1992 Breda M-42 "Duster" SPAAG Upgrade - AI visualisation

They're photorealistic enough by now, with release of "Gemini 2.5 flash image (nano banana)", but good quality AI-generated pictures of vehicles with complex geometry, rivaling photos or renders or scalemodels of the same vehicles, still could take hundreds of attempts, which takes hours - at least on regular accounts on free websites everyone's using; and they still require some, or even a lot, of work in preparation.
One might need to clean drawing by hands, or make new drawing showing vehicle from any other side required, or switch to making a simple 3d model and then asking AI to make it photorealistic.

I recently made a picture of Wegmann IFV from mid-90s NGP program in Germany, while having only 1 isometric view of it, and side drawing of hull without turret, and it took 6 hours straight with varying promts and additional images (of gun and APS launcher), while probably more then 200-300 images were generated and rejected (dozens of them - because AI frequently refused to generate anything new at all, spitting back one of the images I've uploaded, unchanged; or because it completely failed to follow any of the provided examples; or because it failed to follow requests to add elements from first image to second by switching from designating one image "first" to another), as I was saving best ones - and only 24th among those finally got all geometry and scale (as I understood them at the time) sufficiently correct.

(May be premium accounts, which allow to generate more images within the same conversation then free accounts' 5 per 1 hr,
could improve it faster, as more and more changes and corrections are asked for within the same conversation and more context on what's needed is provided, but IIUC judging by complaints on say Reddit, it's not guaranteed at all, as the same frequent problems are still present).
 
Last edited:
whatismoo said:
This seems misleading in the extreme to the point of not being useful?
Click to expand...

Agreed. Given what we know of how AI will turn around and use this image as a source for a future "AI summary" posting this sort of AI image in a supposedly research-driven thread just makes future searches for info on this vehicle even worse.
 
skylancer-3441 said:
They're photorealistic enough by now, with release of "Gemini 2.5 flash image (nano banana)", but good quality AI-generated pictures of vehicles with complex geometry, rivaling photos or renders or scalemodels of the same vehicles, still could take hundreds of attempts, which takes hours - at least on regular accounts on free websites everyone's using; and they still require some, or even a lot, of work in preparation.
One might need to clean drawing by hands, or make new drawing showing vehicle from any other side required, or switch to making a simple 3d model and then asking AI to make it photorealistic.
Click to expand...
True. But they are enormous help for visualization of images.
 
Sorry but I would remain in topic, the AI images generation is interesting but we could move the discussion into a dedicated new thread?
 

Similar threads

Back
Top Bottom