Download
1 variant available
License:
AnimaThe Anima Model is licensed by CircleStone Labs LLC. Copyright CircleStone Labs LLC. IN NO EVENT SHALL CIRCLESTONE LABS LLC BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.
Built on NVIDIA Cosmos
Kuchiki Rukia - Bleach
Only trained on Season 1 with https://github.com/negaga53/neme-anima on Anima Base v1.
Only trained on images without post processing (bright/dark scenes, flashbacks, etc.)
=== Triggers ===
Character: rkia, official style, kuchiki rukia,
Hair/Eyes: black hair, hair between eyes, purple eyes
Outfit 1: black kimono, black hakama, sandals, // katana
Outfit 2: blazer, school uniform, skirt, kneehighs, bow, ribbon
Outift 3: white shirt, school uniform, skirt, kneehighs, bow, ribbon
Extra: pajamas
Preview images have been upcsaled with x2 Hires Fix and (partially) with Adetailer (Face and Hand).
Use strength 0.8 - 1.0
This was one of the things I REALLY missed from Illustrious and I finally beat my laziness and trained it myself. It's not perfect by no means but it gets the job done pretty well.
If you think you can create a better lora, PLEASE do it! ðŸ˜
I attached the training data.
=== Technical details ===
Training parameters from run.toml and dataset.toml:
epochs = 60
micro_batch_size_per_gpu = 1
pipeline_stages = 1
gradient_accumulation_steps = 2
gradient_clipping = 1.0
warmup_steps = 100
eval_every_n_epochs = 5
eval_before_first_step = true
eval_micro_batch_size_per_gpu = 1
eval_gradient_accumulation_steps = 1
save_every_n_epochs = 10
activation_checkpointing = true
partition_method = "parameters"
save_dtype = "bfloat16"
caching_batch_size = 1
map_num_proc = 8
steps_per_print = 1
compile = true
[model]
type = "anima"
transformer_path = "/home/main/.cache/neme-anima/models/anima-base-v1.0.safetensors"
vae_path = "/home/main/.cache/neme-anima/models/qwen_image_vae.safetensors"
llm_path = "/home/main/.cache/neme-anima/models/qwen_3_06b_base.safetensors"
dtype = "bfloat16"
llm_adapter_lr = 0.0
sigmoid_scale = 1.3
[adapter]
type = "lora"
rank = 32
dtype = "bfloat16"
[optimizer]
type = "adamw_optimi"
lr = 5e-05
betas = [0.9, 0.99]
weight_decay = 0.01
eps = 1e-08
resolutions = [512, 1024, 1536]
enable_ar_bucket = true
min_ar = 0.5
max_ar = 2.0
num_ar_buckets = 9
num_repeats = 1.0