vunderba

4 ms·

1.
by vunderba 5d ago
I agree. In fact the entire reason I initially built GenAI Showdown was because many of the comparative tests on places like Image Arena Leaderboard [1] are not designed to challenge models on prompt adherence. Even when they are, a conside
2.
by vunderba 5d ago
If you want to look at a company that does pixel art right for the modern era, Yacht Club Games is probably the gold standard with games like Shovel Knight and Mina the Hollower. https://www.yachtclubgames.com/blog/the-
3.
by vunderba 5d ago
Well, the results are in, at least for text-to-image (the editing bench will come later). Qwen-Image 2.1 is definitely a pretty big leap over the last open-weight version, Qwen-Image 1.0, released back in August of last year and managed to
4.
by vunderba 6d ago
Agreed. There's also a lot of bad tinging/yellow saturation that very much reminds me of early gpt-image outputs on a lot of the non-cherry picked stuff I've been seeing on Twitter/Reddit. A lot of people were putting Zi
5.
by vunderba 6d ago
Well this is HN - original home of the "ummm actually..." - so I appreciate when people pick all the nits. :) Even though I prompted for a crucible in the prompt, I think the fact that the prompt also contained terms like “blacksm
6.
by vunderba 6d ago
Wait... is that true? I don't think the original Queen Image 2.0 was ever released beyond an API. At least, I don't remember a public weights release.
7.
by vunderba 6d ago
Can do! GPT-Image-2 already scored unsurprisingly very high: 12 out of 15 on text-to-image, and 10 out of 12 on image-to-image. The three benchmarks it failed on (D20, Flat Earth, and Banded Snake) are pretty difficult, so I'd be surpr
8.
by vunderba 6d ago
That’s a good catch. Yes, all scoring is done through manual review since relying on a VL model for these kinds of meta-metrics is a sort of loose equivalent of gödel's second incompleteness theorem. I’ll have to think about this one.
9.
by vunderba 6d ago
This is my experience as well. Ideogram4 (assuming you are willing to put in the work to use the proper structured JSON input) is very accurate when it comes to text rendering in an image.
10.
by vunderba 6d ago
So thoughts Positives • It's a heck of a lot smaller than Qwen-Image 1 (20b parameters) at only 7b, making it one of the smaller open-weight models available (Z-Image Turbo is one of the few that is smaller at 6b) when compared to Ideo
11.
by vunderba 6d ago
Boogu-Image has the Apache 2.0 License [1] (good coherence, but outputs can look synthetic). And Krea 2 has a community license [2] that is fairly permissive - I think commercial usage is allowed under $1 million. Boogu-Image scored 6/
12.
by vunderba 8d ago
Nice. Some Feedback • Wish this was a web app though, I have a general policy of not isntalling unknown software on my computer. • Website needs a demonstration video (or at the very least some screenshots) If you're interested in impr
13.
by vunderba 9d ago
Yue2 is absurdly fun to mess around with if you're a musician and basically feels like band-in-a-box on steroids. It's very easy to take old midi tracks you've written and then get Yue2 to do melodically accurate covers with
14.
by vunderba 11d ago
As a kid I vaguely remember our family had a stack of Dr. Dobb’s Journals with one particular article actually warning about the dangers of doublespace, and since it was completely optional (despite being touted as a flagship feature) we ki
15.
by vunderba 11d ago
When I lived overseas (Asia), I applied for FEIE every year which lets you exclude ~120k of earned income from the year. Made having to pay taxes in two countries far less painful. Unless you intend on making very little (such that the stan
16.
by vunderba 12d ago
Nitpicking a bit - once again another FOV of ~110. Sigh. Maybe I’m just being picky, but I feel like I’m playing a submarine periscope simulator when I use these things. It’s been over a decade since the early VR headsets like the Rift, and
17.
by vunderba 12d ago
The “glasshole” vibe wasn’t antisocial; it was an understandably human reaction to the fact that you couldn’t tell whether the wearer (and/or google) was surveilling or recording you any time they so much as glanced in your direction i
18.
by vunderba 12d ago
Wrt entertainment I think we're still pretty far off. You have to do quite a bit of handholding (particularly maintaining coherence across multiple minutes of video) so its mostly the hobbyists that are making stuff right now. Minimax
19.
by vunderba 12d ago
Nice job, but I'm surprised you're not a bit more worried about DMCA takedowns, given that Nintendo has a history of being pretty litigious about this sort of thing. Doldecomp (and most reverse engineering projects) work around th
20.
by vunderba 13d ago
I think we’re in agreement. I make heavy use of smaller local models on a daily basis (Qwen3-VL for auto-captioning images, Gemma3:27b for some translation work, etc.). Gemma3:27b is a good example of a very capable general purpose multimo
21.
by vunderba 13d ago
I read about a study a while back in which researchers brought in a person who had synesthesia and asked him to identify the exact colors associated with 500 different random numbers drawn from a range of up to a million. (He did not know w
22.
by vunderba 13d ago
> People are realizing that what they need isn't more general intelligence, it's more specialization. A small but well tuned coding model... It’s not quite as simple as that. Several studies have shown the opposite: models tr
23.
by vunderba 15d ago
Nice job. You should make it so there’s a chance that if you try to navigate to another area too quickly the narrator says, “You must gather your party before venturing forth.” This is super nostalgic, but BG1 is still actively sold particu
24.
by vunderba 15d ago
Related 2025 paper on how GenAI running in an unconstrained loop tends toward convergent collapse: https://pubmed.ncbi.nlm.nih.gov/41583980
25.
by vunderba 15d ago
I always run stairs when given an opportunity (big proponent of "incidental exercise"), so when I used to live in apartments I would deliberately rent a room in the highest available floor. It's shocking how much HIIT you can
26.
by vunderba 15d ago
Same! I used to play Max 300 in heavy mode (on repeat) at my local arcade without hugging the bar - your feet end up moving like Donald Duck from Mathemagic Land. 30 minutes and you are gassed... Nowadays I switched over to Pump It Up - I f
27.
by vunderba 15d ago
This seems to be one of the few open-weight models that isn't just another "text to audio" generator. You can take existing sheet music, convert it to ABC notation, and then experiment with various covers. I don't have a
28.
by vunderba 16d ago
There's a great demo video of it as well. https://youtu.be/uu7jA8EHi_0 Some of the details in the paper are super cool. Since the game pulls "c" down to near walking speed, you receive photons at different ra
29.
by vunderba 17d ago
Reminds me of wired elements (which uses Rough.js under the covers). https://github.com/rough-stuff/wired-elements
30.
by vunderba 17d ago
> everything is period-correct EXCEPT for the hard drive and floppy drive Hence why I couched my assertion with "except". IMHO its the best way to enjoy the original hardware experience without all the inevitable failures of

More ›