bleach-jojo-multistyle-noobai-xl-vpred-lora

Bleach and JoJo Multistyle NoobAI XL V-Pred LoRA

This SDXL LoRA provides three separately prompted manga registers on NoobAI XL V-Pred. It is provided only for personal, research, and educational use.

Artifact

• Repository: sleephashira/bleach-jojo-multistyle-noobai-xl-vpred-lora
• File: multi3_lora.safetensors
• Size: 456519428 bytes
• SHA-256: bdedd00bcbd33deff9a4159d7db977a5def6cfb9b2bad11887d1e27b0aa1fba5
• Base model: Laxhar/noobai-XL-Vpred-1.0
• Triggers: blcmanga, blccolor, jjbamanga

Training Summary

The training set contained 1,800 cleaned, extracted, copyrighted panels:

• 600 monochrome Bleach panels for blcmanga
• 600 official-color Bleach panels for blccolor
• 600 JoJo panels for jjbamanga

Checkpoint step 3600 was selected for this release.

Tested Usage

NoobAI XL V-Pred requires Euler, prediction_type="v_prediction", rescale_betas_zero_snr=True, guidance_rescale=0.7, and CFG 4-5. Other samplers or epsilon-prediction settings do not provide the intended inference behavior. Start with one trigger; prompts may combine triggers experimentally. Use monochrome, greyscale with blcmanga or jjbamanga, and explicit color tags with blccolor. A tested starting point is 28 steps, CFG 5, and LoRA scale 0.75 at approximately 1024x1024 total pixel area.

from diffusers import EulerDiscreteScheduler

pipe.scheduler = EulerDiscreteScheduler.from_config(
    pipe.scheduler.config,
    prediction_type="v_prediction",
    rescale_betas_zero_snr=True,
)
pipe.load_lora_weights(
    "sleephashira/bleach-jojo-multistyle-noobai-xl-vpred-lora",
    weight_name="multi3_lora.safetensors",
)
pipe.fuse_lora(lora_scale=0.75)

image = pipe(
    "blcmanga, monochrome, greyscale, 1boy, sword, rain",
    width=832,
    height=1216,
    num_inference_steps=28,
    guidance_scale=5.0,
    guidance_rescale=0.7,
).images[0]

Results vary by prompt, seed, resolution, and adapter strength. The model can reproduce biases, visual motifs, or character associations from its training set and may produce malformed anatomy, text, borders, or panel details.

Terms, Rights, and Disclaimer

This adapter was trained on extracted Bleach and JoJo panels. The source artwork is copyrighted and was not licensed by its copyright holders for this training or release. This is an unofficial release with no affiliation with or endorsement by the authors, artists, publishers, rights holders, or upstream model authors.

Permitted scope: personal, research, and educational use only.

Prohibited scope: commercial or monetized use, advertising, paid services, resale, and any other revenue-generating use are prohibited.

license: other identifies non-standard model terms. These terms are a custom use restriction, not an OSI-approved open-source license. No rights are granted to any source artwork, characters, names, trademarks, datasets, or other third-party material. Downloading the weights does not supply permission from any copyright, trademark, publicity-right, or other rights holder.

The downloader and user are solely responsible for determining legality, obtaining all necessary permissions, complying with local law, and following all applicable platform and upstream terms. The software and artifact are provided as-is, without warranties of any kind. To the maximum extent permitted by law, the maintainer excludes liability for use, outputs, claims, damages, or other consequences arising from the artifact.

This disclaimer and the custom restriction do not override applicable law or upstream licenses. All applicable upstream base-model terms also govern. Review the NoobAI XL V-Pred 1.0 model card and terms, the upstream Illustrious XL model card, its terms of use, and the linked Fair AI Public License 1.0-SD before downloading or using this adapter.