# Motion added to the algorithmic-art skill comparison

27 September 2026. This is an evidence-backed extension of the earlier survey, not a new performance ranking.

## What changed

The reference set now includes Gemini’s breathing light; DeepMind Institute’s moving geometry; DeepSeek’s procedural background and typing; Claude’s changing typographic grid; OpenAI DevDay’s modular motion; Grok’s voice orb and simulated thinking states; and Astra’s filmed demo as a contrast case. Every entry has moving evidence in MP4 and GIF form.

## Best next test briefs

1. **Breathing presence:** stable form, slow expansion/contraction of light, a natural rest; distinguish perceived life from actual task progress.
2. **Interrupted order:** a persistent typographic or geometric structure, localized variations, specified holds and transitions.
3. **Coherent field:** a real, named noise implementation, smooth temporal sampling and a reproducible seed. DeepSeek supplies the strongest public-code evidence in this batch.
4. **Waiting as performance:** distinguish idle, submitting, waiting, streaming, completion and failure. Evaluate transitions and interruption, not just a looping spinner. Current Grok/DeepSeek captures are marketing simulations, not live chat tests.

## What to compare

Use Anthropic and Pardesco with the same model, brief, budget and seeds. Judge original work for motion rhythm, temporal continuity, usefulness of controls, reproducibility, source/technique honesty, and the ability to explain a look as person + concrete operations. Show each complete cycle and several transitions. Keep technical correctness separate from aesthetic preference; do not assign influence or authorship based on resemblance.

The Vibes companion supplies the people, portraits and prompting language. The art skills supply implementations. These marketing references supply observable test cases. No controlled generation benchmark was executed in this turn, so the earlier workflow recommendation remains unchanged.

## Evidence ladder

- **Published implementation:** DeepSeek’s loaded chunk contains a WebGL2 renderer, simplex-style noise, u_time updates, draw calls, and distortion/swirl parameters. Static inspection; not a runtime trace.
- **Live rendered surface:** Grok’s orb appears at a canvas location and changes over the recording. Exact algorithm not established.
- **Observed motion:** Claude typography, Gemini light, DeepMind geometry, DevDay reel. Movement is captured; algorithms and historical influences are not inferred as facts.
- **Documented production authorship:** Play Studio’s DevDay case study names its team and OpenAI collaborators.
- **Contrast case:** Astra teaser is a filmed interaction demonstration, not established algorithmic art.

## Sources

- Anthropic skill: https://github.com/anthropics/skills/tree/main/skills/algorithmic-art
- Pardesco: https://github.com/Pardesco/algorithmic-art-skill
- DeepSeek renderer: https://www.deepseek.com/_next/static/chunks/4097-1b51b1ad13616f56.js
- DevDay makers: https://play.studio/work/open-ai-dev
- Perlin’s research: https://cs.nyu.edu/~perlin/

The source-code fingerprint index and capture notes accompany the report. Keyword presence in a framework alone is not accepted as proof of a particular on-screen algorithm.