code, which is then run as a preview. The code remains as the deliverable and can be shared and re-edited.
• Claude Imagine: Generates a working screen directly without going through code as an intermediate product. No deliverable remains; the experience itself is the demo.

Artifacts is superior in terms of practicality, but Claude Imagine holds significance in demonstrating a future vision that "software doesn't have to consist of pre-written code."

Future Outlook — The Direction of "On-the-Fly Generated Software"

DataCamp's verifiers assessed its potential, saying "it could be revolutionary if integrated into development editors" and "integration with design tools like Figma is conceivable." As model inference speed improves and costs come down, applications where the screen is assembled each time according to the user's intent — instead of fixed menus and forms — become increasingly realistic.

Anthropic tends to release this type of experimental feature at model milestones, and Claude Imagine was also provided as a demonstration of Claude Sonnet 4.5's capabilities. The best way to follow how this concept develops with future superior models is through Anthropic's official news.

Summary

• Claude Imagine (Imagine with Claude) is Anthropic's research preview that generates software on the fly without pre-written code
• Released simultaneously with Claude Sonnet 4.5 and offered to Max plan subscribers for a limited time
• Tell it what you want to build on a desktop-style screen, and the UI and functionality are implemented in real time
• There are many limitations including no saving, monotonous output, and bugs — for practical use, Artifacts or Claude Code are more appropriate
• It is valuable as an experiment that showed a future vision of software where "screens are generated each time according to the user's intent"

Sources: Anthropic Official Announcement (Claude Sonnet 4.5) / DataCamp: Imagine with Claude Verification / Testing Catalog: Real-Time UI Generation Experiment