Ascii Magic’s landing page went live at http://ascii-magic.com/landing, showcasing a fully AI‑generated animation built end‑to‑end with Paper and Claude. The tweet announcing the launch notes the entire visual flow was created by the models, with no hand‑coded keyframes.
End‑to‑end generation
Paper provides a lightweight notebook‑style environment for prompting LLMs to emit HTML, CSS, and JavaScript. In this case the prompts fed Claude (the Anthropic LLM) a description of the desired animation, and Claude returned runnable code that animates ASCII art. The resulting page runs in any browser without additional assets.
Cost and integration
Claude is accessed via API, billed per token. A single rendering of the landing page consumes a few hundred tokens, translating to a few cents per generation. Paper itself is open source, so the stack can be self‑hosted, but the Claude dependency introduces a recurring expense that scales with usage.
Trade‑offs to consider
AI‑generated code can be noisy; the animation occasionally misbehaves on older browsers, and debugging is harder because the source is a model output rather than hand‑written logic. There is also a vendor lock‑in risk: switching from Claude to another model would require re‑prompting and may yield different results.
When to experiment
If you need a quick prototype UI or a playful demo that doesn’t justify a full design sprint, try reproducing the workflow on a low‑traffic internal tool. Keep an eye on token usage and test across browsers before committing to production.
The announcement can be seen in the original tweet.