Testing GPT image tiers: high vs ultra barely differ, quality drops hard on low settings
kevinkern · x · 2026-09-13
Browsing the GPT Observ site, the author compared image quality tiers: high and ultra look nearly identical, while medium and especially low show visible quality drops. The sweet spot appears to be around Med/High. Useful guidance for users paying per-generation.
More from Multimodal
- Crafting an Invisible Video Transition in MiniMax H3: Match and Sun Swapped Under White — LudovicCreator · 2026-09-13
- Agent builds 15x14km 3D tile of Marbella in ~1h: 4,393 houses, 2,492 pools, 450km drivable roads — LinusEkenstam · 2026-09-13
- Seedream team unveils VoT: visual thinking before pixel rendering for image generation — JingxiangSun42 · 2026-09-13
- Wan 2.2 Suddenly Outputs Wavy Abstract Video; Blackwell fp16 Instability Suspected — deviruchii · 2026-09-13
- Krea 2 Recognizes Most Source Characters Without LoRAs, User Finds — magik_koopa990 · 2026-09-13
- Tencent's Open-Source WeMM-Embedding Runs Video Retrieval on an RTX 3060 — huangyun_122 · 2026-09-13