Fire Emblem Sprite Generator
If You Like This Model, Give It a ❤️
The LoRA model is trained on sprites from Fire Emblem (and some from Advance Wars) sized at 64x64 pixels, which can be upscaled to 512x512 while utilizing BLIP image captions.
Quick Tips
No Trigger Word Required: This model operates without the need for a trigger word.
Recommended Resolution: We suggest setting your image resolution to 512x512 and then downscaling it 8 times to 64x64 for optimal results. Any image editing programs will work, but we recommend using plug-in for post-processing. (plug in : https://github.com/mrreplicart/sd-webui-pixelart)
Class Tags: This model is trained with character class tags, such as
archer,assassin,cavalier,great knight,paladin, andDark Druid.
Enjoy your sprite!
Description
Let forget about V1.0 and stat at v2.0 here.
FAQ
Comments (8)
Awesome lora!! What models work best for this?
Amazing! I'd like to train a similar LoRA for Final Fantasy Tactics Advance, could you share how you did it for this one?
I use ChatGPT to write code to download sprite img from internet and process them into the format that I want. After that I use Kohya_ss to train LoRA.
Anyone have ANY luck with openpose/controlnet? I just cannot get it to behave with this one, which is too bad, because it's 1000% perfect for sprite sheets, but generating poses closely doesn't seem that obvious
I have a test on openopse-controlnet but on a different sprite model: https://civitai.com/images/3848860
It's not good but work at some level. For this model, the dataset is mostly focus on the standard pose but already include other moving pose. (The ratio of standard and moving pose on training is about 50:50 but the moving pose contain a lot more variation so it is more focus on standard ).
For creating sprite sheets, I think you can try img2img with other existing sprite sheets or draw a draft by yourself. However, I believe the real problem would be on how to make consistent character.
these look awesome i love the gba FE sprites, definitely gonna use it more
Any chance of this getting an SDXL/Pony update?
Creating a sprite pixel art model like this takes a considerable amount of time. And consider the small resolution (64x64), using SDXL might be overkill. However, if there's significant demand from users and I can allocate the necessary time, I may consider updating the project in the future. For now, there's no definite timeline for such an update.