Image Model Test: MAI-Image-2.6 Beats GPT Image 2.5 on Knitted Toy Prompt at Half the Price
Fun_Walk_4965 · reddit · 2026-10-29
The author ran the viral "photo to knitted designer toy" prompt on the same photo across three edit models via Atlas Cloud API:
| Model | Billed | Time | Result |
|---|---|---|---|
| GPT Image 2.5 Sunburst | $0.10 | 3m 59s | Beautiful knitted bust, normal head proportions |
| GPT Image 2.5 Flare | $0.10 | 1m 04s | Nearly identical to Sunburst |
| MAI-Image-2.6 | $0.06 | 1m 01s | Only one that built the oversized wide head |
- The key line is "oversized, horizontally broad head with tiny features" — both GPT Image tiers ignored it and produced faithful but normal-proportion portraits, while MAI-Image-2.6 actually delivered the exaggerated toy shape.
- Flare matches Sunburst on price but is nearly 4x faster; MAI is the cheapest, though capped at 1024px output.
More from Multimodal
- Open-Genjutsu: a free open-source alternative to $150/mo video cloning apps — matchaman11 · 2026-10-29
- 'The Dream Goes On': an AI-generated video with cinematic night-running shots — Mindless_Plum_2571 · 2026-10-29
- Magnific Move launches with templates for product ads and motion graphics — charis_ai · 2026-10-29
- Runway CEO revisits 2019 AI short film, making 'Will Smith spaghetti' look HD — c_valenzuelab · 2026-10-29
- Open-source pipeline turns a song into a beat-locked music video on one RTX 5080 — Adebrantes · 2026-10-29
- Open-source EffectCraft panel runs ComfyUI workflows directly on a video timeline — HomoLapis · 2026-10-29