• There is NO official Otland's Discord server and NO official Otland's server list. The Otland's Staff does not manage any Discord server or server list. Moderators or administrator of any Discord server or server lists have NO connection to the Otland's Staff. Do not get scammed!

Using Bing AI DALL-E to create Tibia like items

MagicWall

Veteran OT User
Joined
Oct 12, 2008
Messages
124
Solutions
6
Reaction score
344
Hi, today I tested new Bing AI functionality for generated images focusing on its possible use for creating Tibia sprites, effects:

All_items.png
If anyone wants to try:
1) Open Microsoft Edge browser.
2) Go to site bing.com
3) Login in bing to Microsoft account.
4) Open Chat
5) Then write in chat something like this:
Create image item poison axe with Tibia style
6) Wait for the response (image) and then write:
Can you make it more tibia pixel art style?
Like in images below (examples):
BingAI_Egzo_Spellbook.pngBingAI_Hussars_Axe.pngBingAI_Poison_Axe.pngBingAI_Winged_Hussars_Axe.png
 
Very cool, are those images 32x32 pixels in size?
 
This is so cool. I've been waiting to see how much progress would be made with AI and generating tibia content. This makes me want to create an OT just so I can add unlimited amounts of custom sprites lol.
 
U Could easily remake those in lower resolution looks dope detail in the handles omg
 
Trying to create a full set with a similar style (Bing AI DALL-E):

Query:
Create image item <style> <item type> game asset with Tibia game pixel art style
Example query:
Create image item dark magician purple hat game asset with Tibia game pixel art style

Result with compare (96x96/64x64/32x32):
1684957061546.png

Examples:
1684957100987.png
 
Yeah, definitely doesn't fit (downscaling is just too visible). Would be cool for any other pixel art game tho.
 
Last edited:
@matiszek30
Can you share the queries you use to generate these images?
Can leonardo.ai generate 32x32px images?
 
@matiszek30
Can you share the queries you use to generate these images?
Can leonardo.ai generate 32x32px images?
No it doesn't support 32x32 you new to downscaled the images to that size.

Leonardo by itself won't generate it out of the box. You need to train your own model, for free you can only train one but if you pay you can do more. It's quiet easy to train as you just provide pictures and wait till training is done. I have done it on pictures with size 360x360 but make sure that the pictures of armor you provide have proper naming like "plate armor" and you need to set up your instance prompt which you should sort of use it in your final prompt. For example, you set up instance prompt as "a tibia armor" and in your train data you added a picture of a golden armor and dragon scale armor but you want it to be mixed then you use prompt like 'a tibia golden scale armor" and you generate pictures. I would say sometimes you don't have a successful generation and you end with some picture of dragon for example so it's a bit of hit and miss. After I generate the images I use creative upscale provided by Leonardo AI for best results as images can be blurry, then I remove the background and downscale them to 32x32. As the last step I use a pixelator program to work on the saturation to get final result.

You can remove background in Leonardo but to not waste a tokens I use Login (https://express.adobe.com/sp/tools/remove-background).

For downscaling I use a website 32x32 pixel art converter - safeimagekit (https://safeimagekit.com/32x32-pixel-art-converter).

Here is the link for pixelator: Pixelator App (http://pixelatorapp.com) - for commercial use you need a licence.

You can try on cipsoft data if you want just type in google: tibia fandom armor and you should get some pics in size of 360x360 on google images but I recommend to use your own ones or some free ones. For upscaling you can try and use Nero AI explores more professional AI solutions. Innovation, efficiency, endless potential. (https://ai.nero.com/enhance) it can give quiet good results but could be too smooth.

This is my set up for generating pictures

obraz_2023-06-02_000018491.png

This is are generated pictures after creative upscale:
nG0zU7u.png


This is the final result:
pixel art armor.png
 
Last edited:
No it doesn't support 32x32 you new to downscaled the images to that size.

Leonardo by itself won't generate it out of the box. You need to train your own model, for free you can only train one but if you pay you can do more. It's quiet easy to train as you just provide pictures and wait till training is done. I have done it on pictures with size 360x360 but make sure that the pictures of armor you provide have proper naming like "plate armor" and you need to set up your instance prompt which you should sort of use it in your final prompt. For example, you set up instance prompt as "a tibia armor" and in your train data you added a picture of a golden armor and dragon scale armor but you want it to be mixed then you use prompt like 'a tibia golden scale armor" and you generate pictures. I would say sometimes you don't have a successful generation and you end with some picture of dragon for example so it's a bit of hit and miss. After I generate the images I use creative upscale provided by Leonardo AI for best results as images can be blurry, then I remove the background and downscale them to 32x32. As the last step I use a pixelator program to work on the saturation to get final result.

You can remove background in Leonardo but to not waste a tokens I use Login (https://express.adobe.com/sp/tools/remove-background).

For downscaling I use a website 32x32 pixel art converter - safeimagekit (https://safeimagekit.com/32x32-pixel-art-converter).

Here is the link for pixelator: Pixelator App (http://pixelatorapp.com) - for commercial use you need a licence.

You can try on cipsoft data if you want just type in google: tibia fandom armor and you should get some pics in size of 360x360 on google images but I recommend to use your own ones or some free ones. For upscaling you can try and use Nero AI explores more professional AI solutions. Innovation, efficiency, endless potential. (https://ai.nero.com/enhance) it can give quiet good results but could be too smooth.

This is my set up for generating pictures

View attachment 75910

This is are generated pictures after creative upscale:
nG0zU7u.png


This is the final result:
View attachment 75907

That's a very detailed guide, thanks for sharing. Wondering if we could train our own models in Stable Diffusion to achieve similar outputs to leonardo.ai (that's what they use under the hood, don't they?)
 
Back
Top