Everyone is sleeping on Gemini 2.5 Flash Image / Nano Banana. As shown in the OP, it's substantially more powerful than most other models while at the same price-per-image, and due to its text encoder it can handle significantly larger and more nuanced prompts to get exactly what you want. I open-sourced a Python package for generating from it with examples (https://github.com/minimaxir/gemimg) and am currently working on a blog post with even more representative examples. Google also allows generations for free with aspect ratio control in AI Studio: https://aistudio.google.com/prompts/new_chat
That said, I am surprised Seedream 4.0 beat it in these tests.
I don't think people are really sleeping on it - nano-banana more or less went viral when it first came out. I'd argue that aside from the capabilities built into ChatGPT (with the Ghibli craze and whatnot) craze it's the best known image editing model.
It's a weird situation where the Gemini mobile app hit #2 on the App Stores because of free Nano Banana, but no one ever talks about it and most disclosed image generations I've seen are still ChatGPT.
> That said, I am surprised Seedream 4.0 beat it in these tests.
OP here. While Seedream did have the edge in adherence it also tends to introduce slight (but noticeable) color gradation changes. It's not a huge deal for me, but it might be for other people depending on their goals in which case NanoBanana would be the better choice.
Gemini is great when it gets it right, but in my experience, it sometimes gives you completely unexpected results and won't get it right no matter what. You can see that in some of the examples (eg the Girl with the pearl earring one). I'm constantly surprised by how good Flux is, but the tragedy is most people (me included) will just default to whatever they normally use (chatgpt and gemini, in my case), so it doesn't really matter that it's better
Agreed, to the point where I built my own UI where I can simultaneously generate three images and see a before/after. Most often only one of three is what I actually wanted.
Flux kontext quality is noticeably worse that nano banana, Qwen image 2509 and Seedream 4 most of the times. For pure image generation instead Hunyuan image is scarily good.
I was trying to use gemini 2.5 flash image / nano banana to tidy up a picture of my messy kitchen. It failed horribly on my first attempt. I was quite surprised how much trouble it had with this simple task (similar to cleaning up the street in the post). On my second attempt I had it first analyze the image to point out all the items that clutter the space, and then on a second prompt had it remove all those items. That worked much better, showing how important prompt engineering is.
Yeah, that's part of the reason I list the number of attempts as part of the stats for each model + respective prompt. It's a loose metric of how "steerable" a given model is, or put another way, how much I had to fight with it before we were able to get it to follow the prompt directives.
No one is sleeping on nano-banana/Gemini Flash, it's highly over-tuned for editing vs novel generation and maxes out at a pretty low resolution.
Seedream 4.0 is somewhat slept on for being 4k at the same cost as nano-banana. It's not as great at perfect 1:1 edits, but it's aesthetics are much better and it's significantly more reliable in production for me.
Models with LLM backbones/omni-modal models are not rare anymore, even Qwen Image Edit is out there for open-weights.
Meh, most Google AI products look great on paper but fail in actual real scenarios. And that ranges from their Claude Code clone to their buggy storybook thing which I really wanted to like.
I've been using Nano Banana quite a lot, and I know that it absolutely struggles at exterior architecture and landscaping. Getting it to add or remove things like curbs, walkways, gutters, etc, or to ask to match colors is almost futile.
I am trying Qwen Image Edit for turning day photos into night, mostly architecture etc. Most models are struggling, and Nano Banana misses edges and stuff, making the pictures align poorly.
It is fun being one of the elderly who set their standards back in distant 2022. All these demos look incredible compared to SD1, 2 & 3. We've entered a very different era where the models seem to actually understand both the prompt and the image instead of throwing paint at the wall in a statistically interesting manner.
I think this was fairly predictable, but as engineering improvements keep happening and the prompt adherence rate tightens up we're enjoying a wild era of unleashed creativity.
Here's a post I wrote on the Replicate blog putting these image editing models head-to-head. Generally, I found Qwen Image Edit to be the cheapest and fastest model that was also quite capable of most image editing tasks.
If I were to make an image editing app, this would be the model I'd choose.
I still feel varying the prompt text, number of tries, and varying strictness combined with only showing the result most liked dilute most of the value in these test. It would be better if there was one prompt 8/10 human editors understood and implemented correctly and then every model got 5 generation attempts with that exact prompt on different seeds or something. If it were about "who can create the best image with a given model" then I'd see it more, but most of it seems aimed at preventing that sort of thing and it ends up in an awkward middle zone.
E.g. Gemini 2.5 Flash is given extreme leeway with how much it edits the image and changes the style in "Girl with Pearl Earring" only to have OpenAI gpt-image-1 do a (comparatively) much better job yet still be declared failed after 8 attempts, while having been given fewer attempts than Seedream 4 (passed) and less than half the attempts of OmniGen2 (which still looks way farther off in comparison).
OP here. That's kind of the idea of listing the number of attempts alongside failure/successes. It's a loose metric for how "compliant" a model is - e.g. how much work you have to put it in order to get a nominally successful result.
Neat comparison. The only qualm I have is giving a pass on that last giraffe... it's not visibly any shorter, just bent awkwardly.
Even so, Gemini would lose by 1, but I found that I would often choose it as the winner(especially say, The Wave surfer). Would love to see a x/10 instead of pass/fail.
I do not use ai image generating much lately. It seemed like there was a burst of activity a year and half ago with self hosted models and using some localhost web guis. But now it seems like it is moving more and more to online hosted models.
Still, to my eye, ai generated images still feel a bit off when doing with real world photographs.
George's hair, for example, looks over the top, or brushed on.
The tree added to the sleeping person on the ground photo... the tree looks plastic or too homogenized.
> But now it seems like it is moving more and more to online hosted models.
It's mostly because image model size and required compute for both training and inference have grown faster than self-hosted compute capability for hobbyists. Sure, you can run Flux Kontext locally, but if you have to use a heavily quantized model and wait forever for the generation to actually run, the economics are harder to justify. That's not counting the "you can generate images from ChatGPT for free" factor.
> George's hair, for example, looks over the top, or brushed on.
IMO, the judge was being too generous with the passes for that test. The only one that really passes is Gemini 2.5 Flash Image:
Flux Kontext: In addition to the hair looking too slick, it does not match the VHS-esque color grading of the image.
Qwen-Image-Edit: The hair is too slick and the sharpness/saturation of the face unnecessarily increases.
Seedream 4: Color grading of the entire image changes, which is the case with most of the Seedream 4 edits shown in this post, and why I don't like it.
For 99% of my use cases I’ll just use ChatGPT or Gemini due to convenience. But if you want something with a specific style, Flux LoRAs are much better, in which case I’ll boot up the old 4090.
The economics 1000% do not justify me owning a GPU to do this. I just happen to own one.
Thank you for the pointer. I was struggling with Nanobanana for editing an image which it had created earlier, but Reve gave me the edit result exactly the way I wanted in the first pass.
My usecase: An image of a cartoon character, holding an object and looking at it. Wanted to edit so that the character no longer has the object in her hand and now looking towards the camera.
Result Nanobanana: At first pass it only removed the object that the character was holding, however there was no change in her eyeline, she was still looking down at her now empty hand. Second prompt explicitly asked to change the eyeline to look at camera. Unsuccessful. Third attempt asked the character to look towards ceiling. Success but unusable edit as I wanted the character to look at the camera.
Result Reve: At first attempt it gave me 4 options and all 4 are usable. It not only removed the object and changed the eyeline of the character to look at the camera, but it also made posture changes so that the empty hands were appropriately positioned, and now since the character is in a different situation (sans the object that was holding her attention) Reve posed the character in different ways which were very appropriate - which I didn't think of prompting for earlier (maybe because my focus was on immediate need - object removal and change in eyeline).
On a little more digging found this writeup which will make me to signup for their product.
Some might critique the prompts and say this or that would have done better, but they were the kind of prompt your dad would type in not knowing how to push the right buttons.
Good effort, somewhat marred by poor prompting. Passing in “the tower in the image is leaning to the right,” for example, is a big mistake. That context is already in the image, and passing that as a prompt will only make the model apt to lean the tower in the result.
Kontext is very good. Get yourself a 5060 ti 16GB and never have to pay for API calls again for this purpose, at least not when you have the time spare. If you need this sort of editing at the speed of gui-clicking + 10s, then you'll need to pay API tolls, or capex for > 5070/80.
You have to REALLY be into AI to do this for generation/API cost reasons (or willing to have this as a hacking project of the month expense). Even ignoring electricity, a 16 GB 5060 Ti is more expensive than 16,000 image generations. Assuming you do one every 15 seconds, that's 240,000 seconds -> more than 2 months of usage at an hour a day of generations.
If you've already got a decent GPU (or were going to get one anyways) then cost isn't really a consideration, it's just that you can already do it. For everyone else, you can probably get by just using things like Google's AI Studio for free.
I wonder how much longer those annoying stock photo database will continue. They are great for press photography and such. But stock pics of people in offices for a website are nothing, I would buy a min 3 month subscription for anymore
As generative AI eats away at the high royalty, restrictive license, consent evading, stereotype reinforcing business model of stock photo companies, it will be a challenge to resist the schadenfreude.
I'm pretty sure that "replace the homeless man with a park bench" image was a reference to some TV show making a gentrification joke, but I can't put my finger on it. Anyone recall?
This is not the point of this post, but is anyone else getting tired of this front end style that Claude creates? I see it on web apps everywhere and (just like with AI writing and images) I get that funny "is this slop?" feeling
Actually 5. The gemini result is pretty correct. And for that test, IMO only gemini properly preserved the original aesthetic. All others don't have the dark/scary mood.
Is there anything like this comparison for nsfw images? I'm married to a boudoir photographer who sometimes wants to use ai tools for things, and they are all _awfull_ if there is nudity on photos. It's like some sort of neo puritanism has taken over.
That said, I am surprised Seedream 4.0 beat it in these tests.
reply