I buy a fair amount of wetlook and slapstick here but its hard getting exactly what I want. As is pretty obvious, I think a glamorous woman wearing long gloves is great but it doesnt come up that often.

I've found that using Whisk in particular I can "improve" some of my favorite pictures.

I can't show actual examples here because modified pictures of real people aren't allowed. So I created a completely picture of a glamorous woman (#3 here). She also has a tattoo between her breasts, which I don't like. Not a big fan of tattoos. In picture #1, I've removed the tattoo and added long matching gloves. #2 adds rain and in #4 she's been hit by a pie. I'm in heaven.

It doesn't always work or work well. Whisk balks if the woman's outfit is too low-cut or shows too much boob or nipple. I use "Add Images", then "Upload Image" (lower part of the yellow box). The icon in the lower left after the image is "analyzed" creates a text description of the image. I check the description because Whisk loves describing lipstick as "nude" which trips the alarms. I delete it or substitute pink. Then I have fun in the lower "describe your edits" box.

I sometimes take the original (modified if necessary) description and my edits text to other programs such as Gemini but the Whisk results as you can see are the closest to the original that caught my attention.

However I don't show those close to the original results anywhere to avoid the Anti-AI Vigilantes.

She's a particularly lovely lady! Would you mind if I added her to my growing group of assistants, which I dip into when I want a specific (AI) person, rather than a newly-generated one?

Needless to say, if you're not happy with that I won't do it.

I amend already-generated images in Whisk often. It's actually very powerful and will let you alter quite a lot of things.

Sure. I'm flattered.

That's great, thank you. In return here's my AI assistant, Jenna, who you're welcome to add to your group if you wish. The first video I've made featuring her is waiting in the wings to be released!

Dude I don't wanna be offensive but doing this with photos of real people is actually disgusting. They didn't consent to having totally unrelated photos being used as fetish content. It's the same thing with deepfake porn, there isn't any fucking consent or knowledge of it. It's the same thing as a random guys likeness being stolen to put on a billboard in china.

AI photo editing software can edit photos, who knew?

Is the photo editor in Whisk the same as Nanobanana? I find Google's range of AI tools quite confusing in terms of branding. There's the Google Labs selection of ImageFX, Flow, etc, then there's Whisk, and then there's Google AI Suite with Gemini and Nanobanana. I'm "still" using ImageFX for still images for video training, it's quick and free and good enough for what I need.

You are on morally dubious ground with editing photos of real people, obviously. When it comes to manipulating images without consent, it's not "anti-AI vigilantism" to be against it.

and


Which is why I tread carefully here and only posted an original photo of an AI created woman.


Wait: They didn't? They were paid to pose for a wetlook shoot but didn't know it was going to be used as fetish content? Really?

I was gonna post a picture of a (real-life) wetlook model from one of the producers, with gloves added, but now... I'm done.

To the one serious question, I find the add images section of Whisk gets much closer results to the original than the regular area. And not asked but if the generated description includes polished fingernails I have to edit that out before adding gloves!!

ok now I'm done.

I agree with the confusion of google products but I find it easier to distill it down to the core models first and then how they're delivered.

Imagen4 - Google's pure text to image model.
Nanobanana Pro - Google's latest omni model. Functionally it's an edit model that allows multiple images to be provided for consistency or editing. In total, it accepts up to 14 reference images. These 14 images can include the following: Up to 6 images of objects with high-fidelity to include in the final image, Up to 5 images of humans to maintain character consistency.

Even without providing reference images, nanobanana is a generation ahead of imagen4 from a pure Text to image capability.

How these models are then utilised cross google's cloud platforms differ depending on the product. In the case of flow and whisk, they're providing a front end to allow the user to input styles or reference images to be used. Whisk will accept input images and analyse them to provide a blended prompt that is used my the model. Personally, I'm not a fan of this, safety constraints are dialled up in this product and feel there is a lack of control. There is nothing you can't do better with GPT or Gemini models with your own instructional prompts.

ImageFX is just a wrapper for the imaging models. Again, just a way of making it easier for the end user and delivering a product. The safety features are static and non controllable.

Where is gets more interesting is AI studio. This is effectively a front end for the API and you get the purest access to the model along with controls around system prompt and guardrail settings. It's less user intuative but way more powerful. You can also create an API key to user in your own scripts.

Vertex, is google's AI platform, it has features similar to AIStudio as well as way more models and products. It's aimed more towards product development and delivery.