Show HN: I made a site that easily lets you generate AI images using templates
phantasmagoria.stavros.io
phantasmagoria.stavros.io
Some questions/feedback that come to mind:
- How do I know how many credits my generation is going to cost? I ran one D&D template and the cost was 2.4 credits. Also a question relevant on the normal "Generate" tab where I have options to adjust - presumably higher quality costs more credits since it takes more GPU time to run more iterations?
- When generating from a template, there's no option to make the image private until it's been generated. If I were paying for a tool like this I'd expect a way to default to private. Perhaps making images private and defaulting to private can be features for paying users, so that the "public" images for free users are also free advertising for your frontpage?
- Rights issues around the images are unclear. Are these 100% mine to do with as I please, just as if I had generated them running SD myself? Clarification around this is probably important for anybody wanting to use a tool like this in commercial creative endeavors.
- Maybe there should be a username filter? I was able to register as "admin" but if you need that changed I'll happily take "root" or "system" instead ;)
1. That's a bit of an omission right now, you're right. I need to add a bit of JS to the generation page, thanks for the feedback.
2. You can click the "Private" button while it's generating, and it will be hidden. Though, you're right, anyone who sees it before will be able to see it after too. I'll add a checkbox to the form right now.
3. You're right, I need to add a license page. It's the same license as SD, so IIRC it's a perpetual license to use them as you see fit.
4. Yeah I should probably have a username blacklist, you're right :P
Thanks, this is very useful!
-----
> anyone who sees it before will be able to see it after too
I can see where opinions might differ, but to me "Private" means nobody could see it even with a link, while "Unlisted" means I could send the link to anybody but it won't appear in my profile/the front page.
Maybe the middle ground if you don't want to bikeshed on naming things (relatable) would be including a small explanation alongside the checkbox during image generation, and also in the FAQ?
I didn’t create an account because it seemed to be saying I need to put in money first before I could try anything. Other users might have similar interpretations.
If you haven’t seen it, take a look at artbreeder which by the way has a free level.
Right now you can't change templates, you can only add info to generate the image you want, but in the future I want to add the ability for everyone to generate templates.
EDIT: I've added the free tier to the pricing page and a bit of an explanation to the templates page, thanks for the feedback!
One bug though: After I wrote my prompt I discovered that there are some 'Advance options', clicked on the button, more advanced options unraveled, but I lost my prompt.
In another comment you point out that this service expands into a much larger prompt and then here you wonder why someone would pay for that. Meanwhile I think it's obvious money on the table when I have to write a paragraph when I really just want permutations of ork portraits for the next Shadowrun game or whatever.
Having to become an experienced prompt engineer to use these tools productively is serious friction.
I'd like to see tools/services made for professionals and I think this is the direction it would need to go.
Or even better running stable diffusion locally?
Huggingface can't effectively build a consumer product. They're building for B2B to be a github for models.
Huggingface explicitly rejected building B2C tools.
That huggingface UI will forever remain unfriendly and unable to scale.
B2C / creator tools are going to start disrupting businesses and B2B. I'm actually extremely bullish on B2C/tools now more than ever.
I like this approach, it's far more intuitive.
I love generating generic cartoons from input. Very useful when creating icons / avatars!
https://phantasmagoria.stavros.io/images/i3VCADPjZ9RY6K7q/
Some of these images I want to hang on my wall:
However, it's AI, so you don't always get what you expect.
Can you generate something that is recognisably that, but inserts a random person, say batman or Jesus into that scene in place of spiderman so that people will recognize the reference?
Image generation takes 5 sec, but loading the model takes 120 sec. There's some time you can keep the model "alive" for, to wait for another image generation request, but if I have too few users, it means everyone gets image generation times of 120 sec and I pay for 120 sec per image, which is a lot.
With enough users, though, I can always keep the model up, and everyone gets 5 sec generation times and the economics for this make sense. Hopefully if this picks up steam I'll be able to rent a dedicated server to generate things on, but it's just a sideproject for now.
At least I hope it will be useful/fun to some people.
EDIT: Added, good shout, thanks!
A few days ago I was generating images using my home desktop, and had a Web push notification notifying people when it was on. That was much cheaper, but very bad UX, as it could be hours before you got your image (and you were long gone by then).
However, I can see why you might choose not to bother with that and find another way to keep the model alive without the try prompt.
Edit: didn't you mention it takes 5 sec if model is alive?
Also, Inferrd doesn't really look cheaper, a 16GB GPU is $600/mo, which is about the same as Banana, except Banana doesn't bill for the time the model isn't working.
For my needs, with the few images I need to generate, the difference is between paying $50/mo (admittedly, with worse UX for my customers) and paying $600/mo for a card that's mostly sitting idle.