Credit:
Trained on this awesome dataset (safe for work) : https://huggingface.co/datasets/codeShare/lora-training-data/blob/main/claymore_lora_training_data.zip
(Created by Adcom: https://tensor.art/u/754389913230900026 )
His Chroma Version of this LoRa: https://tensor.art/models/922660820934998215/Claymore-e52
----------------------------------------------------------------------------------------------------
cla 20: trained on synta-xl illustrious mix
cla 20 vpred trained on illustrious-xl v0.1 with zero snr enabled. Idk if this is a real v-pred lora but works pretty good with vpred models, so...
Style description: Psychodelic, old school video game graphics, claymore manga influence,
I wanted to see if SDXL is able to be trained on these, as to my knowledge training with colllages is really only smth that works fine with Flux (and Chroma now).
Trained directly on my own model Synta-XL Illustrious Mix V2. Images have mostly natural language captioning and imo NL works the best in combination with booru tags.
Not some easy to use style, but cool for experimental stuff. However if you prompt for offcial art and claymore \(series\), then this LoRa is quite useful and precise in covering this artstyle.
Depending on what you want you can add some of these in your negative prompt:
(spot color, greyscale, comic, 2koma, 3koma, multiple views, speech bubble:1.1),
(gameplay mechanics, user_interface, health bar, heads-up_display:1.1)
add more weight if needed
For postive prompts besides trigger word you can also add some of these:
video game graphics, digital art, Norihiro Yagi, claymore \(series\), claymore manga by norihiro yagi