This is using imagen-4.0-ultra-generate-preview-06-06
the AI still gets upset over certain key words (like shower for example). I hope they come out with an image editing version instead of text to image in the future. That would help avoid getting so many errors.
I hope they come out with an image editing version instead of text to image in the future.
Nano Banana Pro within Gemini may probably be what you're looking for. It's a stunning model, not the best I've seen so far in terms of realism and quality, but is has also great editing capabilities allowing you to make very consinstent series of images.
Nano Banana Pro within Gemini may probably be what you're looking for. It's a stunning model, not the best I've seen so far in terms of realism and quality, but is has also great editing capabilities allowing you to make very consinstent series of images.
Yes Nano Banana is the best image editor. second and maybe equal to it is Qwen from Alibaba. NB allows multiple images to get uploaded at a time which is how I have been able to make more complex poses and include multiple people etc. However Imagen seems to be the most realistic for text-to-image IMO. Plus it is way faster like maybe 10x faster.
I'm talking about the latest Nano Banana Pro, which is even more realistic and detailed than Imagen and even (much) better in editing than the "normal" NB, let's say a blend of both worlds, honestly I haven't seen another model so far that even comes close to it. Here's a series of images I did today, I think you'll se what I mean.
WF1 said: I'm talking about the latest Nano Banana Pro, which is even more realistic and detailed than Imagen and even (much) better in editing than the "normal" NB, let's say a blend of both worlds, honestly I haven't seen another model so far that even comes close to it. Here's a series of images I did today, I think you'll se what I mean.
Well, if you try to make a woman in a shower using NB you just can't make it do it (regardless what clothes you specify). It just does not accept the term "woman in a shower" as being anything but pornographic. Also getting see thru wet clothes in NB is a real game of hit-or-miss whereas imagen does both without a care.
HOWEVER... if you upload a pic of a woman wearing see thru wet clothes (no R-rated stuff) then upload a pic of a woman in the shower and tell it to make a new image combining the two then it seems to work.
But as for just realism and detail then latest NB is best.
It was not so much a question of getting the right prompt like you had to do months ago with Bing. It is a question of using the right AI image generator with all the safety checks tuned the right way.
I was using Imagen 4 to generate the ones I posted. Prompts were something like "a millennial woman wearing a pink cap sleeve shirt, white skirt, white tights, and hoop earrings emerges from the pool using the pool ladder. her hands are on top of the ladder. clothes and hair are soaking wet"
Well, if you try to make a woman in a shower using NB you just can't make it do it (regardless what clothes you specify). It just does not accept the term "woman in a shower" as being anything but pornographic. Also getting see thru wet clothes in NB is a real game of hit-or-miss whereas imagen does both without a care.
Hm that's strange. Haven't used the "normal" NB since NBP came out and usually do more pools or lakes, but when I ask for showers it will give me showers. I just tried, beginning the prompt with "woman in a shower, ..." and got these.
wow you must have the magic touch with Gemini, WF!
It won't even try giving me images it just says "you have violated the content guidelines" and does nothing. I can get it to work by combining images only, such as put this woman in the pose of the uploaded picture (some stock photo of a woman in a shower). Or I say something like 'water drips down on her'. Honestly I think they are tracking us when we use the AI tools so possibly it is different for different people?
Very nice photoset with the gloves and red dress BTW!
Well that's strange. Yeah I think that AI keeps constantly learning and that may also be user-based... I suspected it also might be a bit looser for paid users, but when I tried the same prompts with my other Google account using the free subscription it would give me results as well (although not being as refined as on my other account, it seems).
Love you, too
