Google just put out their new "nano banana" image model on some of their products, like Gemini and AI Studio. The biggest thing is that you can edit the image (change angles, backgrounds, etc.) without your photo subject looking all different.

So of course I'd experiment with it!

I've been workshopping a future custom video idea with Veo3 - as @skirtpie recently created an AI thread based on a creative studio's famous staff photoshoot from a decade ago, and I got some inspiration from that - with my own twist.

The GIFs are from Veo3 videos I plan to use in the workshopping - I took the last frame of the video (essentially the aftermath), and asked "nano banana" in Gemini to re-orient the image to a more frontal view, make it a full-body shot (this is important - the second example never showed the top of her head in the video), and up the quality to make it look like a professional camera. And of course, show the woman smiling.

Not too bad for what I was trying to do.

Still experimenting with this on Whisk and i'm pretty much blown away. I do realize they increased the censorship compared to pure text to image generation.

make it look like the woman has been running a mud race in that outfit.

make it look like the woman has been to holi party in that outfit.

Doesn't like pie though.



PS It didn't seem to understand gunge

Pic and text removed for referencing children's TV programme.

One of my Google accounts in Gemini got "upgraded" to the new model a couple of days ago, but given the experience I made so far - mainly with wet images - unfortunately it's a huge step backwards for me compared to Imagen 4:

- realism has gotten much worse, giving more of the Bing-like artifical looking results.
- resolution has drastically been dropped to 1024x1024px compared to the (upscaled) 2048x2048px, or others like 2560x1792px depending on aspect ratio.
- only square images can be generated again, one thing I *really* loved about Imagen 4 was its ability to prompt for different aspect ratios.
- filters are WAY more strict, almost none of my latest prompts are working anymore. not even the slightest hint of partial nudity allowed, where Imagen 4 had become quite forgiving.

The only postive thing really is its consistency and ability to edit or develop images, making it nice for doing series or before/after pictures with consistent persons and/or backdrops. But I hate pretty much everything else about it. Like I said initially my other account is still using Imagen4 for generation, so I have direct comparison and even prompts which are working with both will usually give much better results using the older model. I hope they'll address at least some of these issues.

Fundamentally, you're misunderstanding what the model is. Nano banana isn't a replacement for Imagen4, it's a complementary model that allows editing of images. That's it's strength, the weaknesses are what you described.
It's not a replacement for Imagen, choose what you want achieve and use the right model.

I'd also recommend using https://aistudio.google.com rather than gemini or other front ends. It'll give you the flexibility to easily choose.

This is correct - while "nano banana" can create images, it's intended as an image editor, while Imagen is intended as an image generator. I've attached an image created in each with the same prompt below. Try and guess which one is generated by which!

"Woman #1: age 27, slender and toned, fair-skinned, feminine features, 5 fingers on each hand, 5 toes on each foot, no nail polish on her fingernails or toenails, chest-length blonde hair, styled in a high ponytail with crimped waves, wearing a lace V-neck bridal gown and floral drop earrings
Woman #1 is smiling while sitting in a lawn chair in a grassy area next to a sidewalk on a busy city street on a summer night, legs extended, with her bare feet propped up on an upside-down blue bucket. A pair of white high-heel pumps are on the ground next to the bucket. The woman is holding a bouquet of white roses.
The soles, toes, and tops of the woman's bare feet are completely covered in a layer of thick, sticky, gooey, creamy Cool Whip. Splatters of Cool Whip go up to her calves, as well as splatters on and around the bucket, including on the ground. A blue plastic plate, topped with a residual amount of Cool Whip, is on the ground next to the bucket."

Be sure you're using the correct tools for your job.

As far as the subject matter for this prompt? Yes, I'm running with another crazy idea!

These images were created from existing photos. The woman wearing glasses is based on a photo of a real person, whom I can use because it's actually a picture of myself . However, the face was significantly enhanced and altered. The clothing is also from the original photo.