Chat-GPT Images 1.5 VS. 2.0, What’s new?
As a 3D visualizer, I think it’s important to stay up-to-date with the AI tools that are changing the 3D visualization industry. So when OpenAI launched their new image model, ChatGPT Images 2.0, I had to put it to the test and see what its uses would be for landscape visualization.
I compared the new OpenAI image model, ChatGPT Images 2.0 to the older version, 1.5 to see what’s new. And instead of telling you what has changed, I thought it would be better to show you! I made a prompt based on a location, described a realistic street scene and used this prompt to generate images with both the 1.5 and 2.0 version.
Upgrade: the model ‘’thinks’’ before generating
My reaction: it indeed seems to think more, because generating images take A LOT longer…What took only 47 seconds in version 1.5, took me 3:06 minutes (!) in its newer version 2.0. And I even got results after 4:31 minutes (!) in version 2.0. It’s worth the wait, just make sure you have something to do in the meantime…
Upgrade: better text in images
My reaction: big signs with text that face the camera seem to be alright in 1.5, but once the text is either small or in a different perspective, it messes it up. 2.0 does this a lot better, most text, big or small is readable and it’s still correct when it’s not facing the camera. It’s not perfect though, when you zoom in you do notice that it’s not pixel perfect. But that seems to be the rule with fully generated images anyway: Don’t zoom in too much! ;)
Upgrade: Multiple images at once and more aspect ratios available
my reaction: yes, there are more aspect ratios available in 2.0, seven to be precise, compared to the three in 1.5 that’s a nice upgrade. You may not use them all, but it’s nice to have available. It should also be able to generate 8 images at once, however that option apparently isn’t available in KREA. Also, 8 images seem like a lot to choose from. I wonder if it’s a case of ‘’quantity over quality’’ or if it’s actually helpful to have a whole range of images to choose from.
Upgrade: Better spatial understanding, fixed the ‘’AI-look’’
My reaction: definitely, it’s clear that the newer model understands scale, ratio, distances, movement, placement a lot better than before, which makes your images make a lot more sense. Getting rid of the ‘’AI-look’’ is a substantial upgrade. You’d have to look at least twice now to be sure that it is AI-generated.
Conclusion
I’d say the new 2.0 version of ChatGPT Images is in many ways a good (and necessary) upgrade as it now comes close to other models like Nano Banana Pro. Whether it will become the new standard is to be seen… The wait time is definitely something that might become annoying with longer or more intense use.
What does this mean for landscape visualization?
I think the biggest change is that the model now tries to understand a space before it makes the image. Instead of just mixing together visual patterns taken from other images, it actually thinks about size, distance, positioning and how things relate to each other. For landscapes, that’s a big deal. Roads, trees, and buildings are placed in a logical way and that makes the image much more realistic. This makes it quite a useful tool for generating reference images or mood images, for example when you are in need for something that is really specific. It is, however, a long way off in replacing us as landscape designers and visualizers, for the ones that want correct images at least ;)
Over de auteur
Timo Hoogstra
Ik ben een landschapsontwerper die zich heeft gespecialiseerd in 3D-visualisatie. Ik maak gedetailleerde modellen voor landschaps- en stedelijk ontwerp. Ik werk met SketchUp, Rhino en Revit, en breng modellen tot leven met Lumion of V-Ray. Je krijgt nauwkeurige, meeslepende 3D-beelden—met duidelijke communicatie en ruimte voor revisies.