It feels bolted on to an otherwise cool thing. Even angling the board, or smudging, dimming color, localizing language, etc. All would make it feel like it fits.
I know this is going to sound ridiculous, but the way the dithering is used feels very "generated" and chaotic.
Human pixel art tends to be very deliberate about the exact colors of every pixel and their placement tends to follow idealized (but not necessarily geometric) shapes.
The last time I saw dithering this sloppy was decades ago when people would scan photos as low color images. It is its own vibe, but it's a cold and rough one. These artifacts feel more "realistic", but it felt like slop back then too.
For an example of just how much of a rabbit hole this is, Lucas Pope's forum post[0] about the dithering in Obra Dinn shows how much time can be spent on getting it right.
To be fair, most of the complication there is because it's dynamically generated in 3D, where moving the camera makes the dithering unstable. In 2D hand-drawn art, it's a skill, but it's not quite the same challenge as described in that post.
Occasionally it switches from product hunt to a "Your Ad Here" ad that links to a popup with the following (slop) text:
> Put your café, shop, studio or brand on the billboard in Paris: drawn by hand in the same pixel-art style, and glowing in the scene while people study, work and unwind to the music.
It's clear this is a money-making endeavor, not something created out of passion or goodness of heart. And not art.
Thanks! I’ll change it to “drawn in the same pixel-art style.”
This is a passion project for me. I’m paying to keep it running, and the billboards are meant to help cover those costs. I’ll add an option to hide them.
Ads are the worst and they're antithetical to the vibe of something like this
If it's an actual ad.. You'd get more goodwill (and maybe clicks) out of a tiny 'if you appreciate this maybe check out product hunt to support me' message somewhere. At least it's not manipulative that way.
There's a lot to like here, but too AI-generated for me. The pixel art for Tokyo & Hong Kong don't use proper Chinese/Japanese characters. The pulsating badges in the non-pixel UI are also pretty AI-indicative in a way that is distracting.
Yeah, the city search got cut off on mobile. Vibey and janky, but still it's a pretty creative idea.
I would have been very impressed if it made pixel art for cities on demand. If I were able to put in my hometown and see a delightful pixel art memory, I'd swoon and point to it and say "wow wee LLMs are pretty dope aren't they", but instead I'm just thinking this is yet another shallow project which will be abandoned tomorrow.
The point of my comment was that it’s unlikely there will be something else in world where AI generated output is free and the alternative is extremely expensive.
Gorgeous, very well executed. I'm guessing the primary request will be more cities but I'll deviate and say I'd love to see more views as well. Would be nice to have the same set of cities by I can see it from a studio view same way it is in the Paris view. Or from an office. Or a cafe.
Maybe to a degree but this isn’t just somebody using Seeddance to generate a first-frame-and-last-frame video loop and calling it a day.
There's definitely some manual curation. They even have assets that are clearly being placed into the scene, like the boats moving in the water.
You could theoretically generate all possible animated sprites (birds, people, vehicles, etc.) for use in any given cityscape, but automatically segmenting/detecting the different valid "paths/waypoints" where those assets should be placed would probably end up looking pretty janky.
Maybe??.. Opus 5.5 is a giant step function up with "manual illustration". Quite a bit better than Astra. It is entirely possible to get this quality of animation "hand drawn" from Opus 5.5...
Honestly I was blown away, it is one capability that blindsided me. I haven't been this surprised by a capability lift in a long time.
I haven't had a chance to play with Opus 5.5 yet, but I guess if there were a way to generate each of the pieces (aka the layers) then it might be possible:
- parallax, which might be some clouds moving in the background.
- weather effects which is mostly just an overlay
- animated sprites that interact with, or at least move along, the different sections of the static scene.
It'd be neat to see how well it could recreate some of the larger set pieces like Westminster palace.
I've definitely seen some pretty impressive automated AI voxel stuff (and that was 6 months ago) and that's having to deal an extra dimension. [1]
Personally I still don’t see a situation in which you could do it completely hands-off though. I think you’d end up with some pretty good scenes and some really terrible scenes requiring post-adjustments. It would be an interesting experiment at any rate!
I had a similar idea yesterday - https://www.lofiwizard.com/ - I mainly wanted to have Opus 5.5 generate random browser scenes to see what it could do. I love how your pixel cities came out! I tried to do similar music via code but it never sounded right so I had Google's Lyria 3 model make the music instead.
Yeah and there were other 3rd party "hacks" for it, but I mean Windows or macOS never made it an official API, so it never caught on, even though Apple's own "Dynamic Wallpapers" use some kind of motion (they're videos but only during screen-savers or time-based)
Just tested and this works great on macOS with Plash [1] by Sindre Sorhus. I couldn't find too many URL parameters to control it, but the FAQ does mention an ?obs parameter to hide the interface by default and ?weather=... to choose the weather. This URL in Plash works as described:
https://loficities.com/tokyo/?obs&weather=leaves
I was also able to omit ?obs and use the website interface controls to configure it via Plash browsing mode and then the interface went away after a moment of inactivity.
That's up to personal preference. Just like with static wallpapers, some people like single subtle colors, some people like noisy multicolored wallpapers.
There's already subtle parallax in iOS etc.
Animated wallpapers could be subtle, like raindrops, or animate only when interacted with, when you're bored or waiting for the AI to finish a run etc. :)
Stated preference perhaps, but unlikely revealed preference.
Depends on context, but for a computer on which you're doing work, moving wallpaper is usually a net negative for usability while you're actively working, though the magnitude depends heavily on the animation.
The problem is primarily peripheral attention. Motions is a strong pre-attentive visual signal: your visual system detects it without you deliberately deciding to look at it. One a laptop where you're close to the display and spending long periods concentrating on text, code, documents, etc., any visible animated region around or through windows can create low-level competition for attention.
A good compromise is wallpaper motion during transitions or stillness during work.
Broader UI principle: motion should explain change, not create change to be noticed.
When using my 32 inch screen sometimes I set the wallpaper to a random walk video in YouTube and it works fantastically. I know that it is counter intuitive, but the contrast between real and moving elements and static and almost flat Windows makes for a nice experience. And helps fighting text editing fatigue.
I was having a very similar thought - finally a nice use case for HTML wallpapers like Windows used to have them. I bet that there's a trick to have this as a desktop background in Wayland, but I'm too lazy to find out right now.
Either way, really nicely made, I love it! Small change request: adjust the CSS to make the controls look less vibecode-y.
I’ll take another pass at the controls so they fit the scenes better. If you find a good way to run it as a wallpaper on Wayland, I’d be interested to see it.
strange to me that "lofi" has graduated from low fidelity to AI generated downtempo music, which I assume (like the art) is based on the work of real artists.
All generated, I don't normally use the ChatGPT/Codex desktop app but in this instance I had it do everything directly.
So GPT 5.6 Sol using GPT images 2.0 (fake transparency by generating pink which it then removed) and it deployed to ChatGPT sites which I also wanted to just see in action for fun. It's basically just CloudFlare with D1 underneath or something.
So tldr yeah just generated with the built in image gen, it does have an advantage there over Claude. With images 2.5 I imagine the result would be more consistent.
I really love pixel-style websites. I plan to redesign these websites I built (www.viralreel.app; www.vfitly.com; www.dressmakergames.com) into pixel art style. What do you guys think? Could you give me some suggestions?
I think viralreel is the best fit for a lofi pixel style. Trending videos plus AI already have a strong tech vibe, so it would look really cool.
It's cute, but the custom weather overlays don't make sense. Where are the autumn leaves falling from on top of the 7th floor in Paris, the secret rooftop forest canopy?
I was really hoping that the music would change with the city. I want Istanbul to give me lofi tarlibasi roma wedding vibes, or at least arabesk taxi driver vibes
I first saw Paris from inside the room, and only later realized that 90% of the cities seem to be outdoors. It did make me kind of miss the indoor theming of the local area.
Also make me think how cool it would be to be able to toggle between being indoors or on the balcony. Shift the volume of the music vs the weather, maybe add some city sounds and shift them as well. Finally, it means you could have twice as much visual variety for only 50% more art.
I loved it. Since i was a child i have something with skylines in general that i can't explain (be it real or not). it's like a nostalgic/depressive feeling in a good and weird way. i also have a similar and weird thing with liminal spaces [0]
<overthinking>I get a feeling it's too cold inside. Maybe it's how thin the walls/windows look (evidenced by how loud the rain can be heard), or how tall the ceiling is, which makes it seem colder, with no light flicker indicating presence of an indoor fireplace.</overthinking>
Yeah, the perspective sort of suggests the foreshore east of Milsons Point, but it isn’t landscaped like that, and Opera House would appear different relative to the Harbour Bridge (not to mention Ms Macquaries Chair)
Quite honestly the procedural music generation algorithm sounds bad. I mean, this is just my opinion but the compositions sound incredibly unbalanced. Almost random. I think you would do better mixing up soundcloud released that have the right licenses and streaming that. Humans really do make better music than algorithms. As it stands: it's hard to call this music, much less lofi.
I’d like to keep the music generated in the browser, but it needs to feel less random. I’ll take another look at the balance and how the different parts fit together.
There's a lot of AI on the author's github, so I would assume so. I have no idea how I feel about these sorts of things. On the one hand I think it's objectively a super neat idea, but it makes me sad that a human (probably) didn't make the art/songs.
I use a compiler to perform a task. I don’t use pixel art to perform a task, other than to feel connection to another human —- something AI makes more challenging.
I've been thinking about this a lot lately. I feel very averse to consuming AI media, but it's not super obvious to me why. In some sense, I agree with what you're saying, but in reality no pixel art of a city I've never been to would elicit that feeling in me. Maybe I would feel differently if it were art of the city I live in and had some pixels dedicated to specific things I think are cool about the city, but for this website in particular it could be the best pixel art of SF ever made and I don't think it would be all that meaningful to me.
The main conclusion I've come to is that it's less that I dislike AI art inherently, and rather that I value the human decisions in media that I personally care about, and AI media threatens both its creation and how much other people value such things.
maybe you are MOSTLY judging what you perceive to be the artist's options and then decisions, and with most AI art, it's super easy to judge both of those quickly and accurately -- they had a billion options, and they chose a really lazy one.
or i don't know if you are but this is how i think i feel a lot of the time haha
I mean pixel art is kind of weird, in that I rarely just look at a single screen of pixel art and say, wow, I feel more connected now. It’s often taken in context of a larger whole, like a game or animation where the artist is usually trying to express something, and the specific art is some small part of the overall expression. It’s kind of like how no individual still image from Spirited Away will convey the meaning of the whole movie.
It's not how most people encounter it, but non-animated, non-game standalone pixel art is a thriving artistic niche. Check out the Hall of Fame on lospec for examples:
It is. That can be fine, but when it isn't clearly disclosed it's not OK IMO.
Pixel art cityscapes is also such a small niche, it wouldn't surprise me if there were some "reference images" from popular pixel artists in the pipeline.
Ah that's a bummer. I'm unsure if this is gonna change over time, but it entirely tanks my enjoyment of any visual art project to know it was generated.
Visited and when I clicked on Paris to select a different city, the music kicked in. I thought this was supposed to be lofi but I was met with what I would describe as basic midi. Immediately hit with loud hard piano and rhodes keyboard. The pixel art has an identity crisis. It's very much generated by AI. This entire thing is slop.
It's interesting, I vibecoded a very similar lofi/mood idle game, but in 2.5D with Codex, I would say that the reception was about 60% positive with a lot of people complaining about AI slop, then I removed the Codex fingerprints like the verbose titles "Neon Rain by Tokyo Tower", the button/field UI that looks like a ChatGPT wrapper startup from a year ago, changed font from the generated one, and then it had about 80% positive interaction.
I think its easier to sell it as actual art/non-slop when you take some time to remove what is generic across newer LLM sessions.
I'd encourage chronicling the effort you put in and be transparent about the methods. If you want your AI art to reach proper recognition, the different type of effort needs to be shown
It's a fair point. I experimented with this, but I still don't think that the general public is there yet. Any mention of AI is already a negative sentiment for a large group of consumers, especially in artistic fields. While this is true, it does not make sense to disclose these things because then you can capture both the customers that care and those who don't if you are sufficiently careful.
Once the technology gets good enough and people can't tell it apart and start to get used to generated content like this, it will perhaps start to make sense to provide this kind of effort if it's still relevant.
Disclosure makes sense when the trust it buys is worth more than the sales it costs. Currently the trade-off is completely dogshit. You don't get good reputation by making any art with AI, nor good sales by disclosure.
When/if people start rewarding disclosure it will make sense to do this, otherwise you are asking me to pay for this abstract principle out of my own pocket.
It's not that weird - Google isn't a search engine company, they're an advertising company.
I'm not just flogging a dead horse here - they don't care how good the search results for these types of queries are, they need to be just good enough to stop you from moving to search via Perplexity, Chatgpt, Claude etc. They also need to be as cheap as possible to generate.
Google makes money when you search for things to buy which is why they'll give you an ad-free experience on things that they never made much money off anyways (eg "why is the sky blue?") - they want you to be around when you search for something they can (eg "jewellery"). Note that they never give you AI slop when you're doing a product search.
Unfortunately people both expect answers phrased in response to their questions, and people have generally poor skills at picking up on authoritative-sounding bullshit. Also if they give you "an answer" and stop you from moving to another site with more detail (and possibly ads from someone else!) you're more likely to ask follow up questions on google.
It’s also not realistic, in one landscape the billboard is taller than a 9 story building.
Human pixel art tends to be very deliberate about the exact colors of every pixel and their placement tends to follow idealized (but not necessarily geometric) shapes.
The last time I saw dithering this sloppy was decades ago when people would scan photos as low color images. It is its own vibe, but it's a cold and rough one. These artifacts feel more "realistic", but it felt like slop back then too.
They are. And they are charging $19 to buy AI generated art (https://safaelmali.gumroad.com/l/lofi-cities-complete-collec...).
Turns out part of the appeal for me is that it appears to be handmade.
[0]: https://dukope.com/devlogs/obra-dinn/tig-32/
> Put your café, shop, studio or brand on the billboard in Paris: drawn by hand in the same pixel-art style, and glowing in the scene while people study, work and unwind to the music.
It's clear this is a money-making endeavor, not something created out of passion or goodness of heart. And not art.
This is a passion project for me. I’m paying to keep it running, and the billboards are meant to help cover those costs. I’ll add an option to hide them.
Ads are the worst and they're antithetical to the vibe of something like this
If it's an actual ad.. You'd get more goodwill (and maybe clicks) out of a tiny 'if you appreciate this maybe check out product hunt to support me' message somewhere. At least it's not manipulative that way.
Crazy times, man.
Only on HN there can be a single pixel that is misplaced or off and it causes ONE person to get upset about it.
This is completely petty. Just stop.
I would have been very impressed if it made pixel art for cities on demand. If I were able to put in my hometown and see a delightful pixel art memory, I'd swoon and point to it and say "wow wee LLMs are pretty dope aren't they", but instead I'm just thinking this is yet another shallow project which will be abandoned tomorrow.
https://www.effectgames.com/demos/worlds/
Actually, shouldn't it be possible to auto-generate any city via AI on demand?
There's definitely some manual curation. They even have assets that are clearly being placed into the scene, like the boats moving in the water.
You could theoretically generate all possible animated sprites (birds, people, vehicles, etc.) for use in any given cityscape, but automatically segmenting/detecting the different valid "paths/waypoints" where those assets should be placed would probably end up looking pretty janky.
Honestly I was blown away, it is one capability that blindsided me. I haven't been this surprised by a capability lift in a long time.
- parallax, which might be some clouds moving in the background.
- weather effects which is mostly just an overlay
- animated sprites that interact with, or at least move along, the different sections of the static scene.
It'd be neat to see how well it could recreate some of the larger set pieces like Westminster palace.
I've definitely seen some pretty impressive automated AI voxel stuff (and that was 6 months ago) and that's having to deal an extra dimension. [1]
Personally I still don’t see a situation in which you could do it completely hands-off though. I think you’d end up with some pretty good scenes and some really terrible scenes requiring post-adjustments. It would be an interesting experiment at any rate!
[1] - https://github.com/Ammaar-Alam/minebench
Why do companies like Apple & Microsoft sterilize themselves so?
[0] https://en.wikipedia.org/wiki/Active_Desktop
[1]: https://sindresorhus.com/plash
There's already subtle parallax in iOS etc.
Animated wallpapers could be subtle, like raindrops, or animate only when interacted with, when you're bored or waiting for the AI to finish a run etc. :)
Stated preference perhaps, but unlikely revealed preference.
Depends on context, but for a computer on which you're doing work, moving wallpaper is usually a net negative for usability while you're actively working, though the magnitude depends heavily on the animation.
The problem is primarily peripheral attention. Motions is a strong pre-attentive visual signal: your visual system detects it without you deliberately deciding to look at it. One a laptop where you're close to the display and spending long periods concentrating on text, code, documents, etc., any visible animated region around or through windows can create low-level competition for attention.
A good compromise is wallpaper motion during transitions or stillness during work.
Broader UI principle: motion should explain change, not create change to be noticed.
No, you don't know me, or other people.
I want every fucking thing moving at all fucking times. It's okay.
Btw for other people who love some motion in their occupation:
https://store.steampowered.com/app/1516350/Textreme_2/
https://ash-k.itch.io/textreme-2
Of course, ymmv. But give it a try.
Either way, really nicely made, I love it! Small change request: adjust the CSS to make the controls look less vibecode-y.
https://cindori.com/backdrop
https://yamanote-window-loop.aljosa-asanovic.chatgpt.site/v3
Sound on for nostalgia if you've been to Tokyo.
Edit: forgot I had added a little behind the scenes page
https://yamanote-window-loop.aljosa-asanovic.chatgpt.site/ma...
So GPT 5.6 Sol using GPT images 2.0 (fake transparency by generating pink which it then removed) and it deployed to ChatGPT sites which I also wanted to just see in action for fun. It's basically just CloudFlare with D1 underneath or something.
So tldr yeah just generated with the built in image gen, it does have an advantage there over Claude. With images 2.5 I imagine the result would be more consistent.
I first saw Paris from inside the room, and only later realized that 90% of the cities seem to be outdoors. It did make me kind of miss the indoor theming of the local area.
Also make me think how cool it would be to be able to toggle between being indoors or on the balcony. Shift the volume of the music vs the weather, maybe add some city sounds and shift them as well. Finally, it means you could have twice as much visual variety for only 50% more art.
[0] https://en.wikipedia.org/wiki/Liminal_space
For any city, a thunderstorm or light drizzle option would be very moody.
The main conclusion I've come to is that it's less that I dislike AI art inherently, and rather that I value the human decisions in media that I personally care about, and AI media threatens both its creation and how much other people value such things.
or i don't know if you are but this is how i think i feel a lot of the time haha
https://lospec.com/gallery/sorting:likes/time:all
With AI, instead you feel a connection to the sum of all humanity.
Pixel art cityscapes is also such a small niche, it wouldn't surprise me if there were some "reference images" from popular pixel artists in the pipeline.
https://safaelmali.gumroad.com/l/lofi-cities-complete-collec...
I think its easier to sell it as actual art/non-slop when you take some time to remove what is generic across newer LLM sessions.
Once the technology gets good enough and people can't tell it apart and start to get used to generated content like this, it will perhaps start to make sense to provide this kind of effort if it's still relevant.
When/if people start rewarding disclosure it will make sense to do this, otherwise you are asking me to pay for this abstract principle out of my own pocket.
I'm not just flogging a dead horse here - they don't care how good the search results for these types of queries are, they need to be just good enough to stop you from moving to search via Perplexity, Chatgpt, Claude etc. They also need to be as cheap as possible to generate.
Google makes money when you search for things to buy which is why they'll give you an ad-free experience on things that they never made much money off anyways (eg "why is the sky blue?") - they want you to be around when you search for something they can (eg "jewellery"). Note that they never give you AI slop when you're doing a product search.
Unfortunately people both expect answers phrased in response to their questions, and people have generally poor skills at picking up on authoritative-sounding bullshit. Also if they give you "an answer" and stop you from moving to another site with more detail (and possibly ads from someone else!) you're more likely to ask follow up questions on google.
This is almost like a reflex, I am not sure why, because when I look, I don't see any pattern...
Anyone else have the same?