Staying focused: the panda is wearing gloves, but then when it does the Hadouken and the Hurrican Kicks, it loses the gloves.
Bigger picture this seems like a focus on product stuff like pricing and demos and navbars that don't really matter to artists or game developers. Your feelings are correct that people here - the root motivation of the negativity - are questioning your sincerity. With AI art this is acutely true, people don't view AI art as a sincere art endeavor. I don't doubt your sincerity. But spending more time on the art thing, and making it free or open source, it's going to look and feel more authentic.
I've been toying around with the exact same idea the last couple weeks. It's mostly GPT 4o image => some image cleanup, but honestly a lot more finicky than I originally expected. Lots of prompt engineering. So OP probably put in a fair bit of effort here.
Also each animation probably costs $1-$2 in GPT costs to make [1], so not something that's easy to throw a free tier on.
Not sure about professional quality. All the results seem to need quite bit of hand editing to look good. Also the examples only show animations ripped from street fighter.
For purposes of generative worlds, I have an additional requirement for interchangable equipment and weapons. I.e. a single sword or armor sprite should fit with all humanoid characters, in all animations. I suspect this could be achieved by training against "clipping" tests.
As a start, there should be projections for all four cardinal directions.
Limit the output to palettes of a given size, for classic recolors via GL shaders.
As someone who is an artist and also a programmer I find the differences between both very different. I never see programmers crying over A.I. chiming “they stole our code so it’s evil and I’ll never use it and neither should you!” Like the artist seem to do.
I don’t really care if the image gen models are trained on my renders or photos as long as they are open source.
Also I have been making the assets for my game which needs at least 2,000 sprites and honestly it’s very tedious and I’m looking to automate as much of the pipeline as I can so I welcome anything that removes the the tedium and pain from the process.
people did complain about it stealing GPL code scraped off github etc and then license laundering a lot, especially during the earlier days. I think now people are just worn down to have the conversation again.
I do also think the threat is bigger to artists. Sloppy code will cause problems, and at some point you'll need someone who knows what's going on to step in. Sloppy art, a lot of people accept.
That's why artist should incorporate gen ai into their workflows. They have the artistic eye and separate the slop from the gold better than non artistic people. Also knowing composition and etc helps getting better results from Stable Diffusion, Flux and etc.
I remember when 3D CG was new and it got so much hate for not being "real art". I heard the same drama happened when Photoshop and Illustrator hit the market, they weren't considered "real art".
From a recent blog shared on HN, among a list of reasons they can’t or won’t use LLMs:
>The training data for LLMs is stolen. I don’t mean like “pirated” in the sense where someone illicitly shares a copy they obtained legitimately; I mean their scrapers are ignoring both norms and laws to obtain copies under false pretenses, destroying other people’s infrastructure. [footnotes omitted]
I do agree that the scraping is annoying and I had to set up some anti scraping measures on one of my image heavy sites. However I am for the freedom of data especially if the models are open source. I only use local models and haven't used Claude or ChatGPT since last year and it's pretty awesome.
I think there's a big difference with how designers and programmers can use AI to enhance their work, and how exploited the average person's work has been. A lot of designers publish their work on social media, or it's visible on their websites etc., which AI models have used to train with. But with coding, most people's code only exists in private repos, or is compiled to a format that LLMs cant easily be trained on.
There is so much code on github and all the other repos that it can compare with the amount of art that is out there. Also all of your Javascript code is public once your site is published.
There is nothing stopping artists from using A.I. to improve their work. The sketch to image work flows are great, the variation work flows can save from the tedium and inpainting can help fix and improve images.
Text to image is lazy and is the cause of most of the slop but no one is saying artist should replace their Wacom tablets with text prompts. I feel like there is a lot of hurt egos going on. I remember back in the day on CGtalk, there was so much elitism and in hindsight it probably held a lot of people back, myself included.
Even as someone who leans generally anti-corp I would love to see Capcom come in and sue over this since there's no way to look at those demo animations and not see the "training" was entirely lifted from the Street Fighter series.
Unless you didn’t want people to see and use it, maybe you shouldn’t have made it public? I don’t think training AI is fundamentally different from humans learning from things they see, and we don’t restrict that either.
The fundamental difference is that computers can do this at a pace and scale that humans could never aspire to. There is a natural limit to the extent a single individual can be informed by previous works. It sets a natural pace to innovation that is sustainable for both the artist and derivative works.
Computers have no such limitation and can consume almost the entirety of a subject’s work in a few weeks or months. I think that alone is enough to say that yes, it is fundamentally different.
Automated assembly lines have the same properties. Same with transportation. Buses, trains, airplanes, ships. These all work tirelessly at a pace that humans cannot match.
I think you missed the point. The claim is that AI training from public information is no different from humans learning from public information.
My argument is precisely that the mechanisation of information is fundamentally different from the scale at which a human can learn. One immediate consequence being there is no longer a natural brake on the scale of what can be sourced for use in a derivative work.
To be clear this is not a value judgement, just to point out that it _is_ different, just as driving is fundamentally different from what one can do with one’s own feet. Of course the mechanisation of transport is history and seems daft to argue against. But it is different. Whether that’s good or bad is a much harder question.
When it’s big copyright holders we have very specific, very granular definitions of what constitutes fair or allowed use. But when it comes to smaller creators the answer is that it’s their fault for trying to promote their work and make a living.
These LLMs primarily distribute intellectual and creative wealth from media conglomerates to anyone on earth with $20. (Without fair compensation, agreed)
Imagine if writers complained that stuff they wrote were being used in literature classes that teaches people how to read and write because how sentences they once toiled over for hours would now be used in mundane letters.
AI can't take away an artists creativity, just like how photography didn't kill painting. Yes, it closes up certain avenues on how artists make money but it will most probably make up for that by making them more productive.
I’m curious if you know of a similar tool for generating tilesets for top down RPG like games? I’ve tried Midjourney, which makes nice art, but it can’t be used in a game.
I’d love to hear a bit about the ML side of things: what was your experience with various models? Do you see a clear cost vs quality tradeoff with current state of the art models? How do open vs closed models compare?
What model(s) does this use? Is it built using open source models and tools? I couldn't find any mention of models in the FAQ (who even asked those questions?) or elsewhere. I doubt you trained your own diffusion model.
The GitHub cat in the footer does not link to GitHub either.
Hand-drawn animation, whether for video games or television or other media, is one of those artistic fields where I can see AI being a huge win for everybody. The reason why is because drawing the in-betweens in an animation is tedious, time-consuming, costly, and, in most 2D animation productions, outsourced to sweatshops abroad. And yet it's most of the work that goes into an animation. I think most animators would rather just do the keyframes and leave the tweens in the hands of the machine. It'd make traditional animation as easy and fast as Flash animation, and create a flourishment of even more indie animations from solo creators with even higher quality.
It kind of does what it says on the can, but you will have to edit the result frame by frame in order to make the lost scales and equipment consistent across all animations. Still, a huge time boost.
All AI tools I've tried have failed to generate animations for slimes in my experiments. (Including Veo3) I'm hesitant to even try this, any chance it would work?
You're going to have a much harder time getting good animations generated for something that isn't a humanoid, since the training sets won't have much to work from. It's possible with a really good prompt for 'slime' you could get something but this doesn't seem to allow prompt customization.
If you're after monsters like slimes I would not drop money on this.
I uploaded a photo of a pixel art electrician and it just completely ruins it and looks nothing like my guy and forces it into a particular style where it looks like a cheap mobile game playable ad and somehow Mario at the same time. Not to mention, the actual animation is wrong too.
It's cool but you can tell the sprites are low quality and AI generated. They have that trademark AI "fuzziness." It's possible that players would think it is a design choice but it is noticable.
> This isn't putting anyone out of work. The games simply would not be made in the first place.
The opposite is also true. Games that should never be made are being made due to the rubbish that can be generated by these tools. Observe the generated samples on the landing page, they are literally just copying street fighter. These tools are so useless without ripping off the hard work of humans. The deluge of slop is a signal to noise problem.
The samples on the landing page are... not good, by any metric. The background jitters, details go missing from frame to frame, and the animations that should cycle (run, walk...) don't look like they do.
This is a prototype, at best. I would be ashamed of asking money for providing such a sloppy service.
Also, in your faq you have "You own the rights to your generated content.", which I don't think is true. AFAIK, you can't copyright AI art.
I hate this. It's cool that it's possible, but I hate it. Generating art solves no real problem and only suppresses a struggling profession without which this wouldn't even be possible...
The problem is that in the indie game market, where most 2d sprites are drawn, there is not enough money going around making many studios profitable. By reducing cost of manual 2d animation, more studios can be profitable and able to deliver better, more finished, games. Then studios can pay decent salaries to employees.
Have to agree. I don’t think we should stop ai progress, just that I think we built a bad thing along the way. I will never buy a video game that I know has ai generated content, I want to see some art, not a soulless prediction of what art might look like.
It's not really practical but I would take the bet that I will never /intentionally/ buy a game with AI generated assets. I just don't see that AI generated art has any value to it. I'm sure you could sneak some past me, it's getting harder to spot all the time.
miaomiaoPixel_v11 can generate sprites that small. Specify a size of 128x128, and that corresponds to an actual size of 16x16. Use an image editor to shrink the sprite down to 16x16, then reduce the palette as needed to remove the near-matching colors.
Curious idea, but doesn't seem to working for me. It's been sitting in "Your job is in queue" state for 10 minutes, and that's it. Reloaded the page just now and it just reset itself.
I'm surprised posts like this aren't instantaneously flagged and removed. The tools for doing this are readily available for free as it is.
This is basically another "throw AI in it" CRUD app that wants your money. They're all really low effort grifts that do not deserve to see the light of day.
They work for a subset of animations that go in commonplace game types. Anything creative or new or different? No chance you'll get a satisfactory result. They are all subtly wrong in ways that anyone can easily spot and will likely get your game filtered out and harshly criticized publically on Steam or other platforms.
There is a lot of negativity on Hacker News.
Staying focused: the panda is wearing gloves, but then when it does the Hadouken and the Hurrican Kicks, it loses the gloves.
Bigger picture this seems like a focus on product stuff like pricing and demos and navbars that don't really matter to artists or game developers. Your feelings are correct that people here - the root motivation of the negativity - are questioning your sincerity. With AI art this is acutely true, people don't view AI art as a sincere art endeavor. I don't doubt your sincerity. But spending more time on the art thing, and making it free or open source, it's going to look and feel more authentic.
I think the negativity is overkill.
I've been toying around with the exact same idea the last couple weeks. It's mostly GPT 4o image => some image cleanup, but honestly a lot more finicky than I originally expected. Lots of prompt engineering. So OP probably put in a fair bit of effort here.
Also each animation probably costs $1-$2 in GPT costs to make [1], so not something that's easy to throw a free tier on.
[1] https://openai.com/api/pricing/#:~:text=Image%20Generation%2...
Not sure about professional quality. All the results seem to need quite bit of hand editing to look good. Also the examples only show animations ripped from street fighter.
Great progress. Bookmarked.
For purposes of generative worlds, I have an additional requirement for interchangable equipment and weapons. I.e. a single sword or armor sprite should fit with all humanoid characters, in all animations. I suspect this could be achieved by training against "clipping" tests.
As a start, there should be projections for all four cardinal directions.
Limit the output to palettes of a given size, for classic recolors via GL shaders.
Thanks for so many valuable points and ideas. Will do.
I like the "interchangable equipment and weapons." idea, very cool. And yes. it's doable. many ways possible could work. I'll experiment.
Hmm The payment link does not require login And I paid and the credits vanished into thin air ...
To the poster, please find the customer service to fix
https://buy.stripe.com/cNicN59Bqf6K4q92Qrbwk00?prefilled_ema...
Also, I cannot find my queuing job after closing the page.
sorry for the inconvenience.
I've fixed it.
ALso given your account 2x more credits.
Shouldn’t it be more like 5x credits? That’s what I would have given to a paying early adopter dealing with these annoying issues.
checking
It’s sad how the existing corpus of artists’ works are being used against them.
The only reason this can generate images like these is because artists previously created artwork like this. And then in return they get:
> you don't need to high an artist or animator to develop you game.
Unless you created all the training material yourself, any model like this is highly unethical, in my opinion.
As someone who is an artist and also a programmer I find the differences between both very different. I never see programmers crying over A.I. chiming “they stole our code so it’s evil and I’ll never use it and neither should you!” Like the artist seem to do.
I don’t really care if the image gen models are trained on my renders or photos as long as they are open source.
Also I have been making the assets for my game which needs at least 2,000 sprites and honestly it’s very tedious and I’m looking to automate as much of the pipeline as I can so I welcome anything that removes the the tedium and pain from the process.
people did complain about it stealing GPL code scraped off github etc and then license laundering a lot, especially during the earlier days. I think now people are just worn down to have the conversation again.
I do also think the threat is bigger to artists. Sloppy code will cause problems, and at some point you'll need someone who knows what's going on to step in. Sloppy art, a lot of people accept.
I’m not sure over time people will continue to accept sloppy art.
The “glitchiness” of the AI sprites can be quite unsettling because one can feel the something-quite-wrong that is difficult to place.
This is a worse state than using even rough illustrations by actual artists.
If one must use AI for this I would recommend having an artist in the pipeline who uses AI to create assets but then makes sure they are seamless.
That's why artist should incorporate gen ai into their workflows. They have the artistic eye and separate the slop from the gold better than non artistic people. Also knowing composition and etc helps getting better results from Stable Diffusion, Flux and etc.
I remember when 3D CG was new and it got so much hate for not being "real art". I heard the same drama happened when Photoshop and Illustrator hit the market, they weren't considered "real art".
From a recent blog shared on HN, among a list of reasons they can’t or won’t use LLMs:
>The training data for LLMs is stolen. I don’t mean like “pirated” in the sense where someone illicitly shares a copy they obtained legitimately; I mean their scrapers are ignoring both norms and laws to obtain copies under false pretenses, destroying other people’s infrastructure. [footnotes omitted]
“I think I'm done thinking about GenAI for now” https://news.ycombinator.com/item?id=44193018
I do agree that the scraping is annoying and I had to set up some anti scraping measures on one of my image heavy sites. However I am for the freedom of data especially if the models are open source. I only use local models and haven't used Claude or ChatGPT since last year and it's pretty awesome.
I think there's a big difference with how designers and programmers can use AI to enhance their work, and how exploited the average person's work has been. A lot of designers publish their work on social media, or it's visible on their websites etc., which AI models have used to train with. But with coding, most people's code only exists in private repos, or is compiled to a format that LLMs cant easily be trained on.
There is so much code on github and all the other repos that it can compare with the amount of art that is out there. Also all of your Javascript code is public once your site is published.
There is nothing stopping artists from using A.I. to improve their work. The sketch to image work flows are great, the variation work flows can save from the tedium and inpainting can help fix and improve images.
Text to image is lazy and is the cause of most of the slop but no one is saying artist should replace their Wacom tablets with text prompts. I feel like there is a lot of hurt egos going on. I remember back in the day on CGtalk, there was so much elitism and in hindsight it probably held a lot of people back, myself included.
I’m not opposed to generative AI, in principle, it’s how it’s being employed that bothers me.
Please elaborate.
Even as someone who leans generally anti-corp I would love to see Capcom come in and sue over this since there's no way to look at those demo animations and not see the "training" was entirely lifted from the Street Fighter series.
Unless you didn’t want people to see and use it, maybe you shouldn’t have made it public? I don’t think training AI is fundamentally different from humans learning from things they see, and we don’t restrict that either.
The fundamental difference is that computers can do this at a pace and scale that humans could never aspire to. There is a natural limit to the extent a single individual can be informed by previous works. It sets a natural pace to innovation that is sustainable for both the artist and derivative works.
Computers have no such limitation and can consume almost the entirety of a subject’s work in a few weeks or months. I think that alone is enough to say that yes, it is fundamentally different.
Automated assembly lines have the same properties. Same with transportation. Buses, trains, airplanes, ships. These all work tirelessly at a pace that humans cannot match.
There is no fundamental difference.
I think you missed the point. The claim is that AI training from public information is no different from humans learning from public information.
My argument is precisely that the mechanisation of information is fundamentally different from the scale at which a human can learn. One immediate consequence being there is no longer a natural brake on the scale of what can be sourced for use in a derivative work.
To be clear this is not a value judgement, just to point out that it _is_ different, just as driving is fundamentally different from what one can do with one’s own feet. Of course the mechanisation of transport is history and seems daft to argue against. But it is different. Whether that’s good or bad is a much harder question.
When it’s big copyright holders we have very specific, very granular definitions of what constitutes fair or allowed use. But when it comes to smaller creators the answer is that it’s their fault for trying to promote their work and make a living.
AI isn’t human.
1. These models have been trained on private data many a time without permission
2. Making something public isn’t always a choice made by the creator
3. Making something public does not denote fair use, this is why copyright (albeit arguably a poor solution) exists
4. These LLMs consolidate wealth into a small group using outputs from a larger, often less wealthy group (creatives) without fair compensation
These LLMs primarily distribute intellectual and creative wealth from media conglomerates to anyone on earth with $20. (Without fair compensation, agreed)
Imagine if writers complained that stuff they wrote were being used in literature classes that teaches people how to read and write because how sentences they once toiled over for hours would now be used in mundane letters.
AI can't take away an artists creativity, just like how photography didn't kill painting. Yes, it closes up certain avenues on how artists make money but it will most probably make up for that by making them more productive.
Comparing teaching other humans to "teaching" a model is absurd.
AI can't take their creativity but it will take their means of survival so a few companies can profit instead. Yeah, great trade-off.
> Yes, it closes up certain avenues on how artists make money but it will most probably make up for that by making them more productive.
How?
I’m curious if you know of a similar tool for generating tilesets for top down RPG like games? I’ve tried Midjourney, which makes nice art, but it can’t be used in a game.
Is this using OpenAI's GPT-image or something different?
Congrats on shipping!
I’d love to hear a bit about the ML side of things: what was your experience with various models? Do you see a clear cost vs quality tradeoff with current state of the art models? How do open vs closed models compare?
What model(s) does this use? Is it built using open source models and tools? I couldn't find any mention of models in the FAQ (who even asked those questions?) or elsewhere. I doubt you trained your own diffusion model.
The GitHub cat in the footer does not link to GitHub either.
This is so depressing.
Also, why does the female ninja suddenly grow a penis?
Hand-drawn animation, whether for video games or television or other media, is one of those artistic fields where I can see AI being a huge win for everybody. The reason why is because drawing the in-betweens in an animation is tedious, time-consuming, costly, and, in most 2D animation productions, outsourced to sweatshops abroad. And yet it's most of the work that goes into an animation. I think most animators would rather just do the keyframes and leave the tweens in the hands of the machine. It'd make traditional animation as easy and fast as Flash animation, and create a flourishment of even more indie animations from solo creators with even higher quality.
It kind of does what it says on the can, but you will have to edit the result frame by frame in order to make the lost scales and equipment consistent across all animations. Still, a huge time boost.
All AI tools I've tried have failed to generate animations for slimes in my experiments. (Including Veo3) I'm hesitant to even try this, any chance it would work?
I really want to see an example with a slime. I’m sure it’d butcher it, but it would be so funny.
You're going to have a much harder time getting good animations generated for something that isn't a humanoid, since the training sets won't have much to work from. It's possible with a really good prompt for 'slime' you could get something but this doesn't seem to allow prompt customization.
If you're after monsters like slimes I would not drop money on this.
Lol it's not functional for most things.
I uploaded a photo of a pixel art electrician and it just completely ruins it and looks nothing like my guy and forces it into a particular style where it looks like a cheap mobile game playable ad and somehow Mario at the same time. Not to mention, the actual animation is wrong too.
It's cool but you can tell the sprites are low quality and AI generated. They have that trademark AI "fuzziness." It's possible that players would think it is a design choice but it is noticable.
will add a quality upscale model to fix it. thanks.
The example sprites aren't loading for me. I see original panda and the female ninja graphics, but the 6 sprite boxes are empty.
thanks for letting me know. I'll put some CDN to the videos.
This is awesome, good luck with this tool. Can it do isometric and top down views?
working on adding those
I would like to know what sort of AI is trained to do this.
You wouldn’t happen to have vibe coded this would you? I can’t get a response from multiple pages.
> This isn't putting anyone out of work. The games simply would not be made in the first place.
The opposite is also true. Games that should never be made are being made due to the rubbish that can be generated by these tools. Observe the generated samples on the landing page, they are literally just copying street fighter. These tools are so useless without ripping off the hard work of humans. The deluge of slop is a signal to noise problem.
The samples on the landing page are... not good, by any metric. The background jitters, details go missing from frame to frame, and the animations that should cycle (run, walk...) don't look like they do.
This is a prototype, at best. I would be ashamed of asking money for providing such a sloppy service.
Also, in your faq you have "You own the rights to your generated content.", which I don't think is true. AFAIK, you can't copyright AI art.
I hate this. It's cool that it's possible, but I hate it. Generating art solves no real problem and only suppresses a struggling profession without which this wouldn't even be possible...
I guess we should also delete Stack Overflow while we're at it because programmers should write everything from scratch for the same reason
> Generating art solves no real problem
It definitely does solve a problem. This tool provides affordable custom sprites with fast delivery.
If it didn't solve any problem there would be zero demand for it and no reason for you to care about it.
The problem is that in the indie game market, where most 2d sprites are drawn, there is not enough money going around making many studios profitable. By reducing cost of manual 2d animation, more studios can be profitable and able to deliver better, more finished, games. Then studios can pay decent salaries to employees.
Have to agree. I don’t think we should stop ai progress, just that I think we built a bad thing along the way. I will never buy a video game that I know has ai generated content, I want to see some art, not a soulless prediction of what art might look like.
> I will never
Bet you $10 that in five years' time you will have.
It's not really practical but I would take the bet that I will never /intentionally/ buy a game with AI generated assets. I just don't see that AI generated art has any value to it. I'm sure you could sneak some past me, it's getting harder to spot all the time.
My guess is that in five years’ time there won’t be any games that don’t include any, and also people will really have stopped caring so much
I think the problem is that the profession is struggling because you pay a lot of money for work that is often of very dubious quality.
I can’t ask a anyone that claims to be a pixel artist for sprites because the chance of getting garbage is just too high.
Btw, if anyone does know someone good, I’m open to paying xD
> solves no real problem
this is very obviously not true.
really the space for 16x16 sprites with AI has not been done very well so far
miaomiaoPixel_v11 can generate sprites that small. Specify a size of 128x128, and that corresponds to an actual size of 16x16. Use an image editor to shrink the sprite down to 16x16, then reduce the palette as needed to remove the near-matching colors.
16x24 sprites end up looking a lot nicer though.
I operated for 30 minutes, but it didn't achieve the result I wanted
Settings page does not work. Profile page does not work. I've been waiting for my first test for 30 minutes now. Not a great first impression...
fixed setting and profile.
sorry for the long waiting, I didn't thought there would be so many requests. Added more powerful GPUs.
Curious idea, but doesn't seem to working for me. It's been sitting in "Your job is in queue" state for 10 minutes, and that's it. Reloaded the page just now and it just reset itself.
sorry, I didn't expect so many requests all queued up too long. I've added some servers it should start to process.
When your request finish it'll send you an email notification.
Three things are known. Slashdot and HN effect and the gervais effect.
Don't really get the mean-spirited comments. Not everyone has the means to exascale their project.
Hope someone makes a celebrity Mortal Kombat game from this: https://www.youtube.com/watch?v=DjYfFDvtZ5s
The privacy pages and legal pages don't link. They all return to the homepage.
Without them I am hesitant in uploading my concept art from my project for testing.
Otherwise it is a cool concept and I would potentially look in to if it turns out you don't go all evil villain and claim ownership over the sprites.
Enabling a preview set of sprites without having an account would do no harm neither.
But creating an account without any legal documentation is a no from me as I don't know whats happening with my IP which leaves me uncomfortable.
Even if it's not a ratified legal document; some draft is better than nothing.
sorry, just fixed it.
I'm surprised posts like this aren't instantaneously flagged and removed. The tools for doing this are readily available for free as it is.
This is basically another "throw AI in it" CRUD app that wants your money. They're all really low effort grifts that do not deserve to see the light of day.
They work for a subset of animations that go in commonplace game types. Anything creative or new or different? No chance you'll get a satisfactory result. They are all subtly wrong in ways that anyone can easily spot and will likely get your game filtered out and harshly criticized publically on Steam or other platforms.
This right here too is so egregious: https://www.godmodeai.cloud/plans