The Orion's Arm Universe Project Forums





Free gemerative AI for images
#1
Exclamation 
I frustrated for a while about where to put this. If there's a more appropriate place, please feel free to move this post.

I've been using three different providers of free AI image generation. Since I'm a lousy manual artist, and 3D modeling can take some effort, they've greatly helped restart my interest in writing SF, including for OA. They all work from "Prompts" -- textual descriptions of what you want to see. Their interpretations of those prompts can sometimes be, well, surprising. Despite all the publicity, generative AI text processing is still rudimentary. Also, they all have restrictions on the types of images that they're willing to generate. You can think of them as being limited to a G rating, although that's an over-simplification.

1. Microsoft's implementation of DALL-E 3 at https://www.bing.com/images/create

Bing/Create requires a Microsoft account. It currently allows up to 15 prompts per day and usually, but not always, generates 4 images for each prompt (sometimes fewer). It seems to be the least restrictive of these three. You can use Bing/Chat as an interface to it if you want.

2. Meta's Imagine, based on Emu, at https://imagine.meta.com/

Meta/Imagine requires a Meta or Facebook account. It currently has no documented limit as to the number of prompts you can use. Like Bing/Create, it creates 4 images per prompt. It's a little faster, but that might be because it just became available a few days ago, so far fewer people are using it. I think it's limited to the US right now, too.

3. NightCafe's implementation of Stable Diffusion XL 1.0 at https://creator.nightcafe.studio/studio

NightCafe has its own user account system. Users accumulate 5 credits/day, sort of the equivalent of 5 images. Some types of operations cost more credits. You can purchase a subscription to get many more credits. Their implementation seems to have the most limitations on what images can be generated. If you have adequate hardware, you can run your own copy of Stable Diffusion on your own computer, but the necessary hardware is quite expensive at the moment.

Apparently there are other free image generation resources, but these are the only ones I've tried.

So far, I've primarily been using Bing/Create, although it looks like I might be using Meta/Imagine too in the future.

Except for the images of the virtual book itself, I used Bing/Create to make the illustrations for https://www.orionsarm.com/fm_store/Halon...is_v2.html

I'm looking forward to trying Google's image generator, but it's still in "closed beta." I.e. you have to apply for an invitation. I applied but never got a response.
Selden
Reply
#2
I've tried using Night Cafe, but mostly the images look too spooky for my taste. Midjourney AI seems to have a similar problem.

Certainly there are many scary and disturbing sides to the OA scenario, but I don't think it should be all dark.
Reply
#3
A lot depends on the prompt you provide. And what you're looking for, of course. I often have to look at more than a dozen images to find one that's close enough to what I want. So far, I've found that DALL-E 3's aesthetics seem to be closest to my own. It also tends toward the colorful. Although I always include "photorealistic" in the prompt, they're not quite. I suspect that's because there aren't actually any photographs of the kinds of things I prompt for.

Here's one I managed to generate for Halloween using Bing/Create:


Attached Files Thumbnail(s)
   
Selden
Reply
#4
This image, Alb greeting Blue from Halonic Symbiosis, is very clear and accomplished. I assume you made this with Bing?
   
Reply
#5
Yes, it is.

Here's the prompt I used:

photorealistic
a red-haired man wearihg green shorts
kneeling in front of a white tiger with blue stripes
scratching the tiger behind its ears
in a grassy field near a large oak tree
Selden
Reply
#6
fwiw, here are the ones I rejected (reduced in size): the tiger was too big and its stripes weren't blue. Every once in a while an acceptable image shows up among the first four. This seems to have been one of those times.

Since then I've learned how to recolor the tiger's stripes manually, in prep for Part 2 of the story. For whatever reason bing/create doesn't like to color the tiger's stripes.


Attached Files Thumbnail(s)
           
Selden
Reply
#7
There are lots of providers. Is currently on the thousands so is difficult to keep track on so many websites using Stable Diffusion, Stable Diffusion XL, Dall-e 3 and the rest of models.

Because there are thousands of webs, different people coded or wrote webs for listing the list of AI providers.

This was the attempt of a friend of mine nicknamed Jakeukalane:
https://tuscriaturas.miraheze.org/wiki/A...o_a_imagen - Title: Anexo:Lista_de_generadores_de_texto_a_imagen
Jakeukalane wrote this list by hand.

Other people uses an LLM to power their list of AI models. For example:
There's An AI For That - https://theresanaiforthat.com/ (it contains more than 10,000 websites of recent creation that use AI and LLMs).

You can search on that web by task: https://theresanaiforthat.com/tasks/#switch

My favorite websites for Generative art are:

We are not limited to use Bing or Dall-e 3.
-----Signature-----------------------------------------------------
Quote: "Nature considers all variables".
Reply
#8
Can you provide a brief review of the servers you've used? e.g. what AI do they use? How fast are they? How many free images per day? What special options do they provide?

Of the three I've used, they all generate only static images, no video. They all produce images in about the same amount of time. I'd rank them in this order of the quality of the images they generate:
1. Bing/create uses DALL-E 3 and provides up to 15 free prompts/day. It generates 1024x1024 pixel images from textual prompts only.
2. Meta/Imagine uses their own Emu and doesn't seem to have a limit to the number of free prompts. It generates 1280x1280 images from textual prompts only.
3. NightCafe primarily uses SDXL 1.0. It provides 5 additional free credits (images) each day. It generates 1024x1024 images from both textual and image prompts.
Selden
Reply
#9
i've generated video with nightcafe- and have used pretty much every tool there- i just had to save credits every day for a few months until i'd built up enough to make video worth it.
and them, if the image is worth making larger, i'll expand it to 2000px or whatever.
Reply
#10
https://playgroundai.com/ has 500 free images per day. Other sources say 1000 free images but i'm quoting this:
https://playgroundai.com/pricing
  • Create 500 images per day
  • Use images commercially
  • Wait times during peak hours
  • Fixed image dimensions
  • Waiting period after 500 images
  • ...
I think this means that once you reached 500 images a day, you can continue making images that day but on another queue with different waiting times.
The usual waiting times of playground are fast. it ranges from 5-10 seconds to 30 or 60 seconds per prompt. it also depends on the dimensions of your image, the model you are using and other characteristics.

There are are overviews and tutorials on how to use:
https://www.youtube.com/watch?v=KmYvQV7sFs0
Video title: Playground AI Tutorial - Make AI Art for FREE!
Channel name: Wade McMaster - Creator Impact
Upload date: 10 mar 2023

But the developers of PlaygroundAI has developed a lot of features since then. For example, you can create a blank canvas and modify the images like in an graphic design program or upload your images, inpaint the image with AI and combine several images.

https://www.youtube.com/watch?v=7GT4zCjib80
Video title: Playground Introduction To Canvas
Channel name: Playground AI
Upload date: 5 ago 2023

This video is part of a series of tutorials on the features of playground canvas:
https://www.youtube.com/playlist?list=PL...IBqwyB3HA3

There are also many other webs that provides hundreds of credits each day and some other who are free and only require a digital line or queue.
There are Telegram bot accounts that let you make AI images just prompting a message to that account and return you the images on your telegram chat.
-----Signature-----------------------------------------------------
Quote: "Nature considers all variables".
Reply


Forum Jump:


Users browsing this thread: 1 Guest(s)