@SeanPedersen96@mmastrac I predict it will be difficult. Although DiffusionGemma's base model is Gemma4 26B A4B, the number of experts activated during inference is much higher than that of Gemma4 26B A4B (84 vs 8). The performance of 8GB VRAM gpu will be severely impacted by RAM bandwidth.
@OfficialLoganK I like Antigravity's undo functionality without relying on Git, and I hope there will be a more advanced snapshot feature to undo changes to non-text files. Of course, I'm also looking forward to the new Gemini model; will there be any surprises this month? :)