I just got a much better version using this command instead, which uses the maximum image size according to https://github.com/openai/openai-cookbook/blob/main/examples... OPENAI_API_KEY="$(llm keys get openai)" \ uv run 'https://raw.githubusercontent.com/simonw/tools/refs/heads/main/python/openai_image.py' \ -m gpt-image-2 \ "Do a where's Waldo style image but it's where is the raccoon holding a ham radio" \ --quali…
The faces...that's nice that it turned a kid's book into an abomination
it's interesting, zoomed out it kind of looks ok, zoomed in.... oh my.
So during my Nano Banana Pro experiments I wrote a very fun prompt that tests the ability for these image generation models to follow heuristics, but still requires domain knowledge and/or use of the search tool: Create a 8x8 contiguous grid of the Pokémon whose National Pokédex numbers correspond to the first 64 prime numbers. Include a black border between the subimages. You MUST obey ALL the FOLLOWING rules for th…
This is an amazing test and it's kinda' funny how terrible gpt-2-image is. I'd take "plagiarized" images (e.g. Google search & copy-paste) any day over how awful the OpenAI result is. Doesn't even seem like they have a sanity checker/post-processing "did I follow the instructions correctly?" step, because the digit-style constraint violation should be easily caught. It's also expensive as shit to just get an image th…
It's always felt like way overhyping to call something "democratization" when it's something I could do as a middle schooler in 2005. It takes some skill to do very well but it's not like basic diagram creation isn't something people already could do for basically free (I create figures for my job all the time now and chatGPT is more expensive than tools I use for design). Commissioning high quality diagrams from a d…
You are making a mistake a lot of people make when talking about genAI helping others do work. I get that to you it is very easy to do, but there are other groups of people that are not able to do it. What you are saying is like a hobbyist carpenter saying that making a bedside table would take him one weekend to do, so he doesn't think it is okay for tables to be made via assembly line instead of hiring a carpenter…
I think you're missing my point, which is pretty narrow here. "Democratization" is fairly grand term implying that the general public now have access to something freeing they didn't before (it generally invokes some idea of liberation, as the term often is used to note a transition from an authoritarian to a democratic government). I don't think there has ever been a particularly high barrier to making good diagrams, in my experience it's an easy to learn skill both in time and money, so it feels like it's cheapening the term "democratization". Maybe I'm being a bit sensitive though because of how the world is right now with people sometimes literally fighting for democracy. Normally I am pretty lax with semantics but I've had some people really rub me the wrong way when overhyping AI.
Because I'm not an artist and can't afford to pay one for whatever business I have? This idea that only experts are allowed to do things is just crazy to me. A band poster doesn't have to be a labor of love artisanal thing. Were you mad when people made band posters with MS word instead of hiring a fucking typesetter? I just don't get it.
I actually love MS word posters. It's a million times more authentic and enjoyable than a slop generation. If a band put up an AI poster I'd assume they lack any kind of taste which is the whole reason I'd want to listen to a band anyway. I know this is controversial in tech spaces. But most people, particularly those in art spaces like music actually appreciate creativity, taste, effort, and personal connection. Not…
Because I'm not an artist and can't afford to pay one for whatever business I have? This idea that only experts are allowed to do things is just crazy to me. A band poster doesn't have to be a labor of love artisanal thing. Were you mad when people made band posters with MS word instead of hiring a fucking typesetter? I just don't get it.
Because I'm not an artist and can't afford to pay one for whatever business I have? If your business can't afford to spend $5 on Fivr, it's not a business. It's not even panhandling.
Why is that better? They're going to use AI anyway. It's fiver.
Image generation? Hmm, would be cool if OpenAI also made a video-generation model someday..
If only there was a social network with solely AI generated videos, I would pay literal money for it...
If I may address this with both skepticism and curiosity, why. I think I speak for everyone when I say I would pay to go back to facebook 2018. No algorithm, no ai.
I love how perfectly this captures the difficulties of using generative AI for detection tasks.
Oh god yes, I've been trying to make a LLM Assisted Magic the Gathering card scanner... its been a hell of a time trying to get it to just OCR card names well....
The quality of the text is really impressive and I can’t seem to see any artefacts at all. The fake desktop is particularly good: Nano Banana would definitely slip up with at least a few bits of the background.
There are a couple of AI-esque misspellings - in the More Myth than Menace wolves image, on the right in the "at a glance" section, it reads "wolves aarely approach people," and in the Typography image the text in the top right is "Type connncts us all."
But yeah the quality is remarkable, and rather scary.
I feel like asking the image generators to mark AI images is the wrong way to go about it. It's like trying to maintain a blocklist. It seems better to me to have the major camera manufacturers or cell phones cryptographically sign their images as real.
I feel like this idea comes up often and in my opinion it doesn't solve anything. Take a picture of an AI image and you've made this approach useless. Which then goes to the argument of "well you'll see it's a picture of a picture" to which I will say there are plenty of ways to make this not appear so, and the ultimate form of this argument is that you can eventually project light directly into the photosensors, or…
SynthID survives basic transforms including screenshots/photos, although it can of course be defeated. Even still it helps with the laziest fakes, which there seem to be a lot of - I've seen several quite widespread misinformative images over the past couple months that failed a synthID check.
Anyways I think approaching the problem from both directions is probably good.