
A Flux.1-dev LoRA capable of creating unique new Pokemon sprites and other objects/characters in the Pokemon style.
The model is flexible enough to be able to produce images with normal backgrounds as well if described. If you want to make sure you get the "sprite style" add 'white background' to your prompt.
Training:
Dataset Size: 1000
Resolution: 512x512
Batch Size: 4
Steps: 7000
Learning Rate: 0.0001
All images use the same caption "pokemon_sprite, a sprite from a pokemon game".
All example images were generated with these settings:
Sampler: [Forge] Flux Realistic (Slow)
Schedule Type: Beta
Steps: 20
Resolution: 1024x1024
Distilled CFG: 3.5
CFG: 1
I already have hand-written new captions for all 1000 images, with tags for the Pokemon's types (type_fire, type_ice, etc.), their evolution stage (evolution_basic, evolution_1, etc.) and their form if applicable (form_mega, form_dynamax, form_galar, etc.) so a new, hopefully more customizable, v2 will follow soon.
Feel free to share your experiments below :)
描述:
训练词语: pokemon_sprite,a sprite from a pokemon game
名称: pokemon_sprite_v1.safetensors
大小 (KB): 167939
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success