← sources

What are you guys hoping to see in upcoming image and video models?

source: reddit (reddit:StableDiffusion)published 2026-10-24 20:20ingested 2026-10-25 02:38

retained text

We've been eating good lately, so I'm curious what you guys wanna see in future models. T2I, I2I, video, anything goes.  
  
Personally, I'm really happy with Krea 2. The model has honestly delivered pretty much everything I've thrown at it. The only thing I'd really want is a better vae and maybe a slightly bigger DiT.

Been looking into more technical stuff lately and looped transformers sound really promising. Makes me wonder how likely we are to get an open-source model using a completely new architecture like the one GPT-2.5 uses.

document fetched 2026-10-25 00:29 · canonical https://reddit.com/r/StableDiffusion/comments/1xeissk/what_are_you_guys_hoping_to_see_in_upcoming_image