Sometimes AI just doesn't cut it - #Project2026 sneak peek

I wasn't going to do this but after today I couldn't help it. I have grand ambitions for 2026 and Hive. Well "GRAND" may be a stretch but for me it is grand so that is what I am calling it.
And to that end, I have now officially (Blockchain official at least) trademarked #Project2026 to better document my project which I originally wasn't going to divulge but it's too late to put the cat back in the bag.
Vague / Brief overview
#Project2026 is the brainchild of mine that I came up with last year. It involves a lot of my hobbies all together in 1 project. The project is not 'for' anyone or to get anything from it. I am creating it for me, to prove to myself that I can do this thing to the level of detail and professionalism that I hope I can do.
I am combining all of the following into 1 project:
- Use my Local AI instances
- Only using local AI LLM and GenAI instances for initial project creation and digital file modification and creation.
- Blender - 3D modelling
- I want to get better at using Blender software for my own 3D printing needs.
- 3D printing
- I have been slowly upgrading my 3D printer (better plates, nozzles, filament types, etc)
- Woodworking
- I have always loved working with Wood and creating things from Wood (thanks to my Grandfather for that).
- Blockchain/Web3
- I love the Web3 space and in 2025 I re-fell in love with the Hive blockchain so this project is focused on just Hive blockchain.
Playing around with local AI
Look, I am not a AI expert. But what I am a expert in is trying something and failing and then trying something else. Rinse and Repeat until I get the process down. To not get too much into the weeds, I run ComfyUI locally on my desktop to utilize my RTX 5070 TI. I was running a specific workflow in ComfyUI using the 'Flux 1' model. Well 'Flux 2' came out recently for local instance via ComfyUI so I am not attempting to work through my, well 'workflow' with the Flux 2 model and dependencies.
So to test it, I start out with a very simple GenAI request. Take a Lore Image from Splinterlands and ask it to remove the background. Which is simple enough to do.... RIGHT?
Starter Image: Molten Elemental

I gave it this exact prompt:
"Remove the red background and leave just the Molten Elemental character"

I don't know WTF that export image is... it's not remotely close lol
So what do I do? I just open up GIMP (open source Photoshop equivalent = FREE software) and open up the image and within 20 seconds I can remove the background manually.

20 seconds later =

But another test I like to run is to morph the character into Lego format. Why Lego format? Because BraveTofu created a post back in May 2025 and it really struck me. So much so that I bookmarked it. You can see it here.
But the main idea is to use AI to take a Splinterlands character and turn them into a LEGO figurine. Which if I paste his cover image here, you can see what I mean.

Image credit to BraveTofu: https://peakd.com/hive-13323/@bravetofu/splinterlands-lego-collection
So I changed my prompt to: "Redesign this character in a LEGO design. It should be standing on a black pedestal."
And well this is my first attempt...

2nd attempt?


Yeah this is going to take a LOT more work
Just to show I can do it...
Here is one I created back on my Flux 1 workflow and with the settings dialed in.

Sure it's not exactly what I had in mind but it's WAY better. So back to the drawing board to refine my new workflow and settings. Silly AI always making more work for me! 😀
But that is all for today. Just a fun little article about Splinterlands characters and local AI and me tanking at it today!
![]() | ![]() |
|---|


AI doens't really understand some things....
Also if you want to make a lego minifig that you can actully just bug, check this out. :) https://www.lego.com/en-us/minifigure-factory
oh that's neat, I didn't know that was even a thing
one of my Dungeons and and dragons people made minifigs of our whole party :)
okay that's super neat!
Very interesting project mate! Your AI is creating pretty images, just not the RIGHT ones for now but like you said, the way to success is just to not give up when you fail.