Those anticipating a new top model were left disappointed this week. The focus was on the surrounding tools, which can be just as beneficial for daily producers.
Midjourney: Editing Gets Serious
On 23 September, a major alpha update was released: live style examples, default parameters you can save, a revamped Create feed, and enhanced editing tools for V8.1 and V8.2.
A day later, improved inpainting and outpainting, faster models in the interface, and
--tilethat now seamlessly continues followed.Most interesting for us: you need to regenerate less frequently and can make more targeted corrections.
Qwen-Image-2.1: Open and Transparent
Alibaba released Qwen-Image-2.1 on 20 September: generating and editing in one model, with direct support for transparent backgrounds.
Freestanding products without separate masking work are a real gain for webshop images. I have not tested this myself.
Small
Adobe Firefly (18 September): GPT Image 2 now supports 10 image ratios instead of 3, and prompts can be longer.
As a reminder, the benchmark for this month: ChatGPT Images 2.5 (8 September) generates twice as fast and retains the subject from reference photos better.
Not for images: Black Forest Labs (FLUX) released FLUX 3 Action on 25 September, but that is a robotics model.
To Follow
No firm release dates found for Nano Banana, FLUX, or Ideogram this week.
Sources
Stories from our work and our minds, in your inbox.
One mail with everything: the blog, AI news and what we are building. No fixed frequency.
