I have been generating a few dozen images per day for storyboarding purposes. The more I try to perfect it, the easier it becomes to control these outputs and even keep the entire visual story as well as their characters consistent over a few dozen different scenes; while even controlling the time of day throughout the story. I am currently working with 7 layers prompts to control for environment, camera, subject, composition, light, colors and overall quality (it might be overkill, but it’s also experimenting).
I also created a small editing suite for myself where I can draw bounding boxes on images when they aren’t perfect, and have them fixed. Either just with a prompt or feeding them to Claude as image and then having it write the prompt to fix the issue for me (as a workflow on the api). It’s been quite a lot of fun to figure out what works. I am incredibly impressed by where this is all going.
Once you do have good storyboards. You can easily do start-to-end GenAI video generation (hopping from scene to scene) and bring them to life and build your own small visual animated universes.
We use nano banana extensively to build video storyboards, which we then turn into full motion video with a combination of img2vid models. It sounds like we're doing similar things, trying to keep images/characters/setting/style consistent across ~dozens of images (~minutes of video). You might like the product depending on what you're doing with the outputs! https://hypernatural.ai
I agree. Bruhcula? Something like that. He's a vampire, but also models and does stunts for Baywatch - too much color and vitality. Joan of Arc is way more pale.
Maybe a little mode collapse away from pale ugliness, not quite getting to the hints of unnatural and corpse-like features of a vampire - interesting what the limitations are. You'd probably have to spend quite a lot of time zeroing in, but Google's image models are supposed to have allowed smooth traversal of those feature spaces generally.
> The more I try to perfect it, the easier it becomes
I have the opposite experience, once it goes off track, its nearly impossible to bring it back on message
How much have you experimented with it? For some stories I may generate 5 image variations of 10-20 different scenes and then spend time writing down what worked and what did not; and running the generation again (this part is mostly for research). It’s certainly advancing my understanding over time and being able to control the output better. But I’m learning that it takes a huge amount of trial and error. So versioning prompts is definitely recommended, especially if you find some nuances that work for you.
> Nano Banana supports a context window of 32,768 tokens: orders of magnitude above T5’s 512 tokens and CLIP’s 77 tokens.
In my pipeline for generating highly complicated images (particularly comics [1]), I take advantage of this by sticking a Mistral 7b LLM in-between that takes a given prompt as an input and creates 4 variations of it before sending them all out.
> Surprisingly, Nano Banana is terrible at style transfer even with prompt engineering shenanigans, which is not the case with any other modern image editing model.
This is true - though I find it works better by providing a minimum of two images. The first image is intended to be transformed, and the second image is used as "stylistic aesthetic reference". This doesn't always work since you're still bound by the original training data, but it is sometimes more effective than attempting to type out a long flavor text description of the style.
I added a CLI to it (using Gemini CLI) and submitted a PR, you can run that like so:
GEMINI_API_KEY="..." \
uv run --with https://github.com/minimaxir/gemimg/archive/d6b9d5bbefa1e2ffc3b09086bc0a3ad70ca4ef22.zip \
python -m gemimg "a racoon holding a hand written sign that says I love trash"
I decided to avoid that purely to keep changes made to the package as minima as possible - adding a project.scripts means installing it adds a new command alias. My approach changes nothing other than making "python -m gemimg" do something useful.
I agree that a project.scripts would be good but that's a decision for the maintainer to take on separately!
The author went to great lengths about open source early on. I wonder if they'll cover the QwenEdit ecosystem.
I'm exceptionally excited about Chinese editing models. They're getting closer and closer to NanoBanana in terms of robustness, and they're open source. This means you can supply masks and kernels and do advanced image operations, integrate them into visual UIs, etc.
You can even fine tune them and create LoRAs that will do the style transferring tasks that Nano Banana falls flat on.
I don't like how closed the frontier US models are, and I hope the Chinese kick our asses.
That said, I love how easy it'll be to distill Nano Banana into a new model. You can pluck training data right out of it: ((any image, any instruction) -> completion) tuples.
The Qwen-Edit images from my GenAI Image Editing Showdown site were all generated from a ComfyUI workflow on my machine - it's shockingly good for an open-weight model. It was also the only model that scored a passing grade on the Van Halen M&M test (even compared against Nanobanana)
> I don't like how closed the frontier US models are, and I hope the Chinese kick our asses.
For imagegen, agreed. But for textgen, Kimi K2 thinking is by far the best chat model at the moment from my experience so far. Not even "one of the best", the best.
It has frontier level capability and the model was made very tastefully: it's significantly less sycophantic and more willing to disagree in a productive, reasonable way rather than immediately shutting you out. It's also way more funny at shitposting.
I'll keep using Claude a lot for multimodality and artifacts but much of my usage has shifted to K2. Claude's sycophancy is particular is tiresome. I don't use ChatGPT/Gemini because they hide the raw thinking tokens, which is really cringe.
Claude Sonnet 4.5 doesn't even feel sycophantic (in the 4o) way, it feels like it has BPD. It switches from desperately agreeing with you to moralizing lectures and then has a breakdown if you point out it's wrong about anything.
Also, yesterday I asked it a question and after the answer it complained about its poorly written system prompt to me.
They're really torturing their poor models over there.
I've been keeping an eye on Qwen-Edit/Wan 2.2 shenanigans and they are interesting: however actually running those types of models is too cumbersome and in the end unclear if it's actually worth it over the $0.04/image for Nano Banana.
I was skeptical about the notion of running similar models locally as well, but the person who did this (https://old.reddit.com/r/StableDiffusion/comments/1osi1q0/wa... ) swears that they generated it locally, just letting a single 5090 crunch away for a week.
If that's true, it seems worth getting past the 'cumbersome' aspects. This tech may not put Hollywood out of business, but it's clear that the process of filmmaking won't be recognizable in 10 years if amateurs can really do this in their basements today.
The author overlooked an interesting error in the second skull pancake image: the strawberry is on the right eye socket (to the left of the image), and the blackberry is on the left eye socket (to the right of the image)!
This looks like it's caused by 99% of the relative directions in image descriptions describing them from the looker's point of view, and that 99% of the ones that aren't it they refer to a human and not to a skull-shaped pancake.
I am a human, and I would have done the same thing as Nano Banana. If the user had wanted a strawberry in the skull's left eye, they should've said, "Put a strawberry in its left eye socket."
Exactly what I was thinking too. I'm a designer, and I'm used to receiving feedback and instructions. "The left eye socket" would to me refer to what I currently see in front of me, while "its left eye socket" instantly shift the perspective from me to the subject.
To be honest this is the sort of thing Nano Bannana is weak at in my experience. It's absolutely amazing - but doesn't understand left/right/up/down/shrink this/move this/rotate this etc.
See below to demonstrate this weakness with the same prompts as the article see the link below, which demonstrates that it is a model weakness and not just a language ambiguity:
Mmh, ime you need to discard the session/rewrite the failing prompt instead of continuing and correcting on failures. Once errors occur you've basically introduced a poison pill which will continuously make things to haywire. Spelling out what it did wrong is the most destructive thing you can do - at least in my experience
I admit I missed this, which is particularly embarrassing because I point out this exact problem with the character JSON later in the post.
For some offline character JSON prompts I ended up adding an additional "any mentions of left and right are from the character's perspective, NOT the camera's perspective" to the prompt, which did seem to improve success.
Came to make exactly the same comment. It was funny that the author specifically said that Nano Banana got all five edit prompts correct, rather than noting this discrepancy, which could be argued either way (although I think the "right eye" of a skull should be interpreted with respect to the skull's POV.)
Use Google AI Studio to submit requests, and to remove watermark, open browser development tools and right click on request to “watermark_4” image and select to block it. And from next generation there will be no watermark!
>Nano Banana is terrible at style transfer even with prompt engineering shenanigans
My context: I'm kind of fixated on visualizing my neighborhood as it would have appeared in the 18th century. I've been doing it in Sketchup, and then in Twinmotion, but neither of those produce "photorealistic" images... Twinmotion can get pretty close with a lot of work, but that's easier with modern architecture than it is with the more hand-made, brick-by-brick structures I'm modeling out.
As different AI image generators have emerged, I've tried them all in an effort to add the proverbial rough edges to snapshots of the models I've created, and it was not until Nano Banana that I ever saw anything even remotely workable.
Nano Banana manages to maintain the geometry of the scene, while applying new styles to it. Sometimes I do this with my Twinmotion renders, but what's really been cool to see is how well it takes a drawing, or engraving, or watercolor - and with as simple a prompt as "make this into a photo" it generates phenomenal results.
Similarly to the Paladin/Starbucks/Pirate example in the link though, I find that sometimes I need to misdirect a little bit, because if I'm peppering the prompt with details about the 18th century, I sometimes get a painterly image back. Instead, I'll tell it I want it to look like a photograph of a well preserved historic neighborhood, or a scene from a period film set in the 18th century.
As fantastic as the results can be, I'm not abandoning my manual modeling of these buildings and scenes. However, Nano Banana's interpretation of contemporary illustrations has helped me reshape how I think about some of the assumptions I made in my own models.
Fair enough! I suppose I've avoided that kind of "style transfer" for a variety of reasons, it hadn't even occurred to me that people were still interested in that. And I don't say that to open up debate on the topic, just explaining away my own ignorance/misinterpretation. Thanks
Yes, that is a serious skill. How many of the woes that we see is because people don't know what they want or are unable to describe it in such a way that others understand it.
I believe prompt engineer to properly convey how complex communication can be, when interacting with a multitude of perspectives, world views, assumptions, presumptions etc.
I believe it works well to counter the over-confidence that people have, from not paying attention to what gaps exist between what is said and what is meant.
Yes, obviously a role involving complex communication while interacting with a multitude of perspectives, world views, assumptions, presumptions, etc needs to be called "engineer."
That is why I always call technical writers "documentation engineers," why I call diplomats "international engineers," why I call managers "team engineers," and why I call historians "hindsight engineers."
Most designers can't, either. Defining a spec is a skill.
It's actually fairly difficult to put to words any specific enough vision such that it becomes understandable outside of your own head. This goes for pretty much anything, too.
Case in point, the final image in this post (the IP bonanza) took 28 iterations of the prompt text to get something maximally interesting, and why that one is very particular about the constraints it invokes, such as specifying "distinct" characters and specifying they are present from "left to right" because the model kept exploiting that ambiguity.
Hey! The author, thank you for this post! QQ, any idea roughly how much this experimentation cost you? I'm having trouble processing their image generation pricing I may just not be finding the right table. I'm just trying to understand if I do like 50 iterations at the quality in the post, how much is that going to cost me?
All generations in the post are $0.04/image (Nano Banana doesn't have a way to increase the resolution, yet), so you can do the math and assume that you can generated about 24 images per dollar: unlike other models, Nano Banana does charge for input tokens but it's neligible.
Discounting the testing around the character JSON which became extremely expensive due to extreme iteration/my own stupidity, I'd wager it took about $5 total including iteration.
I tried asking for a shot from a live-action remake of My Neighbor Totoro. This is a task I’ve been curious about for a while. Like Sonic, Totoro is the kind of stylized cartoon character that can’t be rendered photorealistically without a great deal of subjective interpretation, which (like in Sonic’s case) is famously easy to get wrong even for humans. Unlike Sonic, Totoro hasn’t had an actual live-action remake, so the model would have to come up with a design itself. I was wondering what it might produce – something good? something horrifying? Unfortunately, neither; it just produced a digital-art style image, despite being asked for a photorealistic one, and kept doing so even when I copied some of the keyword-stuffing from the post. At least it tried. I can’t test this with ChatGPT because it trips the copyright filter.
Nano Banana can be frustrating at times. Yesterday I tried to get it to do several edits to an image, and it would return back pretty much the same photo.
Things like: Convert the people to clay figures similar to what one would see in a claymation.
And it would think it did it, but I could not perceive any change.
After several attempts, I added "Make the person 10 years younger". Suddenly it made a clay figure of the person.
For images of people generated from scratch, Nano Banana always adds a background blur, it can't seem to create more realistic or candid images such as those taken via a point and shoot or smartphone, has anyone solved this sort of issue? It seems to work alright if you give it an existing image to edit however. I saw some other threads online about it but I didn't see anyone come up with solutions.
Those are rarely in the captions for the image. They'd have to extract the EXIF for photos and include it in recaptioning. Which they should be doing, but I doubt they thought about it.
I tried that but they don't seem to make much difference for whatever reason, you still can't get a crisp shot such as this [0] where the foreground and background details are all preserved (linked shot was taken with an iPhone which doesn't seem to do shallow depth of field unless you use their portrait mode).
In my own experience, nano banana still has the tendency to:
- make massive, seemingly random edits to images
- adjust image scale
- make very fine grained but pervasive detail changes obvious in an image diff
For instance, I have found that nano-banana will sporadically add a (convincing) fireplace to a room or new garage behind a house. This happens even with explicit "ALL CAPS" instructions not to do so. This happens sporadically, even when the temperature is set to zero, and makes it impossible to build a reliable app.
The "ALL CAPS" part of your comment got me thinking. I imagine most llms understand subtle meanings of upper case text use depending on context. But, as I understand it, ALL CAPS text will tokenize differently than lower case text. Is that right? In that case, won't the upper case be harder to understand and follow for most models since it's less common in datasets?
There's more than enough ALL CAPS text in the corpus of the entire internet, and enough semantic context associated with it for it to be intended to be in the imperative voice.
use it for technical design doc, where i sketch out something on paper and ask nano banana to make flow chat, its incredibly good at this kind of editing (also if want to borrow image from someone and change some bridges usually its hard its embedded image, but nano banana solves that)
Theres lots these models can do but I despise when people suggest they can do edits with "with only the necessary aspects changed".
No, that simply is not true. If you actually compare the before and after you can see it still regenerates all the details on the "unchanged" aspects. Texture, lighting, sharpness, even scale its all different even if varyingly similar to the original.
Sure they're cute for casual edits but it really pains me people suggesting these things are suitable replacements for actual photo editing. Especially when it comes to people, or details outside their training data theres a lot of nuance that can be lost as it regenerates them no matter how you prompt things.
Nano Banana is different and much better at edits without changing texture/lighting/sharpness/color balance, and I am someone that is extremely picky about it. That's why I add the note that Gemini 2.5 Flash is aware of segmentation masks, and that's my hunch why that's the case.
That's probably where things are headed and there are already products trying this (even photoshop already). Just like how code gen AI tools don't replace the entire file on every prompt iteration.
That's on my list of blog-post-worthy things to test, namely text rendering to image in Python directly and passing both input images to the model for compositing.
> It’s one of the best results I’ve seen for this particular test, and it’s one that doesn’t have obvious signs of “AI slop” aside from the ridiculous premise.
It’s pretty good, but one conspicuous thing is that most of the blueberries are pointing upwards.
I really wish that real expert stuff, like how to do controlnet, use regional prompting, or most other advanced ComfyUI stuff got upvoted to the top instead.
The kicker for nano banana is not prompt adherence which is a really nice to have but the fact that it's either working on pixel space or with a really low spatial scaling. It's the only model that doesn't kill your details because of vae encode/decode.
You CREATED something, and I like to think that creating things that I love and enjoy and that others can love and enjoy makes creating things worth it.
Not really since "prompt engineering" can be tossed in the same pile as "vibe coding." Just people coping with not developing the actual skills to produce the desired products.
Try getting a small model to do what you want quickly with high accuracy, high quality, etc, and using few tokens per request. You'll find out that prompt engineering is real and matters.
Gemini is instructed to reply in images first, and if it thinks, to think using the image thinking tags. It cannot seemingly be prompted to show verbatim the result 4+5 without showing the answer 4+5=9. Of course it can show whatever exact text that you want, the question is, does it prompt rewrite (no) or do something else (yes)?
We can do the same exercises with Flux Kontext for editing versus Flash-2.5, if you think that editing is somehow unique in this regard.
Is prompt rewriting "thinking"? My point is, this article can't answer that question without dElViNg into the nuances of what multi-modal models really are.
I have been generating a few dozen images per day for storyboarding purposes. The more I try to perfect it, the easier it becomes to control these outputs and even keep the entire visual story as well as their characters consistent over a few dozen different scenes; while even controlling the time of day throughout the story. I am currently working with 7 layers prompts to control for environment, camera, subject, composition, light, colors and overall quality (it might be overkill, but it’s also experimenting).
I also created a small editing suite for myself where I can draw bounding boxes on images when they aren’t perfect, and have them fixed. Either just with a prompt or feeding them to Claude as image and then having it write the prompt to fix the issue for me (as a workflow on the api). It’s been quite a lot of fun to figure out what works. I am incredibly impressed by where this is all going.
Once you do have good storyboards. You can easily do start-to-end GenAI video generation (hopping from scene to scene) and bring them to life and build your own small visual animated universes.
We use nano banana extensively to build video storyboards, which we then turn into full motion video with a combination of img2vid models. It sounds like we're doing similar things, trying to keep images/characters/setting/style consistent across ~dozens of images (~minutes of video). You might like the product depending on what you're doing with the outputs! https://hypernatural.ai
Yes we are definitely doing the same! For now I’m just familiarizing myself in this space technically and conceptually. https://edwin.genego.io/blog
Your "Dracula" character is possibly the least vampiric Dracula I've ever seen tbh
I agree. Bruhcula? Something like that. He's a vampire, but also models and does stunts for Baywatch - too much color and vitality. Joan of Arc is way more pale.
Maybe a little mode collapse away from pale ugliness, not quite getting to the hints of unnatural and corpse-like features of a vampire - interesting what the limitations are. You'd probably have to spend quite a lot of time zeroing in, but Google's image models are supposed to have allowed smooth traversal of those feature spaces generally.
That looks exactly like the photos on a Spirit Halloween costume.
lol you can make your own Dracula if you want him to look different: https://hypernatural.ai/characters
> The more I try to perfect it, the easier it becomes I have the opposite experience, once it goes off track, its nearly impossible to bring it back on message
How much have you experimented with it? For some stories I may generate 5 image variations of 10-20 different scenes and then spend time writing down what worked and what did not; and running the generation again (this part is mostly for research). It’s certainly advancing my understanding over time and being able to control the output better. But I’m learning that it takes a huge amount of trial and error. So versioning prompts is definitely recommended, especially if you find some nuances that work for you.
Good read minimaxir! From the article:
> Nano Banana supports a context window of 32,768 tokens: orders of magnitude above T5’s 512 tokens and CLIP’s 77 tokens.
In my pipeline for generating highly complicated images (particularly comics [1]), I take advantage of this by sticking a Mistral 7b LLM in-between that takes a given prompt as an input and creates 4 variations of it before sending them all out.
> Surprisingly, Nano Banana is terrible at style transfer even with prompt engineering shenanigans, which is not the case with any other modern image editing model.
This is true - though I find it works better by providing a minimum of two images. The first image is intended to be transformed, and the second image is used as "stylistic aesthetic reference". This doesn't always work since you're still bound by the original training data, but it is sometimes more effective than attempting to type out a long flavor text description of the style.
[1] - https://mordenstar.com/portfolio/zeno-paradox
I like the Python library that accompanies this: https://github.com/minimaxir/gemimg
I added a CLI to it (using Gemini CLI) and submitted a PR, you can run that like so:
Result in this comment: https://github.com/minimaxir/gemimg/pull/7#issuecomment-3529...@simonw: slight tangent but super curious how you managed to generate the preview of that gemini-cli terminal session gist - https://gistpreview.github.io/?17290c1024b0ef7df06e9faa4cb37...
is this just a manual copy/paste into a gist with some html css styling; or do you have a custom tool à la amp-code that does this more easily?
I used this tool: https://tools.simonwillison.net/terminal-to-html
I made a video about building that here: https://simonwillison.net/2025/Oct/23/claude-code-for-web-vi...
It works much better with Claude Code and Codex CLI because they don't mess around with scrolling in the same way as Gemini CLI does.
very cool. frequently, i want to share my prompt + session output; this will make that super easy! thanks again for sharing!
Any reason for not also adding a project.scripts entry for pyproject.toml? That way the CLI (great idea btw) could be installed as a tool by uv.
I decided to avoid that purely to keep changes made to the package as minima as possible - adding a project.scripts means installing it adds a new command alias. My approach changes nothing other than making "python -m gemimg" do something useful.
I agree that a project.scripts would be good but that's a decision for the maintainer to take on separately!
The author went to great lengths about open source early on. I wonder if they'll cover the QwenEdit ecosystem.
I'm exceptionally excited about Chinese editing models. They're getting closer and closer to NanoBanana in terms of robustness, and they're open source. This means you can supply masks and kernels and do advanced image operations, integrate them into visual UIs, etc.
You can even fine tune them and create LoRAs that will do the style transferring tasks that Nano Banana falls flat on.
I don't like how closed the frontier US models are, and I hope the Chinese kick our asses.
That said, I love how easy it'll be to distill Nano Banana into a new model. You can pluck training data right out of it: ((any image, any instruction) -> completion) tuples.
The Qwen-Edit images from my GenAI Image Editing Showdown site were all generated from a ComfyUI workflow on my machine - it's shockingly good for an open-weight model. It was also the only model that scored a passing grade on the Van Halen M&M test (even compared against Nanobanana)
https://genai-showdown.specr.net/image-editing
> I don't like how closed the frontier US models are, and I hope the Chinese kick our asses.
For imagegen, agreed. But for textgen, Kimi K2 thinking is by far the best chat model at the moment from my experience so far. Not even "one of the best", the best.
It has frontier level capability and the model was made very tastefully: it's significantly less sycophantic and more willing to disagree in a productive, reasonable way rather than immediately shutting you out. It's also way more funny at shitposting.
I'll keep using Claude a lot for multimodality and artifacts but much of my usage has shifted to K2. Claude's sycophancy is particular is tiresome. I don't use ChatGPT/Gemini because they hide the raw thinking tokens, which is really cringe.
Claude Sonnet 4.5 doesn't even feel sycophantic (in the 4o) way, it feels like it has BPD. It switches from desperately agreeing with you to moralizing lectures and then has a breakdown if you point out it's wrong about anything.
Also, yesterday I asked it a question and after the answer it complained about its poorly written system prompt to me.
They're really torturing their poor models over there.
It rubs the data on its skin or else it gets the prompt again!
I've been keeping an eye on Qwen-Edit/Wan 2.2 shenanigans and they are interesting: however actually running those types of models is too cumbersome and in the end unclear if it's actually worth it over the $0.04/image for Nano Banana.
Takes a couple mouse clicks in ComfyUI
On that subject - ComfyUI is not the future of image gen. It's an experimental rope bridge.
Adobe's conference last week points to the future of image gen. Visual tools where you mold images like clay. Hands on.
Comfy appeals to the 0.01% that like toolkits like TouchDesigner, Nannou, and ShaderToy.
I was skeptical about the notion of running similar models locally as well, but the person who did this (https://old.reddit.com/r/StableDiffusion/comments/1osi1q0/wa... ) swears that they generated it locally, just letting a single 5090 crunch away for a week.
If that's true, it seems worth getting past the 'cumbersome' aspects. This tech may not put Hollywood out of business, but it's clear that the process of filmmaking won't be recognizable in 10 years if amateurs can really do this in their basements today.
The author overlooked an interesting error in the second skull pancake image: the strawberry is on the right eye socket (to the left of the image), and the blackberry is on the left eye socket (to the right of the image)!
This looks like it's caused by 99% of the relative directions in image descriptions describing them from the looker's point of view, and that 99% of the ones that aren't it they refer to a human and not to a skull-shaped pancake.
I am a human, and I would have done the same thing as Nano Banana. If the user had wanted a strawberry in the skull's left eye, they should've said, "Put a strawberry in its left eye socket."
Exactly what I was thinking too. I'm a designer, and I'm used to receiving feedback and instructions. "The left eye socket" would to me refer to what I currently see in front of me, while "its left eye socket" instantly shift the perspective from me to the subject.
I picked up on that also. I feel that a lot of humans would also get confused about whether you mean the eye on the left, or the subject's left eye.
To be honest this is the sort of thing Nano Bannana is weak at in my experience. It's absolutely amazing - but doesn't understand left/right/up/down/shrink this/move this/rotate this etc.
See below to demonstrate this weakness with the same prompts as the article see the link below, which demonstrates that it is a model weakness and not just a language ambiguity:
https://gemini.google.com/share/a024d11786fc
Mmh, ime you need to discard the session/rewrite the failing prompt instead of continuing and correcting on failures. Once errors occur you've basically introduced a poison pill which will continuously make things to haywire. Spelling out what it did wrong is the most destructive thing you can do - at least in my experience
Almost no image/video models can do "upside-down" either.
I admit I missed this, which is particularly embarrassing because I point out this exact problem with the character JSON later in the post.
For some offline character JSON prompts I ended up adding an additional "any mentions of left and right are from the character's perspective, NOT the camera's perspective" to the prompt, which did seem to improve success.
Came to make exactly the same comment. It was funny that the author specifically said that Nano Banana got all five edit prompts correct, rather than noting this discrepancy, which could be argued either way (although I think the "right eye" of a skull should be interpreted with respect to the skull's POV.)
Use Google AI Studio to submit requests, and to remove watermark, open browser development tools and right click on request to “watermark_4” image and select to block it. And from next generation there will be no watermark!
I was kind of surprised by this line:
>Nano Banana is terrible at style transfer even with prompt engineering shenanigans
My context: I'm kind of fixated on visualizing my neighborhood as it would have appeared in the 18th century. I've been doing it in Sketchup, and then in Twinmotion, but neither of those produce "photorealistic" images... Twinmotion can get pretty close with a lot of work, but that's easier with modern architecture than it is with the more hand-made, brick-by-brick structures I'm modeling out.
As different AI image generators have emerged, I've tried them all in an effort to add the proverbial rough edges to snapshots of the models I've created, and it was not until Nano Banana that I ever saw anything even remotely workable.
Nano Banana manages to maintain the geometry of the scene, while applying new styles to it. Sometimes I do this with my Twinmotion renders, but what's really been cool to see is how well it takes a drawing, or engraving, or watercolor - and with as simple a prompt as "make this into a photo" it generates phenomenal results.
Similarly to the Paladin/Starbucks/Pirate example in the link though, I find that sometimes I need to misdirect a little bit, because if I'm peppering the prompt with details about the 18th century, I sometimes get a painterly image back. Instead, I'll tell it I want it to look like a photograph of a well preserved historic neighborhood, or a scene from a period film set in the 18th century.
As fantastic as the results can be, I'm not abandoning my manual modeling of these buildings and scenes. However, Nano Banana's interpretation of contemporary illustrations has helped me reshape how I think about some of the assumptions I made in my own models.
You can't take a highly artistic image and supply it as a style reference. Nano Banana can't generalize to anything not in its training.
Fair enough! I suppose I've avoided that kind of "style transfer" for a variety of reasons, it hadn't even occurred to me that people were still interested in that. And I don't say that to open up debate on the topic, just explaining away my own ignorance/misinterpretation. Thanks
"prompt engineered"...i.e. by typing in what you want to see.
Yes, that is a serious skill. How many of the woes that we see is because people don't know what they want or are unable to describe it in such a way that others understand it. I believe prompt engineer to properly convey how complex communication can be, when interacting with a multitude of perspectives, world views, assumptions, presumptions etc. I believe it works well to counter the over-confidence that people have, from not paying attention to what gaps exist between what is said and what is meant.
Yes, obviously a role involving complex communication while interacting with a multitude of perspectives, world views, assumptions, presumptions, etc needs to be called "engineer."
That is why I always call technical writers "documentation engineers," why I call diplomats "international engineers," why I call managers "team engineers," and why I call historians "hindsight engineers."
Not all models can actually do that if your prompt is particular
Most designers can't, either. Defining a spec is a skill.
It's actually fairly difficult to put to words any specific enough vision such that it becomes understandable outside of your own head. This goes for pretty much anything, too.
https://habitatchronicles.com/2004/04/you-cant-tell-people-a...
Yep, knowing how and what to ask is a skill.
For anything, even back in the "classical" search days.
... and then iterating on that prompt many times, based on your accumulated knowledge of how best to prompt that particular model.
Case in point, the final image in this post (the IP bonanza) took 28 iterations of the prompt text to get something maximally interesting, and why that one is very particular about the constraints it invokes, such as specifying "distinct" characters and specifying they are present from "left to right" because the model kept exploiting that ambiguity.
Hey! The author, thank you for this post! QQ, any idea roughly how much this experimentation cost you? I'm having trouble processing their image generation pricing I may just not be finding the right table. I'm just trying to understand if I do like 50 iterations at the quality in the post, how much is that going to cost me?
All generations in the post are $0.04/image (Nano Banana doesn't have a way to increase the resolution, yet), so you can do the math and assume that you can generated about 24 images per dollar: unlike other models, Nano Banana does charge for input tokens but it's neligible.
Discounting the testing around the character JSON which became extremely expensive due to extreme iteration/my own stupidity, I'd wager it took about $5 total including iteration.
We understand now that we interface with LLMs using natural and unnatural language as the user interface.
This is a very different fuzzy interface compared to programming languages.
There will be techniques better or worse at interfacing.
This is what the term prompt engineering is alluding to since we don’t have the full suite of language to describe this yet.
Used to be called Google Fu
right? 15 months ago in image models you used to have to designate rendering specifications, and know the art of negative prompting
now you can really use natural language and people want to debate you about how poor they are at articulating a shared concepts, amazing
it's like the people are regressing and the AI is improving
"amenable to highly specific and granular instruction"
I tried asking for a shot from a live-action remake of My Neighbor Totoro. This is a task I’ve been curious about for a while. Like Sonic, Totoro is the kind of stylized cartoon character that can’t be rendered photorealistically without a great deal of subjective interpretation, which (like in Sonic’s case) is famously easy to get wrong even for humans. Unlike Sonic, Totoro hasn’t had an actual live-action remake, so the model would have to come up with a design itself. I was wondering what it might produce – something good? something horrifying? Unfortunately, neither; it just produced a digital-art style image, despite being asked for a photorealistic one, and kept doing so even when I copied some of the keyword-stuffing from the post. At least it tried. I can’t test this with ChatGPT because it trips the copyright filter.
Nano Banana can be frustrating at times. Yesterday I tried to get it to do several edits to an image, and it would return back pretty much the same photo.
Things like: Convert the people to clay figures similar to what one would see in a claymation.
And it would think it did it, but I could not perceive any change.
After several attempts, I added "Make the person 10 years younger". Suddenly it made a clay figure of the person.
The first request is a style transfer, which is why I included the Ghibli failure example.
I've gotten it to make Ghibli transfers by responding to the initial attempt with "I can barely tell the difference. Make the effect STRONGER."
I found this well written. I read it start to finish. The author does a good job of taking you through their process
For images of people generated from scratch, Nano Banana always adds a background blur, it can't seem to create more realistic or candid images such as those taken via a point and shoot or smartphone, has anyone solved this sort of issue? It seems to work alright if you give it an existing image to edit however. I saw some other threads online about it but I didn't see anyone come up with solutions.
Maybe try including “f/16” or “f/22” as those are likely to be in the training set for long depth of field photos.
Those are rarely in the captions for the image. They'd have to extract the EXIF for photos and include it in recaptioning. Which they should be doing, but I doubt they thought about it.
I tried that but they don't seem to make much difference for whatever reason, you still can't get a crisp shot such as this [0] where the foreground and background details are all preserved (linked shot was taken with an iPhone which doesn't seem to do shallow depth of field unless you use their portrait mode).
[0] https://www.lux.camera/content/images/size/w1600/2024/09/IMG...
In my own experience, nano banana still has the tendency to:
- make massive, seemingly random edits to images - adjust image scale - make very fine grained but pervasive detail changes obvious in an image diff
For instance, I have found that nano-banana will sporadically add a (convincing) fireplace to a room or new garage behind a house. This happens even with explicit "ALL CAPS" instructions not to do so. This happens sporadically, even when the temperature is set to zero, and makes it impossible to build a reliable app.
Has anyone had a better experience?
The "ALL CAPS" part of your comment got me thinking. I imagine most llms understand subtle meanings of upper case text use depending on context. But, as I understand it, ALL CAPS text will tokenize differently than lower case text. Is that right? In that case, won't the upper case be harder to understand and follow for most models since it's less common in datasets?
There's more than enough ALL CAPS text in the corpus of the entire internet, and enough semantic context associated with it for it to be intended to be in the imperative voice.
Shouldn't all caps normalised to tokens like low caps? There are no separate tokens for all caps and low caps in Llama, or at least not in the past.
Looking at the tokenizer for the older Llama 2 model, the tokenizer has capital letters in it: https://huggingface.co/meta-llama/Llama-2-7b-hf
I work on the PixLab prompt based photo editor (https://editor.pixlab.io), and it follows exactly what you type with explicit CAPS.
>> "The image style is definitely closer to Vanity Fair (the photographer is reflected in his breastplate!)"
I didn't expect that. I would have definitely counted that as a "probably real" tally mark if grading an image.
use it for technical design doc, where i sketch out something on paper and ask nano banana to make flow chat, its incredibly good at this kind of editing (also if want to borrow image from someone and change some bridges usually its hard its embedded image, but nano banana solves that)
It's really cool how good of a job it did rendering a page given its HTML code. I was not expecting it to do nearly as well.
Same. This must have training from sites that show html next to screenshots of the pages.
The blueberry and strawberry are not actually where they prompted.
Theres lots these models can do but I despise when people suggest they can do edits with "with only the necessary aspects changed".
No, that simply is not true. If you actually compare the before and after you can see it still regenerates all the details on the "unchanged" aspects. Texture, lighting, sharpness, even scale its all different even if varyingly similar to the original.
Sure they're cute for casual edits but it really pains me people suggesting these things are suitable replacements for actual photo editing. Especially when it comes to people, or details outside their training data theres a lot of nuance that can be lost as it regenerates them no matter how you prompt things.
Even if you
Nano Banana is different and much better at edits without changing texture/lighting/sharpness/color balance, and I am someone that is extremely picky about it. That's why I add the note that Gemini 2.5 Flash is aware of segmentation masks, and that's my hunch why that's the case.
Could you just mask out the area you wish to change in more advanced tools, or is there something in the model itself which would prevent this?
That's probably where things are headed and there are already products trying this (even photoshop already). Just like how code gen AI tools don't replace the entire file on every prompt iteration.
That is true for gpt-image-1 but not nano-banana. They can do masked image changes
Nano banana has a really low spatial scaling and doesn't affect details like other models.
> Nano Banana is still bad at rendering text perfectly/without typos as most image generation models.
I figured that if you write the text in Google docs and share the screenshot with banana it will not make any spelling mistake.
So, use something like "can you write my name on this Wimbledon trophy, both images are attached. Use them" will work.
Google's example documentation for Nano Banana does demo that pipeline: https://ai.google.dev/gemini-api/docs/image-generation#pytho...
That's on my list of blog-post-worthy things to test, namely text rendering to image in Python directly and passing both input images to the model for compositing.
Created a tool you can try out!! sorry to self-plug but I launch on Product Hunt next week that lets you do this:)
www.brandimagegen.com
if you want a premium account to try out, you can find my email in my bio!!
Well, I just asked it for a 13-sided irregular polygon (is it that hard?)…
https://imgur.com/a/llN7V0W
I don't feel like I should search for "nano banana" on my work laptop
how did you do NSFW?
> It’s one of the best results I’ve seen for this particular test, and it’s one that doesn’t have obvious signs of “AI slop” aside from the ridiculous premise.
It’s pretty good, but one conspicuous thing is that most of the blueberries are pointing upwards.
Cute. What’s the use case?
NSFW, mostly
I really wish that real expert stuff, like how to do controlnet, use regional prompting, or most other advanced ComfyUI stuff got upvoted to the top instead.
The kicker for nano banana is not prompt adherence which is a really nice to have but the fact that it's either working on pixel space or with a really low spatial scaling. It's the only model that doesn't kill your details because of vae encode/decode.
I haven't paid much attention to image generation models (not my area of interest), but these examples are shockingly good.
Another thing it can't do is remove reflections in windows, it's nearly a no-op.
I'm getting annoyed by using "prompt engineered" as a verb. Does this mean I'm finally old and bitter?
(Do we say we software engineered something?)
I think it’s meant to be engineering in the same sense as “social engineering”.
You're definitely old and bitter, welcome to it.
You CREATED something, and I like to think that creating things that I love and enjoy and that others can love and enjoy makes creating things worth it.
Don't get me wrong, I have nothing against using AI as an expression of creativity :)
Create? So I have created all that code I'm running on my site, yes is bad I know, but thank you very much! Such creative guy I was!
Not really since "prompt engineering" can be tossed in the same pile as "vibe coding." Just people coping with not developing the actual skills to produce the desired products.
Try getting a small model to do what you want quickly with high accuracy, high quality, etc, and using few tokens per request. You'll find out that prompt engineering is real and matters.
Couldn't care less. I don't need to know how to do literally everything. AI fills in my gaps and I'm a ton more productive.
I wouldn't bother trying to convince people who are upset that others have figured out a way to use LLMs. It's not logical.
No it means you can still discern what is BS.
lots of words
okay, look at imagen 4 ultra:
https://aistudio.google.com/app/prompts?state=%7B%22ids%22:%...
In this link, Imagen is instructed to render the verbatim prompt “the result of 4+5”, which shows that text, and not instructed, which renders “4+5=9”
Is Imagen thinking?
Let's compare to gemini 2.5 flash image (nano banana):
look carefully at the system prompt here: https://aistudio.google.com/app/prompts?state=%7B%22ids%22:%...
Gemini is instructed to reply in images first, and if it thinks, to think using the image thinking tags. It cannot seemingly be prompted to show verbatim the result 4+5 without showing the answer 4+5=9. Of course it can show whatever exact text that you want, the question is, does it prompt rewrite (no) or do something else (yes)?
compare to ideogram, with prompt rewriting: https://ideogram.ai/g/GRuZRTY7TmilGUHnks-Mjg/0
without prompt rewriting: https://ideogram.ai/g/yKV3EwULRKOu6LDCsSvZUg/2
We can do the same exercises with Flux Kontext for editing versus Flash-2.5, if you think that editing is somehow unique in this regard.
Is prompt rewriting "thinking"? My point is, this article can't answer that question without dElViNg into the nuances of what multi-modal models really are.
Can you provide screenshots or links that don't require login
sorry, but I don't understand you post. those links don't work.