191–200 of 302 posts
Earlier quoted context omitted.
Did we really get to the point that anything that isn't SaaS seems alien? You know companies sold software that you paid once, and then ran as much as you wanted on your pc?
It doesn't seem odd to me that a product that involves an absurd amount of data and computing power isn't an easily consumable commercial product available for mass download.
Earlier quoted context omitted.
My wife likes impressionists and sunflowers. "A lone sunflower in a grassy field at sunset oil painting claude monet" plus stable-diffusion and a few minutes of tweaking some settings; she now has a new desktop background.
I actually paint and spend a lot of time looking at 'serious' paintings. AI hasn't even scratched the field to a trained eye. Doesn't mean I'm not excited though. This kind-of feels like I'm watching the camera or printing press being invented. Everyone is comparing it to fine art, but I think ultimately it's going in a different and bigger direction.
Earlier quoted context omitted.
Stable Diffusion allows anyone to make kiddie porn with a half-second of curiosity/effort. Maybe you didn't know about that, maybe you think it's NBD, but in any case, that is the tire fire which aspiring AI majors want to avoid.
Pen and paper can do the same. Or Photoshop. Anyone can draw anything! OMG, stop the paper factory.
I'm pretty sure one of the primary arguments for Dall-E and Stable Diffusion existing is that there are lots of people who can't draw anything.
Earlier quoted context omitted.
>that gets my dander up Your w0t m8?
Endangered Words Bureau Agent D23 at your service
"Use Them or we will lose Them!"
I used Twixt (An abbreviation of Betwixt) to replace inbetween in a submission.
Edit: to fix formatting and spelling
Earlier quoted context omitted.
They’re just harder to run on your own resources, since large language models are very large. BLOOM was released a month ago, is likely better than GPT-3 in quality, and requires 8 A100s for inference, which pretty much no one has on their desk.
Can anyone confirm if BLOOM is better than GPT-3 at instruction following? I might have read somewhere that it's not as well behaved.
BLOOM incorporates some new ideas like ALiBi which might make it better in a more general sense. They haven’t released official evaluation numbers yet though so we’ll have to see.
Earlier quoted context omitted.
Stable Diffusion is arguably better, has more features, and is free. OpenAI can't compete with free. Even if you don't want to take the 30 seconds to set it up in a free Google Colab environment, the paid DreamStudio version is still half the price of Dalle.
I find Stable Diffusion better overall, but it has downsides. Stable Diffusion tends to be more creative than DALL-E, but does a lousy job of following directions, especially complex ones. DALL-E is good if I know what I want specifically. I can think of ways to fix Stable Diffusion since it's open-source. I think I could bridge the gaps as I see them in about a weekend of hacking. I'm not sure when I'll get that wee…
This was already possible with DALL-E using the inpainting feature going from defined image to transparent edge; this just automates what was a manual process before. Do wish the inpainting tool had more options, for example to fade a transparency in, since my understanding is it makes a difference; not to mention magic wand selection/deselection tool. In case it is not obvious, every time a user generates an additio…
And while OpenAI released this feature now, it's probably just a matter of days until even better features built on Stable Diffusion will be released, given how much community energy is focussed on it right now.
Earlier quoted context omitted.
Running this at home is only free like mining cryptocurrency is free if you didn't buy your computer and don't pay for the electricity. Plus you can only run it on the computer that has the good graphics card, which probably isn't your laptop. I expect most people aren't going to be generating images all day, so using a cloud-based service for occasional use will still make a lot of sense. Stable Diffusion offers a p…
It does run on Apple silicon. 55 seconds in M1 Pro (vs 15 seconds on RTX 3070).
Earlier quoted context omitted.
Not sure if GP had this in mind, but SD is (more) free in terms of liberty. So yes, you pay with electricity and hardware, but you control the process yourself, which is invaluable. DALL-E could change or go offline at any time.
Considering the threat from DALL-E going offline, it seems quite acceptable. These aren't precious photos since it's all made up anyway, you can download any pictures you make, and you probably already did for the ones you care about. I'd worry more about, say, keeping your photos on Google and losing your account somehow.