vunderba

30 ms·

541.
by vunderba 6mo ago
> https://i.imgur.com/6NXpI2q.png You're killing me Smalls. This one is a 404. I'm really curious what it actually showed. That ring toss is definitely leagues better than its predecessor. I’m not going to fa
542.
by vunderba 6mo ago
OpenAI actually has really good adherence, but occasionally tends to introduce its own almost equivalent of "tone mapping", making hyper-localized edits frustrating. I don’t know how much work it is for you, but one thing a lot of
543.
by vunderba 6mo ago
So the prompts are tuned and adjusted on a per-model basis. If you look at the number of attempts, each receives a specific prompt variation depending on the model. This honestly isn't as much of an issue these days because SOTA models
544.
by vunderba 6mo ago
That's a mistake~ None of the models successfully passed the Flat Earth composition test. I've updated the passing criteria to be more explicit as well. Thanks for catching that!
545.
by vunderba 6mo ago
I’ll have to give it another try. Its predecessor, Hunyuan Image 2.0, scored pretty poorly when I tested it last year: 2 out of 15, so it'll be interesting to see how much it has improved. Here's ZiT, Gpt-Image-2, and Hunyuan Imag
546.
by vunderba 6mo ago
There’s a difference between a side project and a side business. The fact that you’re framing it in terms of monetary reward means you’re always going to be comparing it to your primary source of income. Instead, start with something you wa
547.
by vunderba 6mo ago
> I'm honestly unsure what could be improved at this point. That's because you're focusing a little bit too much on visual fidelity. It's still relatively trivial to create a moderately complex prompt and have it fa
548.
by vunderba 6mo ago
I decided to run gpt-image-2 on some of the custom comics I’ve come up with over the years to see how well it would do, since some of them are pretty unusual. Overall, I was quite impressed with how faithful it adhered to the prompts given
549.
by vunderba 6mo ago
This happens all too frequently when you ask a GenAI model to create an image with a large crowd especially a “Where’s Waldo?” style scenes, where by definition you’re going to be examining individual faces very closely.
550.
by vunderba 6mo ago
OpenAI’s API docs are frustratingly unclear on this. From my experience, you can definitely generate true transparent PNG files through the ChatGPT interface, including with the new GPT-Image-2 model, but I haven’t found any definitive way
551.
by vunderba 6mo ago
I created this little demo of an animated sprite sheet using generative AI. It's not great, but it is passable. https://mordenstar.com/other/hobbes-animation/
552.
by vunderba 6mo ago
> Seems good enough to generate 2D sprites. It’s really not. That's actually a pet peeve of mine as someone who used to spent a lot of time messing with pixel art in Aseprite. Nobody takes the time to understand that the style o
553.
by vunderba 6mo ago
When NB 2 came out I actually had to increase the difficulty of the piano test - reversing the colors of all the accidentals and the naturals, and it still managed it perfectly. https://mordenstar.com/other/nb-pro-2-tes
554.
by vunderba 6mo ago
Are you using Google Gemini directly? I've found the Vertex API seems to be significantly less strict.
555.
by vunderba 6mo ago
It depends on the model. Diffusion models, which are among the more popular approaches, are typically trained at a specific image resolution. For example, SDXL was trained on 1MP images, which is why if you try to generate images much large
556.
by vunderba 6mo ago
LM Arena is a particularly bad comparison site too. Prompts that they use are usually incredibly generic like "A digital render of a sleek, futuristic motorcycle racing through a neon-lit cityscape." I actually built GenAI Showd
557.
by vunderba 6mo ago
There have already been several attempts to procedurally generate Where’s Waldo? style images since the early Stable Diffusion days, including experiments that used a YOLO filter on each face and then processed them with ADetailer. It&#x2
558.
by vunderba 6mo ago
Yep. “Where’s Waldo” has been a classic challenge for generative models for a while because it requires understanding the entire concept (there’s only one Waldo), while also holding up to scrutiny when you examine any individual, ordinary f
559.
by vunderba 6mo ago
OpenAI’s gpt-image-1.5 and Google’s NB2 have been pretty much neck and neck on my comparison site which focuses heavily on prompt adherence, with both hovering around a 70% success rate on the prompts for generative and editing capabilities
560.
by vunderba 6mo ago
It's usually based on what they've been trained on. There aren't very many models that'll do higher resolutions outside of Seedream but adherency is worse.
561.
by vunderba 6mo ago
Nice job. I've seen a couple of "non-mirrored" variations of this where they pre-fill a few of the letters in the grid at random to start you off. Feedback - the heart icon on mobile is right next to the letter A - fat-finger
562.
by vunderba 6mo ago
Back in the days of warez/piracy, I was surprised to never see a "download meter" in exchange for solving CAPTCHAs (that are served from other sites) continuously during the download process if they want to actually get the f
563.
by vunderba 6mo ago
And 100% clientside browser-based PDF tools. I've seen at least 20-30 of them just on Show HN posts since vibe coding became popular.
564.
by vunderba 6mo ago
Nice job. I've seen a couple of these interactive DOM games (breakout is especially popular) but 99% of them have been Firefox/Chrome extensions so it's nice to see one that is a simple bookmark. If you wanted to get really c
565.
by vunderba 6mo ago
Nice job. There's a similar game I used to play for iPads a while back that relied on holding the tablet flat and tilting it to navigate the ball to hit targets. (which itself is kind of like the Monkey Ball games). Feedback CSS3 has
566.
by vunderba 6mo ago
Nice job. Feels a bit like "I Wanna Be the Guy" [1] with trees. You should get an endorsement from Don Cheadle's version of Captain Planet. [1] - https://en.wikipedia.org/wiki/I_Wanna_Be_the_Guy
567.
by vunderba 6mo ago
> You simply cannot do this with the streaming service aggregators or JustWatch. While that might be true, you can also just type "Movies that show smart people under pressure" [1] or similar queries directly into Google, a
568.
by vunderba 6mo ago
They’re completely different projects. ML Sharp is designed to create a 3D Gaussian representation of a depicted scene and output a splat file. Trellis generates model files (GLB I believe) for use with 3D modeling applications.
569.
by vunderba 6mo ago
Nice job. As a fellow edutainment game dev, some feedback: Number 1: As a result of using Gen AI, some of the pictures don’t make sense. In one of the images I saw, you’re looking into the room at an angle, but the directions are still orth
570.
by vunderba 6mo ago
It’s been a while since I heard about it, but IIRC the professor was a stickler for a very specific paper format, so they would distribute a .docx template file with Track Changes already enabled and require students to write their papers

More ›