# I built a comparison PPT across Anthropic, OpenAI, and Chinese AI — retained record

> Retained reddit record from reddit:ClaudeAI.

Canonical: https://highsignal.app/data/records/dce28f26e798cf18

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeAI)  published 2026-10-29 02:28 ingested 2026-10-30 02:36

[original source ↗](https://www.reddit.com/r/ClaudeAI/comments/1xi81nk/i_built_a_comparison_ppt_across_anthropic_openai/)

## retained text



```
the video is a 12-slide comparison of Anthropic, OpenAI and Chinese AI, covering revenue growth and compute spending. I built it with Claude Code in HTML, CSS and SVG, starting with generated images as design references. I wanted to try the workflow across a full deck, with growth curves, company comparisons and spending breakdowns.

For the references, I used atlas cloud’s MCP to try different image models on the same chart briefs. Seedream 5.0 gave me some designs I liked, but I preferred Image 2.5 Sunburst overall. Having actual drafts to compare helped me figure out the layout I wanted before getting into the code.

Sunburst kept adding more text than I wanted, though. I’d like the chart, then notice another heading or paragraph around it. Some slides were turning into little articles. Before generating the final references, I marked what needed to go and specified what should stay: the title, date range, labels, values, units and sources.

The final batch was one request per slide, with three candidates per request. That came to 12 requests and 36 images. My rough estimate for that batch is about $2.47, depending on token usage. That only covers the final references, excluding the earlier model comparisons and the coding work.

Then I gave Claude Code the references and asked it to recreate the layout, spacing, colors and hierarchy while keeping the text and chart elements editable. Once the slides were in code, I could ask for changes to individual elements. I removed a magnified inset from one chart, revised labels and switched the whole presentation from portrait to landscape for desktop viewing.

For the animation, I had Claude Code make the bars grow to their values, draw the lines, count up the numbers and reveal the explanations in sequence. I could scrub through the timeline, find something that appeared too early or stayed on screen too long, and ask it to adjust that part. The exported clip is 68 seconds at 2.5x playback speed, with music added afterward.

The images helped me make the initial design decisions. Once the slides were working, I preferred making the remaining changes in HTML rather than generating more references.

For anyone using Claude Code for presentations, do you start with image references, or work out the layout directly in code?
```



document fetched 2026-10-30 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1xi81nk/i_built_a_comparison_ppt_across_anthropic_openai
