README.md

---
library_name: diffusers
pipeline_tag: image-to-video
base_model: Lightricks/LTX-2.3
tags:
- image-to-video
- text-to-video
- video-to-video
- image-text-to-video
- audio-to-video
- text-to-audio
- video-to-audio
- audio-to-audio
- text-to-audio-video
- image-to-audio-video
- image-text-to-audio-video
- ltx-2
- ltx-2-3
- ltx-video
- ltxv
- lightricks
license: other
license_name: ltx-video-2-open-source-license
license_link: https://huggingface.co/Lightricks/LTX-2.3/blob/main/LICENSE
---

# LTX-2.3 Distilled (Diffusers)

Diffusers-format weights for the distilled LTX-2.3 model from [Lightricks/LTX-2.3](https://huggingface.co/Lightricks/LTX-2.3). Runs in **8 steps with CFG = 1**, trading some flexibility for substantially faster inference.

The non-distilled base model is at [`diffusers/LTX-2.3-Diffusers`](https://huggingface.co/diffusers/LTX-2.3-Diffusers).

## Usage

Requires a recent build of `diffusers` with LTX-2 support:

```bash
pip install -U git+https://github.com/huggingface/diffusers
```

The distilled checkpoint uses a fixed sigma schedule. Always pass `sigmas=DISTILLED_SIGMA_VALUES`, `num_inference_steps=8`, and `guidance_scale=1.0`.

### Text-to-video + audio

```python
import torch
from diffusers import LTX2Pipeline
from diffusers.pipelines.ltx2.export_utils import encode_video
from diffusers.pipelines.ltx2.utils import DEFAULT_NEGATIVE_PROMPT, DISTILLED_SIGMA_VALUES

pipe = LTX2Pipeline.from_pretrained(
    "diffusers/LTX-2.3-Distilled-Diffusers", torch_dtype=torch.bfloat16
)
pipe.enable_model_cpu_offload()

prompt = "A flowing river in a forest at golden hour, gentle wind in the leaves."
frame_rate = 24.0

video, audio = pipe(
    prompt=prompt,
    negative_prompt=DEFAULT_NEGATIVE_PROMPT,
    width=768,
    height=512,
    num_frames=121,
    frame_rate=frame_rate,
    num_inference_steps=8,
    sigmas=DISTILLED_SIGMA_VALUES,
    guidance_scale=1.0,
    output_type="np",
    return_dict=False,
)

encode_video(
    video[0],
    fps=frame_rate,
    audio=audio[0].float().cpu(),
    audio_sample_rate=pipe.vocoder.config.output_sampling_rate,
    output_path="ltx2_distilled_t2v.mp4",
)
```

### First-last-frame-to-video (FLF2V)

```python
import torch
from diffusers import LTX2ConditionPipeline
from diffusers.pipelines.ltx2.pipeline_ltx2_condition import LTX2VideoCondition
from diffusers.pipelines.ltx2.utils import DEFAULT_NEGATIVE_PROMPT, DISTILLED_SIGMA_VALUES
from diffusers.utils import load_image

pipe = LTX2ConditionPipeline.from_pretrained(
    "diffusers/LTX-2.3-Distilled-Diffusers", torch_dtype=torch.bfloat16
)
pipe.enable_model_cpu_offload()

first_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/flf2v_input_first_frame.png")
last_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/flf2v_input_last_frame.png")

conditions = [
    LTX2VideoCondition(frames=first_image, index=0, strength=1.0),
    LTX2VideoCondition(frames=last_image, index=-1, strength=1.0),
]

prompt = "CG animation style, a small blue bird takes off from the ground, flapping its wings."
frame_rate = 24.0

video = pipe(
    conditions=conditions,
    prompt=prompt,
    negative_prompt=DEFAULT_NEGATIVE_PROMPT,
    width=768,
    height=512,
    num_frames=121,
    frame_rate=frame_rate,
    num_inference_steps=8,
    sigmas=DISTILLED_SIGMA_VALUES,
    guidance_scale=1.0,
    output_type="np",
    return_dict=False,
)
```

### HDR generation (IC-LoRA)

```python
import torch
from safetensors import safe_open
from diffusers import LTX2HDRPipeline
from diffusers.pipelines.ltx2.export_utils import encode_hdr_tensor_to_mp4
from diffusers.pipelines.ltx2.pipeline_ltx2_hdr_lora import LTX2HDRReferenceCondition
from diffusers.pipelines.ltx2.utils import DISTILLED_SIGMA_VALUES
from diffusers.utils import load_video

pipe = LTX2HDRPipeline.from_pretrained(
    "diffusers/LTX-2.3-Distilled-Diffusers", torch_dtype=torch.bfloat16
)
pipe.enable_model_cpu_offload()
pipe.load_lora_weights(
    "Lightricks/LTX-2.3-22b-IC-LoRA-HDR",
    adapter_name="hdr_lora",
    weight_name="ltx-2.3-22b-ic-lora-hdr-0.9.safetensors",
)
pipe.set_adapters("hdr_lora", 1.0)

reference_video = load_video("input.mp4")
ref_cond = LTX2HDRReferenceCondition(frames=reference_video, strength=1.0)

with safe_open("ltx-2.3-22b-ic-lora-hdr-scene-emb.safetensors", framework="pt", device="cuda") as f:
    connector_video_embeds = f.get_tensor("video_context")
    connector_audio_embeds = f.get_tensor("audio_context")

hdr_video = pipe(
    reference_conditions=[ref_cond],
    connector_video_embeds=connector_video_embeds,
    connector_audio_embeds=connector_audio_embeds,
    width=768,
    height=512,
    num_frames=121,
    frame_rate=24.0,
    num_inference_steps=8,
    sigmas=DISTILLED_SIGMA_VALUES,
    guidance_scale=1.0,
    output_type="pt",
    return_dict=False,
)[0]

encode_hdr_tensor_to_mp4(hdr_video[0], output_mp4="ltx2_hdr.mp4", frame_rate=24.0)
```

## Notes

- `width` and `height` must be divisible by 32; `num_frames` must equal `8k + 1`.
- See the [Diffusers LTX-2 docs](https://huggingface.co/docs/diffusers/main/en/api/pipelines/ltx2) for multimodal guidance, prompt enhancement, and the upscaling/refinement pipeline.

## License

These weights are released under the [LTX Video 2 Open Source License](https://huggingface.co/Lightricks/LTX-2.3/blob/main/LICENSE).