vunderba

This. The idea of having some kind of cryptographic proof of ownership to a JPG copy of an original piece of artwork was just about the silliest thing I’d ever heard when it first started going viral.

I (and most others) support artists by donating through platforms like Patreon, without needing some kind of arbitrary artificial commodity.

I called it out a while ago but the largest market to get destroyed by gen AI was never the professional market - it was the gig economy.

The quality of generative AI may not be top-notch, but it presents an easy drop-in replacement for one-off tasks that people previously outsourced to platforms like Fiverr (voiceovers, logo design, clip art, copy editing, translation, etc).

There’s a related benchmark along these lines that uses an LLM with a discrete “instructive graphic language” to get models to generate Minecraft voxel structures with varying degrees of success.

https://github.com/Ammaar-Alam/minebench

I’d probably add “Save Early, Save Often, Save Multiple.” because there are so many ways to accidentally softlock yourself making the game totally unwinnable without realizing it until hours later.

There’s a reason most of us as kids started out with descriptive save names like `INNKEEP.SAV` and `BUYSLED.SAV`, and then halfway through the playthrough it's just `1.SAV`, `2.SAV`, `3.SAV`, etc. :)

Flux.2 [dev] was a pretty decent model when it released back in Nov 2025 but was an absolute GPU memory hog to run on local hardware (even with the Turbo LoRA).

That and it's frustrating licensing were the driving reasons why it got heavily overshadowed by Z-Image Turbo which came out around the same time.

Makes me nervous that even if Flux.3 goes open weight it'll be all but impossible to run on consumer hardware.

I have a couple of enthusiast friends who use CompactFlash-to-IDE adapters, and that’s one of the upgrades that almost makes it worth it. You still get the full retro experience (well except for the actuator arm seeking sounds), but on the plus side you don’t have to worry about failing magnetic hard drives.

Also this looks pretty neat as well

https://picoide.com

Totally agreed. Related, I actually wrote a satirical blog post on the brilliant but equally frustrating experiences of playing King's Quest IV: The Perils of Rosella when I was a kid. Just call me Captain Ahab because I hated that damn whale...

https://mordenstar.com/blog/kqiv

It's a proprietary model as a service - it doesn't require anything outside of a browser. But even when they released open-weight (the original Qwen-Image) it's a diffusion model and can't use custom font files.

Well from the images I've seen they seem to be pushing text-heavy "infograph" capabilities pretty hard. Outside of that, I have serious doubts that it'll be better than other proprietary models like NB Pro and gpt-image-2.

After Z-Image and the original Qwen-Image - I think they've pivoted completely to closed source. From the images I've seen, Qwen-Image 3.0 is just a subpar equivalent to other proprietary models like gpt-image-2 and nb-pro.

Yup. There have been weekly “headshot AI” generator Show HNs for practically a year at this point.

A non‑insignificant portion of them claim to be suitable for passport photos. Touching up a passport photo in any way (Photoshop or AI) is a brilliant way to get yourself denied a U.S. passport.

From the travel.state.gov [1] rules:

- Submit the original, unedited photo without filters or digital changes.

[1] https://travel.state.gov/en/passports/apply/help/photos.html

I... don't think so but it's probably been over a decade since I last played it. Completely agree that game was tailor-made for touchscreen interfaces.

After trying this one, I'd agree it seems mechanically identical to Flight Control but with a sort of stylistic low-res pixellated 3d look.

Gameplay from 2011:

https://youtu.be/WLHI5Fu--0I?t=48

Not OP, but it looks to be AI generated which was then heavily downscaled in size, since if you zoom in you can still see the usu. telltale suspects such as weird purple/pink/blue color artifacts, etc.

Here's some quick similar examples I generated using Ideogram v4 running locally on my machine. You could also use a multimodal model such as gpt-image-2, give it a specific example jpg (e.g. tiger swallowtail butterfly) and ask it to re-render it as "8-bit pixel art".

https://imgpb.com/wTHcRqQ

I (think) that Vast [1] did something vaguely similar a few years back where regular people could rent out their nVidia GPUs. Might be worth looking into if that's an option.

[1] https://vast.ai

Nice job - it's a similar mechanic to a game I created many years ago where you steer Aladdin on a carpet up and down to the right "lane" for the correct answer when learning multiplication tables while dodging incoming rocks.

https://mordenstar.com/projects/mathemagic

Small bit of feedback: Recommend hiding the little robot or moving him off the physical lanes, I could that being a distraction for a child trying to focus on answering the questions

Moreover, they have a uniform style, even though your prompt doesn’t ask for one.

This shouldn’t really come as a surprise, particularly to anyone who’s used diffusion models. The same thing happens when you ask an LLM for a short story [1] without providing any specific details.

Even cranking up the temperature or top_p values is no panacea. The more generic your prompt, the more pedestrian the response.

[1] https://news.ycombinator.com/item?id=42093394

Nice job. I lived in Taiwan for years and at the time couldn't read simplified very well, so I used a relatively popular browser extension called 新同文堂 which did something similar.

https://github.com/tongwentang/tongwentang-extension

This is by design. Hacker News tends to frown on people who only use it as a tool to promote their own interests.

When you actively participate in the community in good faith, you’ll generally get more feedback and engagement when you do want to showcase something you’ve created.

Yeah and CRT emulation can be a pretty deep rabbit hole. The amount of work that goes into making things skeuomorphic is pretty wild.

If you want to see a neat example, check out cool-retro-term, it’s been around for a while but is still super impressive.

https://github.com/Swordfish90/cool-retro-term

I don't think they updated it - it still looks a very simplistic combination of slightly high frequency noise + artificial horz scanlines.

I grew up in the era, and a couple of my buddies still keep a few CRTs for playing NES games so its immediately apparent to me. Similar to when laymen use AI to generate "pixel art" without doing any grid alignment or palette reduction.

Retroarch also has some decent approximations of CRT scanline using shaders as well (see crt-geom crt-lottes)

https://www.retroarch.com/?page=shaders

It seems like it’s partially based on LeanChess [1], which is 288 bytes long. I’d be curious to know whether this program was AI-assisted or written entirely from scratch, since Lean Chess was written at a time predating the era of LLMs.

Another thing that amuses me is that these tiny programs often claim to be “complete” chess engines while not actually implementing all the rules. This one doesn’t appear to support en passant, and likely doesn't have pawn promotion either.

If you’re allowed to arbitrarily redefine the scope of chess, then code size stops being as impressive a metric.

[1] https://leanchess.github.io

As somebody who grew up in the era of MS-DOS and QBASIC - it's pretty amusing to see two AI-assisted GORILLA.BAS web games in the past week.

https://news.ycombinator.com/item?id=48856350

Can also play the original as well online

https://classicreload.com/play/qbasic-gorillas.html

Feedback

- the wind/round text should NOT be placed over buildings because.... one of the buildings is gray thus making it almost impossible to read.

- I'd honestly get rid of the pseudo-CRT scanline filter - it looks grainy more than anything. For a better reference see https://github.com/gingerbeardman/webgl-crt-shader

- When implementing camera follow one thing you need is a sort of "grace window". Right now it's so tightly coupled to the movement of the banana it practically gives you whiplash.

Nice job.

A lot of early “fantasy word” generators worked in a similar way - by using n-grams at a character (or character chunk) level rather than the word level even if the breaks didn’t align with syllables.

I don't know if it was ever done - but adapting a Markov model built out against a corpus of word IPAs treating each phoneme as a symbol in the chain would have produced better results as well.

Nice job! In case you're interested in this area of research, there was a related paper/demo by one of the researchers who did a lot of work on ControlNets back in the day. Scroll down past the anime examples to see more traditional paintings being animated.

https://lllyasviel.github.io/pages/paints_undo

Small bit of feedback - be wary of relying too heavily on AI to handle UI/UX (making the assumption that you did in the first place); otherwise they tend to converge on the same set of pastels.

https://www.speedpainter.org/en/ai-drawing-timelapse-generat...

At this point it'd be worth creating an old-school webring [1] for all the HN clock enthusiasts. :)

The clock at: https://clocks.dev/clock/lock-screen

is kind of similar to one of my clocks: the Filling Digit clock [2], which fills the hollow digits with water from the bottom up to represent the seconds in a minute:

Another one that made the rounds here on HN was "Alphabetical Clock" [3] which is pretty amusing.

[1] https://en.wikipedia.org/wiki/Webring

[2] https://clocks.specr.net/filling-digits

[3] https://news.ycombinator.com/item?id=47571401

I don't think it's quite that simple - but I'd recommend reading up on the Whorf hypothesis.

https://plato.stanford.edu/entries/linguistics/whorfianism.h...

Nice job. I'm starting to think we need an old-school webring for all the HN clock enthusiasts. :)

The clock at: https://clocks.dev/clock/lock-screen

is kind of similar to one of my clocks: the Filling Digit clock, which fills with water from the bottom up to represent the seconds in a minute:

https://clocks.specr.net/filling-digits

Nice. I need to go check them out again. I really liked their original Retro Diffusion models but this a while back in the SDXL days. It felt like it was trained on more natural language, so prompting it was quite nice for that era.

But if they’ve been working on it specifically for animation, that’s pretty neat.

Thanks for the info! I'll give this a shot. Most of my knowledge is from over a year ago, when I would set up OpenPose-style positioning with a controlnet at a high strength to guide an image-to-image workflow in comfyui. But it was pretty tedious and prone to errors.