LTX 2 Single File Support

[dg845]

What does this PR do?

This PR adds single file / GGUF support for the LTX 2 video model.

Example script using GGUF files from unsloth/LTX-2-GGUF:

import os
import torch
from diffusers import GGUFQuantizationConfig, LTX2Pipeline, LTX2VideoTransformer3DModel
from diffusers.pipelines.ltx2.export_utils import encode_video

model_id = "Lightricks/LTX-2"
single_file_ckpt = "https://huggingface.co/unsloth/LTX-2-GGUF/blob/main/ltx-2-19b-dev-BF16.gguf"

device = "cuda:0"
dtype = torch.bfloat16
seed = 42

transformer_kwargs = {}
_, single_file_ext = os.path.splitext(single_file_ckpt)
if single_file_ext == ".gguf":
    quantization_config = GGUFQuantizationConfig(compute_dtype=dtype)
    transformer_kwargs["quantization_config"] = quantization_config

transformer = LTX2VideoTransformer3DModel.from_single_file(
    single_file_ckpt,
    config=model_id,
    subfolder="transformer",
    **transformer_kwargs,
)

pipe = LTX2Pipeline.from_pretrained(
    model_id,
    transformer=transformer,
    torch_dtype=dtype,
)
pipe.enable_model_cpu_offload(device=device)

prompt = "a dog dancing to energetic electronic dance music"
negative_prompt = "worst quality, inconsistent motion, blurry, jittery, distorted"
frame_rate = 24.0

video, audio = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=768,
    height=512,
    num_frames=121,
    frame_rate=frame_rate,
    num_inference_steps=40,
    guidance_scale=4.0,
    generator=torch.Generator().manual_seed(seed),
    output_type="np",
    return_dict=False,
)
video = (video * 255).round().astype("uint8")
video = torch.from_numpy(video)

encode_video(
    video[0],
    fps=frame_rate,
    audio=audio[0].float().cpu(),
    audio_sample_rate=pipe.vocoder.config.output_sampling_rate,
    output_path="ltx2_t2v_single_file.mp4",
)

Fixes #12981

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@sayakpaul
@DN6
@AmericanPresidentJimmyCarter

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[sayakpaul]

Sweet!

[DN6]

Thanks 👍🏽

[dg845]

Merging as the failing tests are unrelated.