SageBrush: AI Painter – the simplest way to create with AI | Hacker News Reader

SageBrush: AI Painter – the simplest way to create with AI

apposite.ai

19 pointsgdubs15 comments

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.

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"?

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.

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.

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 :)

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?

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.

I’d totally give this a shot if it wasn’t supported only on ipads and M-chip Macs

It’s unclear how many tokens you have and how many are using as you use the app. It’s also easy to “waste” them as you type a prompt but aren’t finished yet.

Replied re: tokens elsewhere on this page, but good feedback on the live prompt editor.

That was something people really wanted during beta testing, but I can see your point! Perhaps a setting to turn that off for people who don't consider those live updates a feature.

Looks like something I'd love to have on an Android foldable. Web based stuff never handles the pen well enough.

uses tokens, pricing for tokens not available.

They’re in the AppStore page:

225 for $0.99 4500 / month for $9.99

And one token gives you one image of what resolution? Also why isn't that on the app page too? Why is pricing hidden? I consider hiding pricing to be a big dark pattern and makes me wary of any sort of interaction.

That's helpful feedback, thanks.

Sounds like it feels like a dark pattern to you. My goal here isn't to trick people. That's why I decided to offer a $1 token pack – I figured it's low stakes for people to try it out, see how much drawing time they get out of it, and decide if it's worth continuing / subscribing for a better value on tokens.

It's also a fundamentally different experience than something like midjourney etc, where you prompt, wait, get a finished picture. The focus of this app is more about the somewhat 'real time' nature of seeing the drawing evolve. And I didn't want to make the token count so prominent in the UI that people were just fixated on it all the time.

FWIW, in version 1, 1 token = 1 brushstroke / one prompt change / one random seed change, producing an image at 768px. Both the resolution options and the value of a token are something I'm aiming to offer more flexibility on in the next iterations.

So, technically speaking $1 = 250 images. But, I didn't want to explicitly quantify it like that, because it typically takes some brushwork to get to an image you're happy with.

But your feedback is helpful. I'm hoping that with a broader set of users I should be able to get something more like "average drawing time" for the different sets of tokens. And like I mentioned elsewhere, once I'm certain I'm not going to break the bank on my hard costs, I'd love to both come down on price and offer free 'relaxed' usage.

Reply on news.ycombinator.com