ltxvideo-disney

This is a LyCORIS adapter derived from Lightricks/LTX-Video.

The main validation prompt used during training was:

A black and white disney scene in the style of Steamboat Willie

Validation settings

• CFG: 3.8
• CFG Rescale: 0.0
• Steps: 25
• Sampler: FlowMatchEulerDiscreteScheduler
• Seed: 42
• Resolution: 768x512

Note: The validation settings are not necessarily the same as the training settings.

You can find some example images in the following gallery:

Prompt

unconditional (blank prompt)

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

At midnight in a neon-lit cityscape, an agile anime protagonist leaps between rooftops, evading shadowy figures with fluid, acrobatic action in a striking black and white animated style.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

At dawn in a rustic village kitchen, a dedicated sous-chef passionately chops and stirs ingredients, crafting a traditional dish with precision and care in a refined black and white animated scene.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

In the soft morning light of a quiet garden, a reflective individual gazes into the distance, their subtle expressions and gentle actions captured in a delicate black and white animated portrait.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

Under the glow of a full moon at an abandoned urban theater, a solitary figure delivers a dramatic monologue, their every movement and emotion rendered in cinematic black and white detail.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

At twilight in a sophisticated art gallery, a poised subject glides through elegant corridors, each graceful step and refined gesture illuminated in a timeless black and white animated scene.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

In the early morning mist on a bustling city street, a daring adventurer sprints through narrow alleys, navigating obstacles with dynamic energy in a high-octane black and white animated sequence.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

During a foggy night on a deserted pier, an enigmatic figure silently watches the dark waters, their slow, deliberate actions and shadowy presence unfolding in a mysterious black and white animated narrative.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

At the break of dawn in a historic train station, a character in period attire rushes across the platform, their hurried actions and nostalgic surroundings brought to life in classic black and white animation.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

In the fading light of an urban alley adorned with abstract murals, a creative soul sketches fleeting impressions, their experimental actions and imaginative process captured in an artistic black and white animated style.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

At twilight in a bustling high-tech metropolis, a futuristic agent navigates a maze of holographic displays and sleek architecture, their rapid movements and cutting-edge surroundings depicted in a bold black and white animated scene.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

At dusk in a serene lakeside park, a graceful woman performs a fluid dance, her elegant movements and gentle expressions perfectly timed with the shifting light in a refined black and white animated sequence.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

At noon in a busy city square, a determined man strides purposefully through the crowd, his assertive actions and commanding presence captured in a dynamic black and white animated scene.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

In the afternoon at a lively playground, a spirited boy races joyfully across a field, his energetic play and mischievous antics rendered in a vibrant black and white animated style.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

At sunrise on a quaint village street, a curious girl skips along the pavement, her bright energy and animated expressions shining through in a charming black and white animated scene.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

During a cozy evening in a warmly lit living room, a family gathers around a table for a shared meal, their tender interactions and joyful connections animated in a heartfelt black and white scene.

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

Prompt

A black and white disney scene in the style of Steamboat Willie

Negative Prompt

ugly, cropped, blurry, low-quality, mediocre average

The text encoder was not trained. You may reuse the base model text encoder for inference.

Training settings

• Training epochs: 2666
• Training steps: 8000
• Learning rate: 5e-05
  • Learning rate schedule: cosine
  • Warmup steps: 400000
• Max grad value: 0.0
• Effective batch size: 24
  • Micro-batch size: 8
  • Gradient accumulation steps: 1
  • Number of GPUs: 3
• Gradient checkpointing: True
• Prediction type: flow-matching (extra parameters=['training_scheduler_timestep_spacing=trailing', 'inference_scheduler_timestep_spacing=trailing'])
• Optimizer: adamw_bf16
• Trainable parameter precision: Pure BF16
• Base model precision: int8-quanto
• Caption dropout probability: 10.0%

LyCORIS Config:

{
    "bypass_mode": true,
    "algo": "lokr",
    "multiplier": 1.0,
    "full_matrix": true,
    "linear_dim": 10000,
    "linear_alpha": 1,
    "factor": 4,
    "apply_preset": {
        "target_module": [
            "Attention",
            "FeedForward"
        ],
        "module_algo_map": {
            "FeedForward": {
                "factor": 4
            },
            "Attention": {
                "factor": 2
            }
        }
    }
}

Datasets

disney-black-and-white

• Repeats: 0
• Total number of images: ~69
• Total number of aspect buckets: 1
• Resolution: 0.2304 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

Inference

import torch
from diffusers import DiffusionPipeline
from lycoris import create_lycoris_from_weights


def download_adapter(repo_id: str):
    import os
    from huggingface_hub import hf_hub_download
    adapter_filename = "pytorch_lora_weights.safetensors"
    cache_dir = os.environ.get('HF_PATH', os.path.expanduser('~/.cache/huggingface/hub/models'))
    cleaned_adapter_path = repo_id.replace("/", "_").replace("\\", "_").replace(":", "_")
    path_to_adapter = os.path.join(cache_dir, cleaned_adapter_path)
    path_to_adapter_file = os.path.join(path_to_adapter, adapter_filename)
    os.makedirs(path_to_adapter, exist_ok=True)
    hf_hub_download(
        repo_id=repo_id, filename=adapter_filename, local_dir=path_to_adapter
    )

    return path_to_adapter_file
    
model_id = 'Lightricks/LTX-Video'
adapter_repo_id = 'bghira/ltxvideo-disney'
adapter_filename = 'pytorch_lora_weights.safetensors'
adapter_file_path = download_adapter(repo_id=adapter_repo_id)
pipeline = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.bfloat16) # loading directly in bf16
lora_scale = 1.0
wrapper, _ = create_lycoris_from_weights(lora_scale, adapter_file_path, pipeline.transformer)
wrapper.merge_to()

prompt = "A black and white disney scene in the style of Steamboat Willie"
negative_prompt = 'ugly, cropped, blurry, low-quality, mediocre average'

## Optional: quantise the model to save on vram.
## Note: The model was quantised during training, and so it is recommended to do the same during inference time.
from optimum.quanto import quantize, freeze, qint8
quantize(pipeline.transformer, weights=qint8)
freeze(pipeline.transformer)
    
pipeline.to('cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu') # the pipeline is already in its target precision level
model_output = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    num_inference_steps=25,
    generator=torch.Generator(device='cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu').manual_seed(42),
    width=768,
    height=512,
    guidance_scale=3.8,
).frames[0]

from diffusers.utils.export_utils import export_to_gif
export_to_gif(model_output, "output.gif", fps=25)

Exponential Moving Average (EMA)

SimpleTuner generates a safetensors variant of the EMA weights and a pt file.

The safetensors file is intended to be used for inference, and the pt file is for continuing finetuning.

The EMA model may provide a more well-rounded result, but typically will feel undertrained compared to the full model as it is a running decayed average of the model weights.