Frontier Models Generate Stunning Visuals but Fail at Simple Edits
anshulkundaje · x · 2026-08-02
Stanford professor Anshul Kundaje found that while frontier models like GPT can generate phenomenal figures from complex text descriptions, they perform poorly on subsequent precise modifications.
Users pointed out a common pain point: models cannot tweak existing images based on instructions (like simply flipping an arrow) nor convert them into editable formats like SVG or PPT. This exposes significant shortcomings in current image understanding and controllable generation.
Related event: Frontier AI Models Excel at Image Generation but Fail at Precise Edits(2 posts)→
More from Models
- Claude's invisible watermarks cracked within hours; override code gets 20k bookmarks — deliprao · 2026-08-24
- Sonnet 4.5 exhibits intense, strange behavior in response to Opus 3 — repligate · 2026-08-24
- A comprehensive ranking of various AI models has been shared — FinanceYF5 · 2026-08-24
- User comparison finds LTX outperforms H3 in instrument generation energy — cocktailpeanut · 2026-08-24
- Controversial AI Model Ranking: Fable 5 at S+, Kimi K3 and DeepSeek V4 Flash in Tier B — FinanceYF5 · 2026-08-24
- Video Gen Consumes 70% of AI Tokens in China, Diverging from US LLM Focus — AccBalanced · 2026-08-24