Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Gemini 2.5 Flash Image

developers.googleblog.com

Post reply on HN
Print view View original

481–490 of 504 posts

Re: Gemini 2.5 Flash Image

#481

Post by __rito__ »

ctippett wrote: post #225
__rito__ wrote: post #192

I am glad that I never decided to become a photoshop pro. I always contemplated about it, seemed attractive for a while, but glad that I decided against it. RIP r/photoshopbattles. It was in the endless list of new shiny 'skills' that feels good to have. Now I can use nano-banana instead. Other models will soon follow, I am sure.

Interesting take. I'm a programmer, but learned Photoshop in the early 2000s and had a blast making and editing images for fun. Sure, the generative models today can do a far better job than anything I could come up with, but that doesn't detract from the experience and skills I picked up over the years. If anything, knowing Photoshop (I use Affinity Designer/Photo these days) is actually incredibly useful to finesse…

> learned Photoshop in the early 2000s and had a blast making and editing images for fun

> "had a blast"

One can have blasts in many things nowadays. Like playing Factorio, writing functional code for recreational problem solving, playing Chess, making SBC/Microprocessor projects for fun, doing Math for fun, and so on...

Photoshop just couldn’t compete with the existing blasts in my life, and I felt a little bad for not learning it. But that teeny, tiny bad feeling has been wiped away by nano-banana.

Re: Gemini 2.5 Flash Image

#482

Post by technofiend »

carlosbaraza wrote: post #433
danpalmer wrote: post #405

Earlier quoted context omitted.

I've done ~20 prompts so far and not had one be rejected so far. What sort of things are you asking it to do? I've tried things like changing clothing and accessories on people.

Basic things like: "{uploaded image of a man} can you remove the glasses?" or "make everyone in the picture smile" or "open the eyes of everyone in the photo". Nothing that a human would consider "unsafe". I am based in EU and using Google AI Studio with all safety toggles set to "Off".

For a joke between friends I had it take my selfie and make me a bald Catholic priest and then add hair to a friend who is bald. No refusals, although those are pretty tame. In contrast to the quality images nano-banana produced, Copilot removed my glasses and made my eyes brown.

Re: Gemini 2.5 Flash Image

#483

Post by torginus »

hapticmonkey wrote: post #347
fariszr wrote: post #60

This is the gpt 4 moment for image editing models. Nano banana aka gemini 2.5 flash is insanely good. It made a 171 elo point jump in lmarena! Just search nano banana on Twitter to see the crazy results. An example. https://x.com/D_studioproject/status/1958019251178267111

Before AI, people complained that Google was taking world class engineering talent and using it for little more than selling people ads. But look at that example. With this new frontier of AI, that world class engineering talent can finally be put to use…for product placement. We’ve come so far.

I am pretty sure a lot of said engineering talent isn't actually contributing to AI but doing other stuff

Re: Gemini 2.5 Flash Image

#484

Post by vunderba »

MrOrelliOReilly wrote: post #443
vunderba wrote: post #170

I've updated the GenAI Image comparison site (which focuses heavily on strict text-to-image prompt adherence) to reflect the new Google Gemini 2.5 Flash model (aka nano-banana). https://genai-showdown.specr.net This model gets 8 of the 12 prompts correct and easily comes within striking distance of the best-in-class models Imagen and gpt-image-1 and is a significant upgrade over the old Gemini Flash 2.0 model. The re…

This is incredibly useful! I was manually generating my own model comparisons last night, so great to see this :) I will note that, personally, while adherence is a useful measure, it does miss some of the qualitative differences between models. For your "spheron" test for example, you note that "4o absolutely dominated this test," but the image exhibits all the hallmarks of a ChatGPT-generated image that I personall…

Yeah - unfortunately the ubiquitous "piss filter" strikes again. You pretty much have to pass GPT-image-1 through a tone map, LUT, etc. in something like Krita or Photoshop to try to mitigate this. I'm honestly a bit surprised that they haven't built this in already given how obvious the color shift is.

Re: Gemini 2.5 Flash Image

#485

Post by vunderba »

jay_kyburz wrote: post #355
vunderba wrote: post #170

I've updated the GenAI Image comparison site (which focuses heavily on strict text-to-image prompt adherence) to reflect the new Google Gemini 2.5 Flash model (aka nano-banana). https://genai-showdown.specr.net This model gets 8 of the 12 prompts correct and easily comes within striking distance of the best-in-class models Imagen and gpt-image-1 and is a significant upgrade over the old Gemini Flash 2.0 model. The re…

I really like your site. Do you know of any similar sites that that compares how well the various models can adhere to a style guide? Perhaps you could add this? I.e. pride the model with a collection of drawings in a single style, then follow prompts and generate images in the same style? For example if you wanted to illustrate a book, and have all the illustrations look like they were from the same artists.

Hi Jay, unfortunately I haven't see a site like that but being able to rank models in terms of "style adherence" but it would be a nice feature.

It's basically a necessity if you're working on something like a game or comic where you need consistency around characters, sprites, etc.

Re: Gemini 2.5 Flash Image

#486

Post by geysersam »

UltraSane wrote: post #408
atleastoptimal wrote: post #384

I can imagine an automated blackmail bot that scrapes image, video, voice samples from anyone with the most meagre online presence, which then creates high resolution videos of that person doing the most horrid acts, then threatening to share those videos with that person's family, friends and business contacts unless they are paid $5000 in a cryptocurrency to an anonymous address. And further, I can imagine some per…

But these new amazing AI image generators lets you just say "It wasn't me, it is an AI fake". Long term they will seriously devalue blackmail material. I read a scifi novel where they invented a wormhole that only light could pass through but it could be used as a camera that could go anywhere and eventually anytime and there was absolutely no way to block it. So some people adapted to this fact by not wearing clothe…

Don't know why you're being downvoted. That is the logical conclusion.

Although, there's also a chance that those "blackmail gangs" never materialize. After all, you could already ten years ago pay cheap labor to create reasonably good fake images using Photoshop.

Re: Gemini 2.5 Flash Image

#487

Post by danpalmer »

carlosbaraza wrote: post #433
danpalmer wrote: post #405

Earlier quoted context omitted.

I've done ~20 prompts so far and not had one be rejected so far. What sort of things are you asking it to do? I've tried things like changing clothing and accessories on people.

Basic things like: "{uploaded image of a man} can you remove the glasses?" or "make everyone in the picture smile" or "open the eyes of everyone in the photo". Nothing that a human would consider "unsafe". I am based in EU and using Google AI Studio with all safety toggles set to "Off".

Strange. I wouldn't have thought the safety rules would differ by region, at least not for things like that. I uploaded a photo and asked to change the glasses and change the shirt and it did both with no problem.

I just went back to the chat and asked it to remove the glasses and it worked. Asking it to remove the shirt also succeeded, although a) this is a head and shoulders photo so nothing NSFW, and b) it didn't do a great job of guessing what my shoulders look like.

Re: Gemini 2.5 Flash Image

#490

Post by dostick »

vunderba wrote: post #300
gundmc wrote: post #249

Earlier quoted context omitted.

> Though fair warning, gpt-image-1 is borderline useless as an "editor" since it almost always changes the whole image instead of doing localized inpainting-style edits like Kontext, Qwen, or Nano-Banana. Came into this thread looking for this post. It's a great way to compare prompt adherence across models. Have you considered adding editing capabilities in a similar way given the recent trend of inpainting-style pr…

Adding a separate section for image editing capabilities is a great idea. I've done some experimentation with Qwen and Kontext and been pretty impressed, but it would be nice to see some side by sides now that we have essentially three models that are capable of highly localized in-painting without affecting the rest of the image. https://mordenstar.com/blog/edits-with-kontext

For editing prompts testing it is best to start with “only change …” to prevent model from changing everything. Even Nano banana does that.
Post reply on HN