I'm no image gen expert but these prompts are downright terrible even by my standards.
Are you really complaining that ", from the British Museum." leads to it a painting in the actual British Museum? Just remove the sentence, and you'll be fine. Now good luck trying to make Midjourney place the image at the museum!
I'm a paying MJ user and am impressed by Nano Banana. They're different models. They each serve their purpose.
This analysis is just noise. Yawn.
Ironically, even an LLM with its fake reasoning capabilities can point out the issue with the prompts if you ask it to critique this article.
It is interesting what the nbp model takes away from the prompt, though
Eg instead of focusing on the artist, it focuses on the location
This makes sense! I imagine it was trained in some sort of rlvr like way where you give it a prompt and then interrogate "does this image ..." (where each question examines a different aspect of the prompt)
It's obviously an incredible model. I think there's a limit to how useful another article praising it is in contrast with one expressing frustration
I would also welcome someone writing a short takedown where they fix the prompts and get better-than-2022 results from nbp
> I would also welcome someone writing a short takedown where they fix the prompts and get better-than-2022 results from nbp
NBP (and the new ChatGPT generator) are integrated with LLMs to various degrees, so seems like the obvious starting point is a reverse approach: ask them to describe the old images which has the esthetics that Fernando Borretti likes, and start generating from those prompts. If you can recover the old images, then it was just a prompting issue. ("Sampling can show the presence of knowledge but not the absence.") If you can't even with their own 'native' descriptions, then that points to mode-collapse (especially all of the 'esthetic tuning' like DPO everyone does now) as being the biggest problem.
These sorts of prompts used to be quite important when DALL-E was new. I do feel like a lot of the article is just that prompts should be written differently though I think there’s some truth in the idea that nanobanana feels less artistic in some ways.
The author is using special prompts exploiting flaws of the old models, and doesn't like that new models interpret the hacks literally instead.
The new models have prompt adherence precise enough to distinguish
what "British Museum" or "auction at Christie's" is from the art itself, instead of blending a bag of words together into a single vector and implicitly copying all of the features of all works containing "museum" or "ArtStation" in their description.
> It's the sound of failure: so much modern art is the sound of things going out of control, of a medium pushing to its limits and breaking apart. The distorted guitar sound is the sound of something too loud for the medium supposed to carry it. The blues singer with the cracked voice is the sound of an emotional cry too powerful for the throat that releases it. The excitement of grainy film, of bleached-out black and white, is the excitement of witnessing events too momentous for the medium assigned to record them.
The OP would likely prefer Disco Diffusion if they want their art to remain coarse. Modern models possess advanced spatial understanding and adhere strictly to prompts, whereas the OP is using unstructured inputs better suited for older models with CLIP or T5 encoders that lack that spatial awareness. These legacy prompting styles are incompatible with Gen3 models that utilize VLMs as text encoders. If the OP wants to explore modern architecture, they should use Flux.2 with a LoRA or perhaps a coarser model like Zit if they prefer to rely solely on text conditioning. Nano Banana Pro requires extremely long and distinctive prompting to achieve specific aesthetics. His blog post shows a lack of understanding and a lack of adaption to modern architecture which would be fine if it wasn't that dismissive.
Midjourney is optimized for beautiful images, while Nano Banana is optimized for better prompt adherence and (more importantly) image editing. It should be obvious for anyone who spent 20 minutes trying out these models.
If your goal is to replace human designers with cheaper options[0], Nano Banana / ChatGPT is indefinitely more useful than Midjourney. I'd argue Midjourney is completely useless except for social media clout or making concept art for experienced designers.
[0]: A hideous goal, I know. But we shouldn't sugarcoat it: this is what underpin the whole AI scheme now.
It is what has underpinned all of human progress towards automation. It isn't a bad thing. Every time we automate something the luddites cry out about the coming mass unemployment. It has never happened.
>Every time we automate something the luddites cry out about the coming mass unemployment. It has never happened
It has happened each and every time, it just haven't affected you personally. Starting of course with the original luddites - they didn't complain out of some philosophical opposition to automation.
Each time in changes like this a huge number of people lost their jobs and took big hits in their quality of life. The "new jobs", when they arrive, arrive for others.
This includes the post 1990s switch to service and digital economies and outsourcing, which obliterated countless factory towns in the US - and those people didn't magically turn to coders and creatives. At best they took unemployment, big decreases in job prospects, shitty "gig" economy jobs, or, well, worse, including alcohol and opiods.
With AI it's even worse, since it has the capacity to replace jobs without adding new ones, or a tiny handful at a hugely smaller rate.
It literally happens every single time - people DO lose jobs. They might get new jobs, but they definitely lose their old ones.
And not everyone gets new jobs, because usually the new job is fundamentally different and might not be compatible with the person or their original desire out of their employment.
Except all the manufacturing jobs got shipped overseas and now those people are Walmart greeters or similar unskilled labor. Having a shit job isn’t unemployment but it’s not a huge step up
While I don’t disagree with the author, these are simply two completely different tools with different use cases. Nano Banana Pro throws out fantastic images you can actually use in your marketing right away. It’s not an art tool - it’s a business tool
As long as the older tools still exist to make art, I don’t see what the problem is. Use NBP to make your marketing pics, MJv2 for your art
You’re definitely on to something, people wouldn’t criticize as much as they are otherwise, they’d ignore it.
I think the whole point is that in optimizing for instruction following and boring realism we’ve lost what could have been some unique artistic elements of a new medium, but anyway.
Readit News
Are you really complaining that ", from the British Museum." leads to it a painting in the actual British Museum? Just remove the sentence, and you'll be fine. Now good luck trying to make Midjourney place the image at the museum!
I'm a paying MJ user and am impressed by Nano Banana. They're different models. They each serve their purpose.
This analysis is just noise. Yawn.
Ironically, even an LLM with its fake reasoning capabilities can point out the issue with the prompts if you ask it to critique this article.
Eg instead of focusing on the artist, it focuses on the location
This makes sense! I imagine it was trained in some sort of rlvr like way where you give it a prompt and then interrogate "does this image ..." (where each question examines a different aspect of the prompt)
It's obviously an incredible model. I think there's a limit to how useful another article praising it is in contrast with one expressing frustration
I would also welcome someone writing a short takedown where they fix the prompts and get better-than-2022 results from nbp
NBP (and the new ChatGPT generator) are integrated with LLMs to various degrees, so seems like the obvious starting point is a reverse approach: ask them to describe the old images which has the esthetics that Fernando Borretti likes, and start generating from those prompts. If you can recover the old images, then it was just a prompting issue. ("Sampling can show the presence of knowledge but not the absence.") If you can't even with their own 'native' descriptions, then that points to mode-collapse (especially all of the 'esthetic tuning' like DPO everyone does now) as being the biggest problem.
The new models have prompt adherence precise enough to distinguish what "British Museum" or "auction at Christie's" is from the art itself, instead of blending a bag of words together into a single vector and implicitly copying all of the features of all works containing "museum" or "ArtStation" in their description.
> A painting sold at Sotheby's
and
> A painting in the style of something that would be sold at Sotheby's
convey very different meaning (to me).
Deleted Comment
> It's the sound of failure: so much modern art is the sound of things going out of control, of a medium pushing to its limits and breaking apart. The distorted guitar sound is the sound of something too loud for the medium supposed to carry it. The blues singer with the cracked voice is the sound of an emotional cry too powerful for the throat that releases it. The excitement of grainy film, of bleached-out black and white, is the excitement of witnessing events too momentous for the medium assigned to record them.
> "By the time a whole technology exists for something it probably isn't the most interesting thing to be doing."
Here is an image from NBP with an adapted prompt for Italian futurism: https://imgur.com/a/4pN0I0R
and for Kowloon:
https://imgur.com/a/rDT8dfP
Midjourney is optimized for beautiful images, while Nano Banana is optimized for better prompt adherence and (more importantly) image editing. It should be obvious for anyone who spent 20 minutes trying out these models.
If your goal is to replace human designers with cheaper options[0], Nano Banana / ChatGPT is indefinitely more useful than Midjourney. I'd argue Midjourney is completely useless except for social media clout or making concept art for experienced designers.
[0]: A hideous goal, I know. But we shouldn't sugarcoat it: this is what underpin the whole AI scheme now.
It has happened each and every time, it just haven't affected you personally. Starting of course with the original luddites - they didn't complain out of some philosophical opposition to automation.
Each time in changes like this a huge number of people lost their jobs and took big hits in their quality of life. The "new jobs", when they arrive, arrive for others.
This includes the post 1990s switch to service and digital economies and outsourcing, which obliterated countless factory towns in the US - and those people didn't magically turn to coders and creatives. At best they took unemployment, big decreases in job prospects, shitty "gig" economy jobs, or, well, worse, including alcohol and opiods.
With AI it's even worse, since it has the capacity to replace jobs without adding new ones, or a tiny handful at a hugely smaller rate.
And not everyone gets new jobs, because usually the new job is fundamentally different and might not be compatible with the person or their original desire out of their employment.
Whether or not it comes to fruition, it's making large portions of society feel uneasy, and not just programmers, or artists, or teachers.
Like, you know... creating art.
It has happened every single time.
As long as the older tools still exist to make art, I don’t see what the problem is. Use NBP to make your marketing pics, MJv2 for your art
I think the whole point is that in optimizing for instruction following and boring realism we’ve lost what could have been some unique artistic elements of a new medium, but anyway.