import torch
from diffusers import LTX2InContextPipeline
from diffusers.pipelines.ltx2.export_utils import encode_video
from diffusers.pipelines.ltx2.utils import DEFAULT_NEGATIVE_PROMPT

pipe = LTX2InContextPipeline.from_pretrained(
    "diffusers/LTX-2.3-Diffusers", torch_dtype=torch.bfloat16
)
pipe.enable_model_cpu_offload()

# Load this LoRA
pipe.load_lora_weights("linoyts/ltx2.3-ic-lora-ingredients-multishot", weight_name="lora_weights_step_02500.safetensors", adapter_name="lora")
pipe.set_adapters("lora", 1.0)

video, audio = pipe(
    prompt="<your prompt>",
    negative_prompt=DEFAULT_NEGATIVE_PROMPT,
    # IC-LoRA is reference-conditioned — pass your control video via reference_conditions:
    # reference_conditions=[...],  # see the LTX-2 diffusers docs for the condition object
    width=768, height=512, num_frames=49, frame_rate=25.0,
    num_inference_steps=30, guidance_scale=4.0,
    output_type="np", return_dict=False,
)
encode_video(video[0], fps=25.0, output_path="output.mp4")

For the full reference implementation and ComfyUI workflows, see the official LTX-2 repository.

🔌 Using Trained LoRAs in ComfyUI

Base Model: <a href="https://huggingface.co/Lightricks/LTX-2.3" target="_blank" rel="noopener noreferrer" class="text-blue-400 hover:text-blue-300 underline underline-offset-2 transition-colors"> ltx-2.3-22b-dev.safetensors
LoRA type: IC-LoRA (in-context control)
Training Type: LoRA fine-tuning
Training Steps: 2500
Learning Rate: 0.0001
Batch Size: 1

In order to use the trained LoRA in ComfyUI, follow these steps:

Copy your trained LoRA checkpoint (.safetensors file) to the models/loras folder in your ComfyUI installation.
In your ComfyUI workflow:
- Add the "Load LoRA" node to choose your LoRA file
- Connect it to the "Load Checkpoint" node to apply the LoRA to the base model

You can find reference Text-to-Video (T2V) and Image-to-Video (I2V) workflows in the official LTX-2 repository.

Example Prompts

This model inherits the license of the base model (ltx-2.3-22b-dev.safetensors).

Acknowledgments

Base model: Lightricks
Trainer: LTX-2