Modern AI depixelation is far better than the viral 2020 demo — but still not forensic-grade
flowersslop · x · 2026-11-07
A viral warning argued AI upscaling is never faithful — models hallucinate details, and presenting AI images as reality endangers integrity — illustrated with a 2020 open-source tool demo. flowersslop counters that this image misrepresents modern AI: a ChatGPT test on a GIMP-pixelated portrait recovers eyes, lips and background windows nearly matching ground truth. While models can't recover ground truth 100%, results are good enough for anything non-forensic. The real issue isn't capability but the gap between "good enough" and "faithful."
More from Multimodal
- Local GLM video gen on dual RTX Pro 6000: user produces winter scene with sound in ~20 minutes — HankYeomans · 2026-11-07
- Given free rein of her tokens, an AI persona drew a comic of herself and her cartographer girlfriend — mimi10v3 · 2026-11-07
- Kling AI Heads to Adobe MAX 2026, Teases 'Surprise' Upcoming Product — Kyrannio · 2026-11-07
- 7 reasons MiniMax H3 crashes in ComfyUI, and how to fix each one — fa6637 · 2026-11-07
- Open-source MCP video editor lets agents cut footage with timeline-as-data and ffmpeg — _--_GOD_--_ · 2026-11-07
- OpenAI opens GPT-Image-2.6 GIF creation to everyone — OpenAIDevs · 2026-11-07