rootonchair/LTX-2.3-spatial-upscaler-x2-v1.1-Diffusers

LTX-2.3 Spatial Upscaler x2 v1.1 (Diffusers)

Diffusers-format weights for the v1.1 x2 spatial latent upsampler of Lightricks/LTX-2.3, packaged as an LTX2LatentUpsamplePipeline (VAE + latent upsampler).

This is an unofficial staging conversion — there is no existing Diffusers-format repo for any version of this component on the Hub. The VAE submodule is reused as-is from diffusers/LTX-2.3-Diffusers (unchanged between v1.0 and v1.1); only the latent-upsampler weights were converted from the v1.1 checkpoint.

Source checkpoint

• File: ltx-2.3-spatial-upscaler-x2-1.1.safetensors from Lightricks/LTX-2.3
• Converted with scripts/convert_ltx2_to_diffusers.py --version 2.3 --latent_upsampler from a local huggingface/diffusers checkout (0.40.0.dev0).

Usage

import torch
from diffusers import LTX2LatentUpsamplePipeline

pipe = LTX2LatentUpsamplePipeline.from_pretrained(
    "rootonchair/LTX-2.3-spatial-upscaler-x2-v1.1-Diffusers", torch_dtype=torch.bfloat16
).to("cuda")

# `latents` are the video latents produced by an LTX2Pipeline generation
upsampled_latents = pipe(latents=latents, output_type="latent", return_dict=False)[0]

See the Diffusers LTX-2 docs for the full upscaling/refinement workflow alongside LTX2Pipeline.

License

These weights are released under the LTX Video 2 Open Source License.