YesNoOk
avatar

Using AI to aid in Mugen/Ikemen creations (Read 73562 times)

Started by PotS, November 27, 2023, 06:10:43 pm
Share this topic:
Using AI to aid in Mugen/Ikemen creations
#1  November 27, 2023, 06:10:43 pm
  • ******
    • Portugal
    • network.mugenguild.com/pots/
There's probably enough to talk about here to have a thread for it.

To start with you can use AI to generate stages.

Some tools:

Bing Image Creator
Stable Diffusion
Playground AI

The main things to keep in mind when having them generate fighting game stages are probably the floor, which should be flat, and the perspective. The horizon line should be a little above the floor. Often times just asking them to generate a fighting game stage will make them respect those rules, but other times you have to add prompts like "flat floor" etc. To make the results more interesting you can make the prompts more descriptive. The more adjectives, the better.

Currently they only output a single layer. If you want to separate them into several layers you'd probably make your life easier by prompting them to generate in certain art styles that use outlines.

Many examples:

Stages created by myself (with a few colaborations)
Stages created by Kamui_De_Los_Vientos
Stages created by MatreroG
You can help with Ikemen GO's development by trying out the latest development build and reporting any bugs on GitHub.
My Mugen and Ikemen content can also be found here.
Last Edit: December 07, 2023, 12:21:32 pm by PotS
Re: Using AI to generate stages for Mugen/Ikemen
#2  November 27, 2023, 08:15:13 pm
  • **
  • "That's the spirit. Prepare yourself!"
    • Brazil
It's not easy create floor nice, aligned and clean. Continuing testing with multiple prompts and using images to inspire A.I. to work better  :whip:

Spoiler, click to toggle visibilty
Now Loading...
Re: Using AI to generate stages for Mugen/Ikemen
#3  November 28, 2023, 02:15:59 am
  • **
My Create ai art generator

Tool by img2go.com

Here My Stages












https://discord.com/invite/vrcS6k7yWz

I Hate R-Bismut NSFW Artist
Re: Using AI to generate stages for Mugen/Ikemen
#4  December 03, 2023, 04:53:01 am
  • **
    • Vietnam
Sorry, when i post AI portrait here.

From this

AI Result


and


i use https://playground.com

My promt is
Adon street fighter fullbody punch with transparent background

Expand is

Adon from Street Fighter in mid-motion delivering a powerful full-body punch, stance wide and intense expression on face, rendered in ultra-clear digital art with transparent background, dynamic pose capturing movement and muscles tensed, illuminated by dramatic lighting, octane rendering.
Re: Using AI to generate stages for Mugen/Ikemen
#5  December 03, 2023, 09:02:54 am
  • **
Sorry, when i post AI portrait here.

From this

AI Result


and


i use https://playground.com

My promt is
Adon street fighter fullbody punch with transparent background

Expand is

Adon from Street Fighter in mid-motion delivering a powerful full-body punch, stance wide and intense expression on face, rendered in ultra-clear digital art with transparent background, dynamic pose capturing movement and muscles tensed, illuminated by dramatic lighting, octane rendering.

Wrong.
Bro your not is portrait ai,
Because stage ai.
https://discord.com/invite/vrcS6k7yWz

I Hate R-Bismut NSFW Artist
Re: Using AI to generate stages for Mugen/Ikemen
#6  December 03, 2023, 01:32:37 pm
  • ******
If he said he used AI to convert the original picture into a different one and gives proof of it both with the prompt and the site he used, maybe you shouldn't question it and say "no it's not", I don't believe you know what he's done better than him.
(:ssh: I suggest you made your posts more readable and understandable)

But I can say the portrait doesn't fit this topic because it's about stages and not portraits. (if that's what you meant, then you should've worded it better)

Unless you use the portrait as a prop, a poster or something in an arena.
Last Edit: December 03, 2023, 01:36:10 pm by AlexSin
Re: Using AI to generate stages for Mugen/Ikemen
#7  December 03, 2023, 07:24:08 pm
  • **
  • "That's the spirit. Prepare yourself!"
    • Brazil
Playground AI used an image from Igniz Stage (KOF 2002 UM).



The Result:












Now Loading...
Re: Using AI to generate stages for Mugen/Ikemen
#8  December 04, 2023, 02:25:22 pm
  • ******
  • Cosplayer by day, MUGEN creator by night
  • The City of the Wolves is coming
    • Chile
    • network.mugenguild.com/basara/

  • Online
Even when it's a bit obvious for some of you and even for me, I never thought that stages could be done by using AI, I found it interesting and maybe I could make some too in a close future.

But by now, I'm using AI in MUGEN but for resize sprites for a secret project I'm working on, maybe I can talk about that later. By now, I want to see more created stages of yours here :D
Re: Using AI to generate stages for Mugen/Ikemen
#9  December 04, 2023, 04:06:03 pm
  • ***
  • Wandering....
Hi guys is it also possible for just making the stage higher definition and to un-pixel it like vector style without altering and hanging objects on the stage?

Can it be also use anime backgrounds to turn into FG stages and making it like pixel art?


Thanks guys
Re: Using AI to generate stages for Mugen/Ikemen
#10  December 04, 2023, 08:44:55 pm
  • *****
  • I must continue my training...
Got some heat up in here. Imma join the fray soon enough. Stay tuned.

Do your best to become stronger. Become stronger, so you can do your best.

Mugen is a way for me to remain nostalgic with all of the dream matches I cooked up as a kid. It's also what I call a "Digital Action Figure Collection."
MaxBeta link=topic=155656.msg1876950#msg1876950 date=13867
Vegaz_Parrelli...also known as The Lord of the Rings.
[/quote said:
Re: Using AI to generate stages for Mugen/Ikemen
#11  December 05, 2023, 07:42:15 pm
  • *
    • USA
What if there was an all-in-one AI tool that could not just generate art for stages but created the stage itself? With the .def and .sff files and all the code too
Re: Using AI to generate stages for Mugen/Ikemen
#12  December 05, 2023, 09:30:11 pm
  • *****
  • I must continue my training...
What if there was an all-in-one AI tool that could not just generate art for stages but created the stage itself? With the .def and .sff files and all the code too

So like have the A.I. Image Creator and Cybaster's Stage Tool do the fusion dance  :yuno:

Do your best to become stronger. Become stronger, so you can do your best.

Mugen is a way for me to remain nostalgic with all of the dream matches I cooked up as a kid. It's also what I call a "Digital Action Figure Collection."
MaxBeta link=topic=155656.msg1876950#msg1876950 date=13867
Vegaz_Parrelli...also known as The Lord of the Rings.
[/quote said:
Re: Using AI to generate stages for Mugen/Ikemen
#13  December 05, 2023, 10:00:13 pm
  • **
  • "That's the spirit. Prepare yourself!"
    • Brazil
Cybaster's Stage Tool is good  :thumbsup: Easy to use  =)



Now Loading...
Re: Using AI to aid in Mugen/Ikemen creations
#14  December 07, 2023, 12:25:20 pm
  • ******
    • Portugal
    • network.mugenguild.com/pots/
But I can say the portrait doesn't fit this topic because it's about stages and not portraits. (if that's what you meant, then you should've worded it better)
You're right so I updated the thread title to include all sorts of AI usage.

I just did some stage upscaling here. Used Pixelcut.

Another thing I did recently was use ChatGPT to add win quote variety to some characters. You tell it what the character is like (sometimes it even already knows the character), maybe supply some existing win quotes for inspiration then you just pick the better ones it creates.
You can help with Ikemen GO's development by trying out the latest development build and reporting any bugs on GitHub.
My Mugen and Ikemen content can also be found here.
Re: Using AI to aid in Mugen/Ikemen creations
#15  December 07, 2023, 05:22:53 pm
  • ******
  • Cosplayer by day, MUGEN creator by night
  • The City of the Wolves is coming
    • Chile
    • network.mugenguild.com/basara/

  • Online
I found time ago a "swiss army knife" AI website that will help you with a lot of things, not just MUGEN, and maybe some of you have heard about it, it's TinyWow and it's very useful and free. For MUGEN I used the Image section for upscale sprites (for the secret project I mentioned before) and in stages I used the "removed objects for photo" for some ClayFighter stages, the ones from 63 1/3 I got shoots and for some of CF3 prototype screenshots. Even I got a couple unreleased because... some C&D, you know, but that would be in my game and eventually I'd release for normal MUGEN... maybe I can release them anyway for X-mas as a gift, dunno

N.

Re: Using AI to aid in Mugen/Ikemen creations
New #16  December 11, 2023, 08:33:10 am
  • avatar
  • ***
Spoiler: An attempt was made. (click to see content)

I have mixed feelings (there's probably a better way to do what I did).  The ones past the ones with the castle in it aren't what I had in mind, but I liked how they looked.
These started out as 1024x1024 images in Bing (seemed more efficient because of getting up to 4 images per generation).  I cropped the watermark and ran those through a Stable Diffusion extension I saw suggested for making pixel art to downscale the image with a pixel size of 3, and then applied a 256 color fast octree palette using another extension (I didn't like how it looked when I used it to pixelate the image, but I run out of memory if I'm not careful how I mess with larger images.  Also, after clicking enable (because for some reason I thought I could use the palette part independently) I had to set the pixel size to 1 and have rescale to original size checked.

*One of the comments to a checkpoint on civitai suggested doing something along those lines for a more accurate pixel art look (shrink to 1/8 size, then upscale?).  It probably doesn't apply here.


Extensions for Stable Diffusion (you use these in the extras section):
https://github.com/AUTOMATIC1111/stable-diffusion-webui-pixelization (does pixelization)
https://github.com/mrreplicart/sd-webui-pixelart (does pixelization, reducing colors, and applying palettes to images)
https://github.com/AUTOMATIC1111/stable-diffusion-webui-rembg (I didn't use this, but it's supposed to remove backgrounds)
To use (I should preface this by mentioning that I'm using the Automatic1111 UI, so you may have to do different things if you're not using it):
- Download and put the folder into stable-diffusion-webui\extensions
or
- Click on Extensions, then Install From URL in the UI. Paste the github link in there, click the button, and then restart the UI when you're done.

LoRas for Stable Diffusion:
https://civitai.com/models/152875/capcom-vs-snk-art-style
https://civitai.com/models/91059/marvel-vs-capcom-2-artstyle
https://civitai.com/models/44278/street-fighter-ii-the-animated-movie-art-style
https://civitai.com/models/140896/shinkiro-snkkof-street-fighter-artist-style
https://civitai.com/models/160296/kinu-nishimura-street-fighterkof-artist-style
https://civitai.com/models/189011/capcom-cps2-sprite-style
https://civitai.com/models/243388/snk-neo-geo-sprite-style
I haven't tested any of these, but you use them to force a specific artstyle or to change an image to that artstyle through img2img (IIRC, you have to do it gradually through img2img by setting a low denoise strength, and running what it generates through img2img, and repeating this until you either like it enough to use or scrap it and start over)
To use:
- Put them in stable-diffusion-webui\models\Lora.  If you don't have that folder, create it.
- Make note of the trigger word (ex: Capcom SNK Style) if it has one, because you may need to use it (it could be less accurate if you don't use it).  If it doesn't have one, but you want to improve the accuracy?  Open the safetensors file with a text editor, and you should see some words/descriptions that were used in the training process.  Copy those and try using them as prompts.
- Click on the 3rd button below the Generate button in the Stable Diffusion UI (if you're using Automatic1111.  I don't know for the other ones) to open up additional networks.  Click Lora, and then click the one you want to add.  Some text will appear in the prompt window.  Something like <lora:EMS-51611-EMS:1>, for example.  You could also copy what it says so you don't have to do that, I guess.
- If you see in the lora description mentioning something about setting the weight to a certain value, what it means is to mess with the prompts by doing something like this: If it suggests 0.8 for the weight, Capcom SNK Style becomes Capcom SNK Style:0.8 and <lora:EMS-51611-EMS:1> becomes <lora:EMS-51611-EMS:0.8>.  You're adjusting how much these will affect the image (sometimes you need to use weaker weights because of overtraining).

Spoiler: Odds and ends (click to see content)

And for the hell of it, Ryu making goofy faces.  (IIRC, Ryu from Street Fighter making a trollface was the prompt.):
Spoiler, click to toggle visibilty


I'm not going to be making anything with them.  You can use the images for whatever if you think they're worth using. 
1 Last Updated: 4/19/23
2 Last Updated: 1/16/21
3 Last Updated: 04/20/23
4 Last Updated: 7/16/17
Last Edit: December 25, 2023, 06:15:28 pm by N.