3 ms·

Hi, everyone! Been a long time since I've launched something on Hacker News. Gregory here, creator of Polychord for iPad. Today my new app SageBrush: AI Painter

by gdubs 3y ago

Hi, everyone! Been a long time since I've launched something on Hacker News. Gregory here, creator of Polychord for iPad. Today my new app SageBrush: AI Painter was approved and released on the App Store. I wanted to make an AI tool that makes people a part of the process. With SageBrush you draw – either with an Apple Pencil or your fingers – and SageBrush turns your scribbles into AI imagery. I also wanted SageBrush to be simple. Something that would appeal equally to my kids, or to an architect who's looking for a new brush in their toolkit. SageBrush is built on Replicate, and I want to shout-out to them again and thank them for building a really nice API, and for their support during the development of this project. Can't wait to hear what you think.

mstolpm 3y ago
Great project, congrats. Can you detail how tokens are used? Is that per generated version, per final (?) image, per time, export, size? Another feedback: I initially had the wrong impression that the left side was the "input" and the right side the "output" and wondered it the app gave initial renderings the user could "uglify". The video of course cleared that confusion. But perhaps more people are like me "reading" the screenshots from left to right as in input -> output? From a UX standpoint, drawing on the side side of the screen for input of course makes a lot of sense. Perhaps the website could label an example with "our input" and "generated result"?
- gdubs 3y ago
  Thanks! Tokens are per generation. Interesting feedback on the canvas orientation. There's a zero state label that tells you where to draw / where the AI will show up. But sounds like maybe it's a little subtle. The biggest reason for how the canvases are arraigned is to support both left and right handed users. There's a button in the top nav that allows you to switch. Next version I'm hoping to add a couple of tool tips to explain some of the icons, or make them somehow a little more self explanatory.
  - mstolpm 3y ago
    Thanks. I'm still a bit confused about how many tokens for example your example video uses up. You draw the body and it generates a first iteration, then you draw the eyes and it seems there are two new generations, then some more for the background. As mentioned elsewhere, the token system (and potential subscription model) is a bit of a showstopper for me, but the concept is great. Another input: Perhaps I have missed it, but you should mention export options and sizes of generated results on your website. Another question that pops up is the usage rights of the generated images.
    - gdubs 3y ago
      Ah, I see. Without seeing hands in the video, it's hard to see every last brushstroke – but that's how tokens are deducted, per-brushstroke. They're also used when you hit the dice button for a random seed, or change the prompt, etc. Anything that causes the rainbow loading indicator to show up and generate a new image is going to deduct a token. Totally get it – and it's fair feedback, and I anticipated that figuring out the right pricing model would take a little time and feedback. I'd rather have the wiggle room to come down on cost once I'm sure that my calculations on hard-costs are on target. That was the motivation for offering a $1 token pack – provide anyone a chance to at least play around with it, have some fun with the app, get some feedback. Good idea to list the output size – hoping to have some additional options there soon as well. As far as copyright of the generated imagery, it's linked in-app under the terms, etc. (Worth reading, and I can make it more prominently accessible on web, etc.) Gist is that whatever the user makes is theirs to use, but the app doesn't make any claims or guarantees – ultimately the user's responsibility to ensure their work doesn't violate someone else's copyright. Still waiting for my coffee to kick in so probably worth saying that the actual terms in the app superseded any early-morning Internet forum replies of mine :)
sgslo 3y ago
Token-based pricing gives me a bit of anxiety. Concern that'll never use them, or that I'll use them too quickly. Any thoughts on a pricing model where I pay an upfront cost, but get to generate images locally on my own machine?
- gdubs 3y ago
  A local model and upfront cost isn't something that's likely in the short term, but not out of the cards entirely. I'm hoping to eventually add a free tier generation mode for once you've run out of "fast" generations. For version 1 I went with the simple token approach — my main priority was to make sure I was covering costs as I scale things up and get feedback for the next iteration.

by gdubs 3y ago

mstolpm 3y ago

Great project, congrats. Can you detail how tokens are used? Is that per generated version, per final (?) image, per time, export, size? Another feedback: I initially had the wrong impression that the left side was the "input" and the right side the "output" and wondered it the app gave initial renderings the user could "uglify". The video of course cleared that confusion. But perhaps more people are like me "reading" the screenshots from left to right as in input -> output? From a UX standpoint, drawing on the side side of the screen for input of course makes a lot of sense. Perhaps the website could label an example with "our input" and "generated result"?

gdubs 3y ago
Thanks! Tokens are per generation. Interesting feedback on the canvas orientation. There's a zero state label that tells you where to draw / where the AI will show up. But sounds like maybe it's a little subtle. The biggest reason for how the canvases are arraigned is to support both left and right handed users. There's a button in the top nav that allows you to switch. Next version I'm hoping to add a couple of tool tips to explain some of the icons, or make them somehow a little more self explanatory.
- mstolpm 3y ago
  Thanks. I'm still a bit confused about how many tokens for example your example video uses up. You draw the body and it generates a first iteration, then you draw the eyes and it seems there are two new generations, then some more for the background. As mentioned elsewhere, the token system (and potential subscription model) is a bit of a showstopper for me, but the concept is great. Another input: Perhaps I have missed it, but you should mention export options and sizes of generated results on your website. Another question that pops up is the usage rights of the generated images.
  - gdubs 3y ago
    Ah, I see. Without seeing hands in the video, it's hard to see every last brushstroke – but that's how tokens are deducted, per-brushstroke. They're also used when you hit the dice button for a random seed, or change the prompt, etc. Anything that causes the rainbow loading indicator to show up and generate a new image is going to deduct a token. Totally get it – and it's fair feedback, and I anticipated that figuring out the right pricing model would take a little time and feedback. I'd rather have the wiggle room to come down on cost once I'm sure that my calculations on hard-costs are on target. That was the motivation for offering a $1 token pack – provide anyone a chance to at least play around with it, have some fun with the app, get some feedback. Good idea to list the output size – hoping to have some additional options there soon as well. As far as copyright of the generated imagery, it's linked in-app under the terms, etc. (Worth reading, and I can make it more prominently accessible on web, etc.) Gist is that whatever the user makes is theirs to use, but the app doesn't make any claims or guarantees – ultimately the user's responsibility to ensure their work doesn't violate someone else's copyright. Still waiting for my coffee to kick in so probably worth saying that the actual terms in the app superseded any early-morning Internet forum replies of mine :)

sgslo 3y ago

Token-based pricing gives me a bit of anxiety. Concern that'll never use them, or that I'll use them too quickly. Any thoughts on a pricing model where I pay an upfront cost, but get to generate images locally on my own machine?

gdubs 3y ago
A local model and upfront cost isn't something that's likely in the short term, but not out of the cards entirely. I'm hoping to eventually add a free tier generation mode for once you've run out of "fast" generations. For version 1 I went with the simple token approach — my main priority was to make sure I was covering costs as I scale things up and get feedback for the next iteration.