Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Stable Diffusion 3 API Now Available

stability.ai

Post reply on HN
Print view View original

21–30 of 66 posts

Re: Stable Diffusion 3 API Now Available

#21

Post by satvikpendem »

Turing_Machine wrote: post #17
MyFirstSass wrote: post #15

Earlier quoted context omitted.

A tangent but i believe all pricing should public information by law above some threshold to battle corruption, nepotism, cartel making and monopolies. I've seen too many governments and even medium sized companies paying absurd amounts to some shitty locked in cloud platform with lots of better alternatives, because someone got kickbacks, gifts, vacations or a seat on some board, and everyone should be more enraged.…

Unless there's absolutely no alternative, and it's something I desperately need, I basically won't buy anything that says "call for pricing" or some such, because I interpret that as "call so one of our trained liars (i.e., salesmen) can figure out how hard we can screw you". In many cases I'd even pay a premium to avoid talking to one of those people.

You're not the target market, the CTO, CIO, or business development representatives are. So many developers say they won't ever buy "call for pricing" software, like, yeah, they know you won't, in fact that's what they bank on.

Re: Stable Diffusion 3 API Now Available

#22

Post by wongarsu »

MyFirstSass wrote: post #2

- Models will be released - Free to use for non commercial projects - 20$/m subscription needed if project has less than $1m annual revenue or - Undisclosed "Enterprise" subscription needed for projects larger than above To me the $240/y sounds pretty awesome for a smaller commercial project, and i like they've set the threshold pretty high up. But still don't like "undisclosed" black box enterprise pricing though wh…

> so in effect only viable to create small companies with this tech

I assume stability wants to be in the role of building AI models not applications, so if you provide a clear and nontrivial value add beyond their API you should be fine. If you build a tool that allows architects to make their building renderings more lifelike by adding realistic crowds, vegetation, storefronts etc right from the comfort of their existing tooling it's in Stability's interest that you are successful. If you only resell their API but in green they might not be as invested in your success.

Re: Stable Diffusion 3 API Now Available

#23

Post by wongarsu »

the_duke wrote: post #6

This Reddit thread from someone with early access has some sample images: https://old.reddit.com/r/StableDiffusion/comments/1c2je28/i_... Maybe the early access beta was limiting the available resources, or they used bad settings, bud juding by that thread it looks like the model got worse during training or the earlier examples were quite cherry-picked.

To be fair, this has been my initial impression with basically all image generation models. The current generation is finally at the point where throwing untuned prompts at untuned models gives good results, but those never match the results of finely tuned (positive and negative) prompts with parameters adjusted from experience; ideally with model fine-tunes added.

If you want the best results there is still skill and work involved. Consequently a showcase by people experienced with the model far surpasses what you get when you shout prompts over the internet for somebody else to try

Re: Stable Diffusion 3 API Now Available

#24

Post by zone411 »

I ran the same quick prompt adherence and composition test on which ImageFX by Google surpasses DALL-E 3 by a bit (https://www.astralcodexten.com/p/open-thread-315/comment/493...):

1. "A stained glass picture of a woman in a library with a raven on her shoulder with a key in its mouth"

2 out of 20 tries

2. "An oil painting of a man in a factory looking at a cat wearing a top hat"

First try

3. "A digital art picture of a child riding a llama with a bell on its tail through a desert"

0 out of 20 tries

4. "A 3D render of an astronaut in space holding a fox wearing lipstick"

2 out of 20 tries

5. "Pixel art of a farmer in a cathedral holding a red basketball"

First try

So about even with these models and much better than previous versions of SD. Better than Midjourney v6.

Re: Stable Diffusion 3 API Now Available

#25

Post by alexvitkov »

wongarsu wrote: post #22
MyFirstSass wrote: post #2

- Models will be released - Free to use for non commercial projects - 20$/m subscription needed if project has less than $1m annual revenue or - Undisclosed "Enterprise" subscription needed for projects larger than above To me the $240/y sounds pretty awesome for a smaller commercial project, and i like they've set the threshold pretty high up. But still don't like "undisclosed" black box enterprise pricing though wh…

> so in effect only viable to create small companies with this tech I assume stability wants to be in the role of building AI models not applications, so if you provide a clear and nontrivial value add beyond their API you should be fine. If you build a tool that allows architects to make their building renderings more lifelike by adding realistic crowds, vegetation, storefronts etc right from the comfort of their ex…

"Should be fine" doesn't cut it if you want to base your business on it and have no alternatives.

Re: Stable Diffusion 3 API Now Available

#26

Post by wongarsu »

alexvitkov wrote: post #25
wongarsu wrote: post #22

Earlier quoted context omitted.

> so in effect only viable to create small companies with this tech I assume stability wants to be in the role of building AI models not applications, so if you provide a clear and nontrivial value add beyond their API you should be fine. If you build a tool that allows architects to make their building renderings more lifelike by adding realistic crowds, vegetation, storefronts etc right from the comfort of their ex…

"Should be fine" doesn't cut it if you want to base your business on it and have no alternatives.

Every Android or iOS App is a "should be fine" type of deal. So are lots of other things. If you are really worried I'm sure there's a sales person willing to take your call right now. It's not ideal, but honestly not that exceptional. And if push comes to shove you do have alternatives, you just need to sacrifice a bit of quality, maybe deployment complexity, and tune your prompts differently for one of the other models.

Re: Stable Diffusion 3 API Now Available

#27

Post by Kerbonut »

zone411 wrote: post #24

I ran the same quick prompt adherence and composition test on which ImageFX by Google surpasses DALL-E 3 by a bit ( https://www.astralcodexten.com/p/open-thread-315/comment/493... ): 1. "A stained glass picture of a woman in a library with a raven on her shoulder with a key in its mouth" 2 out of 20 tries 2. "An oil painting of a man in a factory looking at a cat wearing a top hat" First try 3. "A digital art picture…

You're essentially mixing units. "2 out of 20" does not match with "first try". I would have liked to see you run all of them for 20 and added comments in addition like "this got it right on the first try", which could also have been luck. I mean if it got 1 out of 20 but happened to get it right the first try, is that better or worse than 2 out of 20?

Re: Stable Diffusion 3 API Now Available

#28

Post by Sharlin »

satvikpendem wrote: post #21
Turing_Machine wrote: post #17

Earlier quoted context omitted.

Unless there's absolutely no alternative, and it's something I desperately need, I basically won't buy anything that says "call for pricing" or some such, because I interpret that as "call so one of our trained liars (i.e., salesmen) can figure out how hard we can screw you". In many cases I'd even pay a premium to avoid talking to one of those people.

You're not the target market, the CTO, CIO, or business development representatives are. So many developers say they won't ever buy "call for pricing" software, like, yeah, they know you won't, in fact that's what they bank on.

And the execs won’t really worry about "how hard the salesmen can screw them", they know the rules of the game and understand that negotiations are negotiations. It’s a negotiator’s job to try to get a good deal, and that works both ways.

Re: Stable Diffusion 3 API Now Available

#29

Post by zone411 »

Kerbonut wrote: post #27
zone411 wrote: post #24

I ran the same quick prompt adherence and composition test on which ImageFX by Google surpasses DALL-E 3 by a bit ( https://www.astralcodexten.com/p/open-thread-315/comment/493... ): 1. "A stained glass picture of a woman in a library with a raven on her shoulder with a key in its mouth" 2 out of 20 tries 2. "An oil painting of a man in a factory looking at a cat wearing a top hat" First try 3. "A digital art picture…

You're essentially mixing units. "2 out of 20" does not match with "first try". I would have liked to see you run all of them for 20 and added comments in addition like "this got it right on the first try", which could also have been luck. I mean if it got 1 out of 20 but happened to get it right the first try, is that better or worse than 2 out of 20?

Like I said, it's a quick test, not a benchmark. The original question was about getting at least one of out 10 right (https://www.astralcodexten.com/p/a-guide-to-asking-robots-to...). Feel free to run them yourself, takes 5 minutes.

Re: Stable Diffusion 3 API Now Available

#30

Post by satvikpendem »

Sharlin wrote: post #28
satvikpendem wrote: post #21

Earlier quoted context omitted.

You're not the target market, the CTO, CIO, or business development representatives are. So many developers say they won't ever buy "call for pricing" software, like, yeah, they know you won't, in fact that's what they bank on.

And the execs won’t really worry about "how hard the salesmen can screw them", they know the rules of the game and understand that negotiations are negotiations. It’s a negotiator’s job to try to get a good deal, and that works both ways.

Yes. Devs who don't know sales are obviously not their demographic, they'll be speaking to layers of other biz dev people before they even get to the execs, and each one knows exactly how to extract concessions and discounts, and the opposing salesperson knows exactly how much of a discount they can give. In this way, it is much more of an efficient market than most devs think.

Devs' reaction to all this is honestly the same as when a non-technical exec tells a dev why they "can't just get it done in a day, it shouldn't be that hard right?" People have their own competencies and are usually blind to others'.

Post reply on HN