GPT Image 2.6 and multiple reference images
GPT Image 2.6 from OpenAI, up to 4 reference images, and better Tripo 2K and 4K textures.
New
- New image model: GPT Image 2.6 — OpenAI's newest image models now lead the Image tab's model list. Flare is built for fast iteration (80 credits) and Sunburst for detailed work (100 credits). Both support 1:1, 3:2 and 2:3 and are available on paid plans. They're in the API too, and calls that don't name a model keep the same default.
- Use up to 4 reference images — GPT Image 2.6, GPT Image 2 and Nano Banana 2.1 now take more than one. Images are numbered in the order you add them, so refer to them by number in your prompt, like "Redraw 2 in the style of 1". The model list shows how many each model accepts.
- A dot marks results you haven't seen — When a card in the model list on the right has a new result you haven't opened, a dot appears in its top-left corner. Select the card and the dot goes away.
- Setting joint depth is easier in Mark joints — The magnifier now opens when you drag a joint forward or back in the side view, and the arms hide while you drag a center joint (pelvis, chest, neck or head) so the torso stays in view.
Fixes
- Better Tripo 2K and 4K textures — We moved to Tripo's newer texture model (v3.5). Colors come out more accurately, and the back of a model no longer picks up the pattern from the front. The price is the same.
- See lightbox iterations as they run — The thumbnail strip now shows an in-progress card, and you can open the result right there when it's done.
- Smart UV shows up in the mobile tools menu — It was missing from the menu on mobile. That's fixed.