The pay-per-generation system makes sense for a first version, and is easily tunable to reduce load. But if this were a mature tool what payment options would you want?
I’ve been daydreaming about an “unlimited” monthly subscription that limits your rate in bursts, so you can only do 1-2 a minute, only 50 an hour, 300 a day etc. but otherwise can eliminate that credit-anxiety I imagine a lot of users feel.
Has anyone else been thinking about what they would want to see in the future?
I’ve been thinking the same thing! I hate having to run out of credits, but I feel like there would be a better way to pay for the AI to create an image, and it could be an unlimited credits subscription for like…ranging around $10 a month - $30 a month.
They better do something because the competition is looking fierce!
They better do something because the competition is looking fierce!
What competition are you talking about? I know of nothing that’s as close that’s public. Do share what the alternatives are.
This might work for individuals, but OpenAI isn’t interested in individuals, I wouldn’t imagine. They are looking to provide this as a service for developers, in which case a similar model to the GPT3 model would likely be what’s on offer. It will likely be expensive, to begin with until they can optimize the models, again similar to what they’re doing with GPT3.
What competition are you talking about? => Midjourney, stable diffusion.
For comparaison Midjourney offers a 30$ unlimited option.
Bloom, GPTJ, GPT neo, OPT, Google t5, GPTX, and Amazon’s new 20B model. Also, it’s anyone’s guess when Google and DeepMind will release some of their wildly powerful models to the public.
Many of these are currently hosted at forefront.ai for very competitive prices.
There are also many people building private servers and hosting themselves. Lastly, some of these can even be run on consumer grade GPUs at home.
Just realized that I commented on a whole bunch of competitive language only models. Stable diffusion, mid journey, and disco diffusion are the main text image generators currently. That said, lots of big players have not even gotten into the game yet… Amazon, Microsoft, Google, etc
This makes sense to me, and it’s hard to know what they’re thinking inside openai but based on their blog they do understand they’ve created something really valuable to individual creators. Like I see a world where even for developers they maintain their existing UI as a sandbox, and you could just buy a slightly different payment/usage plan if you were planning to exclusively use the UI as an individual as opposed to hitting their API as a developer. The only actual work to do on their side would be fiddling with the credit-buying system which seems easy enough.
Look at NightCafe’s pricing to benchmark. They launched their newest “stable” model this week, and the quality and progress are extraordinary. I’m not a DALLE-E2 beta user so no first hand experience, but I’m a NC addict, and from what I’ve seen of the DALLE-E2 output, NightCafe’s new models are competitive to say the least.
MidJourney is the only one I find competitive w/ Dalle2 right now… their new beta was creating crazy things… arguably better than Dalle2 if I am allowed to say so. I do like Dalle’s experience better (and dalle is faster)
For a pricing model, I think offering X free prompts per month is wise so that EVERYONE can try. Hobbyists will want a few more so have a cheap plan, 5USD/month for “a few more” prompts. After that, with pros I don’t know, I’d have to look at the usage stats. I don’t think “unlimited” is a good idea cause you can expect some people to hit your API 24/7. However if they are willing to pay anything…
And the last consideration is the competition of course.