LTX-Video

LTX-Video est un diffusion transformer conçu pour la génération rapide et en temps réel de vidéos haute résolution à partir de texte et d’images. La fonctionnalité principale de LTX-Video est le Video-VAE. Le Video-VAE présente un ratio de compression pixel vers latent plus élevé (1:192), ce qui permet un traitement plus efficace des données vidéo et une vitesse de génération accrue. Pour soutenir et prévenir la perte de détails fins lors de la génération, le décodeur du Video-VAE effectue la conversion latent vers pixel et la dernière étape de débruitage.

Vous pouvez trouver tous les points de contrôle originaux de LTX-Video sous l'organisation Lightricks.

Cliquez sur les modèles LTX-Video dans la barre latérale droite pour voir d'autres exemples de tâches de génération de vidéos.

L'exemple ci-dessous montre comment générer une vidéo optimisée pour la mémoire ou la vitesse d'inférence.

mémoire

vitesse d'inférence

Notes

Reportez-vous aux paramètres recommandés suivants pour la génération à partir du dépôt LTX-Video.
- Le type de données recommandé pour le transformer, le VAE et l'encodeur de texte est torch.bfloat16. Le VAE et l'encodeur de texte peuvent également être en torch.float32 ou en torch.float16.
- Pour les variantes distillées par guidage de LTX-Video, définissez guidance_scale sur 1.0. Le guidance_scale de tout autre modèle doit être défini sur une valeur plus élevée, comme 5.0, pour une bonne qualité de génération.
- Pour les variantes de VAE prenant en compte le pas de temps (LTX-Video 0.9.1 et versions ultérieures), définissez decode_timestep sur 0.05 et image_cond_noise_scale sur 0.025.
- Pour les variantes qui prennent en charge l’interpolation entre plusieurs images et vidéos de conditionnement (LTX-Video 0.9.5 et versions ultérieures), utilisez des images et des vidéos similaires pour obtenir les meilleurs résultats. Un écart par rapport aux entrées de conditionnement peut entraîner des transitions abruptes dans la vidéo générée.

LTX-Video 0.9.7 intègre un upscaler latent spatial et un transformer de 13 milliards de paramètres. Lors de l'inférence, une vidéo basse résolution est d'abord générée rapidement, puis mise à l'échelle et affinée.

Afficher le code d'exemple

import torch
from diffusers import LTXConditionPipeline, LTXLatentUpsamplePipeline
from diffusers.pipelines.ltx.pipeline_ltx_condition import LTXVideoCondition
from diffusers.utils import export_to_video, load_video

pipeline = LTXConditionPipeline.from_pretrained("Lightricks/LTX-Video-0.9.7-dev", dtype=torch.bfloat16)
pipeline_upsample = LTXLatentUpsamplePipeline.from_pretrained("Lightricks/ltxv-spatial-upscaler-0.9.7", vae=pipeline.vae, dtype=torch.bfloat16)
pipeline.to("cuda")  # or "mps", "xpu", "cpu"
pipe_upsample.to("cuda")
pipeline.vae.enable_tiling()

def round_to_nearest_resolution_acceptable_by_vae(height, width):
    height = height - (height % pipeline.vae_temporal_compression_ratio)
    width = width - (width % pipeline.vae_temporal_compression_ratio)
    return height, width

video = load_video(
    "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cosmos/cosmos-video2world-input-vid.mp4"
)[:21]  # only use the first 21 frames as conditioning
condition1 = LTXVideoCondition(video=video, frame_index=0)

prompt = """
The video depicts a winding mountain road covered in snow, with a single vehicle 
traveling along it. The road is flanked by steep, rocky cliffs and sparse vegetation. 
The landscape is characterized by rugged terrain and a river visible in the distance. 
The scene captures the solitude and beauty of a winter drive through a mountainous region.
"""
negative_prompt = "worst quality, inconsistent motion, blurry, jittery, distorted"
expected_height, expected_width = 768, 1152
downscale_factor = 2 / 3
num_frames = 161

# 1. Generate video at smaller resolution
# Text-only conditioning is also supported without the need to pass `conditions`
downscaled_height, downscaled_width = int(expected_height * downscale_factor), int(expected_width * downscale_factor)
downscaled_height, downscaled_width = round_to_nearest_resolution_acceptable_by_vae(downscaled_height, downscaled_width)
latents = pipeline(
    conditions=[condition1],
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=downscaled_width,
    height=downscaled_height,
    num_frames=num_frames,
    num_inference_steps=30,
    decode_timestep=0.05,
    decode_noise_scale=0.025,
    image_cond_noise_scale=0.0,
    guidance_scale=5.0,
    guidance_rescale=0.7,
    generator=torch.Generator().manual_seed(0),
    output_type="latent",
).frames

# 2. Upscale generated video using latent upsampler with fewer inference steps
# The available latent upsampler upscales the height/width by 2x
upscaled_height, upscaled_width = downscaled_height * 2, downscaled_width * 2
upscaled_latents = pipe_upsample(
    latents=latents,
    output_type="latent"
).frames

# 3. Denoise the upscaled video with few steps to improve texture (optional, but recommended)
video = pipeline(
    conditions=[condition1],
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=upscaled_width,
    height=upscaled_height,
    num_frames=num_frames,
    denoise_strength=0.4,  # Effectively, 4 inference steps out of 10
    num_inference_steps=10,
    latents=upscaled_latents,
    decode_timestep=0.05,
    decode_noise_scale=0.025,
    image_cond_noise_scale=0.0,
    guidance_scale=5.0,
    guidance_rescale=0.7,
    generator=torch.Generator().manual_seed(0),
    output_type="pil",
).frames[0]

# 4. Downscale the video to the expected resolution
video = [frame.resize((expected_width, expected_height)) for frame in video]

export_to_video(video, "output.mp4", fps=24)

Le modèle distillé LTX-Video 0.9.7 est distillé en termes de guidance et de timesteps pour accélérer la génération. Il est nécessaire de définir guidance_scale sur 1.0 et de régler num_inference_steps entre 4 et 10 pour une bonne qualité de génération. Vous devriez également utiliser les timesteps personnalisés suivants pour obtenir les meilleurs résultats.

Inférence du modèle de base pour préparer le redimensionnement : [1000, 993, 987, 981, 975, 909, 725, 0.03].
Upscaling : [1000, 909, 725, 421, 0].

Afficher le code d'exemple

import torch
from diffusers import LTXConditionPipeline, LTXLatentUpsamplePipeline
from diffusers.pipelines.ltx.pipeline_ltx_condition import LTXVideoCondition
from diffusers.utils import export_to_video, load_video

pipeline = LTXConditionPipeline.from_pretrained("Lightricks/LTX-Video-0.9.7-distilled", dtype=torch.bfloat16)
pipe_upsample = LTXLatentUpsamplePipeline.from_pretrained("Lightricks/ltxv-spatial-upscaler-0.9.7", vae=pipeline.vae, dtype=torch.bfloat16)
pipeline.to("cuda")  # or "mps", "xpu", "cpu"
pipe_upsample.to("cuda")
pipeline.vae.enable_tiling()

def round_to_nearest_resolution_acceptable_by_vae(height, width):
    height = height - (height % pipeline.vae_spatial_compression_ratio)
    width = width - (width % pipeline.vae_spatial_compression_ratio)
    return height, width

prompt = """
artistic anatomical 3d render, utlra quality, human half full male body with transparent 
skin revealing structure instead of organs, muscular, intricate creative patterns, 
monochromatic with backlighting, lightning mesh, scientific concept art, blending biology 
with botany, surreal and ethereal quality, unreal engine 5, ray tracing, ultra realistic, 
16K UHD, rich details. camera zooms out in a rotating fashion
"""
negative_prompt = "worst quality, inconsistent motion, blurry, jittery, distorted"
expected_height, expected_width = 768, 1152
downscale_factor = 2 / 3
num_frames = 161

# 1. Generate video at smaller resolution
downscaled_height, downscaled_width = int(expected_height * downscale_factor), int(expected_width * downscale_factor)
downscaled_height, downscaled_width = round_to_nearest_resolution_acceptable_by_vae(downscaled_height, downscaled_width)
latents = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=downscaled_width,
    height=downscaled_height,
    num_frames=num_frames,
    timesteps=[1000, 993, 987, 981, 975, 909, 725, 0.03],
    decode_timestep=0.05,
    decode_noise_scale=0.025,
    image_cond_noise_scale=0.0,
    guidance_scale=1.0,
    guidance_rescale=0.7,
    generator=torch.Generator().manual_seed(0),
    output_type="latent",
).frames

# 2. Upscale generated video using latent upsampler with fewer inference steps
# The available latent upsampler upscales the height/width by 2x
upscaled_height, upscaled_width = downscaled_height * 2, downscaled_width * 2
upscaled_latents = pipe_upsample(
    latents=latents,
    adain_factor=1.0,
    output_type="latent"
).frames

# 3. Denoise the upscaled video with few steps to improve texture (optional, but recommended)
video = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=upscaled_width,
    height=upscaled_height,
    num_frames=num_frames,
    denoise_strength=0.999,  # Effectively, 4 inference steps out of 5
    timesteps=[1000, 909, 725, 421, 0],
    latents=upscaled_latents,
    decode_timestep=0.05,
    decode_noise_scale=0.025,
    image_cond_noise_scale=0.0,
    guidance_scale=1.0,
    guidance_rescale=0.7,
    generator=torch.Generator().manual_seed(0),
    output_type="pil",
).frames[0]

# 4. Downscale the video to the expected resolution
video = [frame.resize((expected_width, expected_height)) for frame in video]

export_to_video(video, "output.mp4", fps=24)

Le modèle distillé LTX-Video 0.9.8 est similaire à la variante 0.9.7. Il est distillé en termes de guidance et de timestep, et un code d'inférence similaire peut être utilisé comme ci-dessus. Une amélioration de cette version est qu'elle prend en charge la génération de vidéos très longues. De plus, elle prend en charge l'utilisation du tone mapping pour améliorer la qualité de la vidéo générée à l'aide du paramètre tone_map_compression_ratio. La valeur par défaut de 0.6 est recommandée.

Afficher le code d'exemple

import torch
from diffusers import LTXConditionPipeline, LTXLatentUpsamplePipeline
from diffusers.pipelines.ltx.pipeline_ltx_condition import LTXVideoCondition
from diffusers.models.latent_upscaler import LTXLatentUpsamplerModel
from diffusers.utils import export_to_video, load_video

pipeline = LTXConditionPipeline.from_pretrained("Lightricks/LTX-Video-0.9.8-13B-distilled", dtype=torch.bfloat16)
# TODO: Update the checkpoint here once updated in LTX org
upsampler = LTXLatentUpsamplerModel.from_pretrained("a-r-r-o-w/LTX-0.9.8-Latent-Upsampler", dtype=torch.bfloat16)
pipe_upsample = LTXLatentUpsamplePipeline(vae=pipeline.vae, latent_upsampler=upsampler).to(torch.bfloat16)
pipeline.to("cuda")  # or "mps", "xpu", "cpu"
pipe_upsample.to("cuda")
pipeline.vae.enable_tiling()

def round_to_nearest_resolution_acceptable_by_vae(height, width):
    height = height - (height % pipeline.vae_spatial_compression_ratio)
    width = width - (width % pipeline.vae_spatial_compression_ratio)
    return height, width

prompt = """The camera pans over a snow-covered mountain range, revealing a vast expanse of snow-capped peaks and valleys.The mountains are covered in a thick layer of snow, with some areas appearing almost white while others have a slightly darker, almost grayish hue. The peaks are jagged and irregular, with some rising sharply into the sky while others are more rounded. The valleys are deep and narrow, with steep slopes that are also covered in snow. The trees in the foreground are mostly bare, with only a few leaves remaining on their branches. The sky is overcast, with thick clouds obscuring the sun. The overall impression is one of peace and tranquility, with the snow-covered mountains standing as a testament to the power and beauty of nature."""
# prompt = """A woman walks away from a white Jeep parked on a city street at night, then ascends a staircase and knocks on a door. The woman, wearing a dark jacket and jeans, walks away from the Jeep parked on the left side of the street, her back to the camera; she walks at a steady pace, her arms swinging slightly by her sides; the street is dimly lit, with streetlights casting pools of light on the wet pavement; a man in a dark jacket and jeans walks past the Jeep in the opposite direction; the camera follows the woman from behind as she walks up a set of stairs towards a building with a green door; she reaches the top of the stairs and turns left, continuing to walk towards the building; she reaches the door and knocks on it with her right hand; the camera remains stationary, focused on the doorway; the scene is captured in real-life footage."""
negative_prompt = "bright colors, symbols, graffiti, watermarks, worst quality, inconsistent motion, blurry, jittery, distorted"
expected_height, expected_width = 480, 832
downscale_factor = 2 / 3
# num_frames = 161
num_frames = 361

# 1. Generate video at smaller resolution
downscaled_height, downscaled_width = int(expected_height * downscale_factor), int(expected_width * downscale_factor)
downscaled_height, downscaled_width = round_to_nearest_resolution_acceptable_by_vae(downscaled_height, downscaled_width)
latents = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=downscaled_width,
    height=downscaled_height,
    num_frames=num_frames,
    timesteps=[1000, 993, 987, 981, 975, 909, 725, 0.03],
    decode_timestep=0.05,
    decode_noise_scale=0.025,
    image_cond_noise_scale=0.0,
    guidance_scale=1.0,
    guidance_rescale=0.7,
    generator=torch.Generator().manual_seed(0),
    output_type="latent",
).frames

# 2. Upscale generated video using latent upsampler with fewer inference steps
# The available latent upsampler upscales the height/width by 2x
upscaled_height, upscaled_width = downscaled_height * 2, downscaled_width * 2
upscaled_latents = pipe_upsample(
    latents=latents,
    adain_factor=1.0,
    tone_map_compression_ratio=0.6,
    output_type="latent"
).frames

# 3. Denoise the upscaled video with few steps to improve texture (optional, but recommended)
video = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=upscaled_width,
    height=upscaled_height,
    num_frames=num_frames,
    denoise_strength=0.999,  # Effectively, 4 inference steps out of 5
    timesteps=[1000, 909, 725, 421, 0],
    latents=upscaled_latents,
    decode_timestep=0.05,
    decode_noise_scale=0.025,
    image_cond_noise_scale=0.0,
    guidance_scale=1.0,
    guidance_rescale=0.7,
    generator=torch.Generator().manual_seed(0),
    output_type="pil",
).frames[0]

# 4. Downscale the video to the expected resolution
video = [frame.resize((expected_width, expected_height)) for frame in video]

export_to_video(video, "output.mp4", fps=24)

LTX-Video prend en charge les LoRA avec load_lora_weights().

Afficher le code d'exemple

import torch
from diffusers import LTXConditionPipeline
from diffusers.utils import export_to_video, load_image

pipeline = LTXConditionPipeline.from_pretrained(
    "Lightricks/LTX-Video-0.9.5", dtype=torch.bfloat16
)

pipeline.load_lora_weights("Lightricks/LTX-Video-Cakeify-LoRA", adapter_name="cakeify")
pipeline.set_adapters("cakeify")

# use "CAKEIFY" to trigger the LoRA
prompt = "CAKEIFY a person using a knife to cut a cake shaped like a Pikachu plushie"
image = load_image("https://huggingface.co/Lightricks/LTX-Video-Cakeify-LoRA/resolve/main/assets/images/pikachu.png")

video = pipeline(
    prompt=prompt,
    image=image,
    width=576,
    height=576,
    num_frames=161,
    decode_timestep=0.03,
    decode_noise_scale=0.025,
    num_inference_steps=50,
).frames[0]
export_to_video(video, "output.mp4", fps=26)

LTX-Video prend en charge le chargement à partir de fichiers uniques, tels que des points de contrôle GGUF, via loaders.FromOriginalModelMixin.from_single_file() ou loaders.FromSingleFileMixin.from_single_file().

Afficher le code d'exemple

import torch
from diffusers.utils import export_to_video
from diffusers import LTXPipeline, AutoModel, GGUFQuantizationConfig

transformer = AutoModel.from_single_file(
    "https://huggingface.co/city96/LTX-Video-gguf/blob/main/ltx-video-2b-v0.9-Q3_K_S.gguf"
    quantization_config=GGUFQuantizationConfig(compute_dtype=torch.bfloat16),
    dtype=torch.bfloat16
)
pipeline = LTXPipeline.from_pretrained(
    "Lightricks/LTX-Video",
    transformer=transformer,
    dtype=torch.bfloat16
)

Diffusers

LTX-Video

Notes

LTXI2VLongMultiPromptPipeline

classe diffusers.LTXI2VLongMultiPromptPipeline

__call__

encode_prompt

prepare_latents

vae_decode_tiled

LTXPipeline

classe diffusers.LTXPipeline

__call__

encode_prompt

LTXImageToVideoPipeline

classe diffusers.LTXImageToVideoPipeline

__call__

encode_prompt

LTXConditionPipeline

classe diffusers.LTXConditionPipeline

__call__

add_noise_to_image_conditioning_latents

encode_prompt

trim_conditioning_sequence

LTXLatentUpsamplePipeline

classe diffusers.LTXLatentUpsamplePipeline

__call__

adain_filter_latent

tone_map_latents

LTXPipelineOutput

classe diffusers.pipelines.ltx.pipeline_output.LTXPipelineOutput

call

call

call

call

call