Technology

Cinematic AI: Lightricks Unveils LTX SDR-to-HDR LoRA

|Author: Viacheslav Vasipenok|3 min read| 9
Cinematic AI: Lightricks Unveils LTX SDR-to-HDR LoRA

The world of generative video just got a massive upgrade. Lightricks has officially released the LTX SDR-to-HDR LoRA, a breakthrough tool that transforms standard video into professional-grade High Dynamic Range (HDR) content.

The TL;DR

  • What it does: Converts any input video into HDR.
  • Color Standard: Full support for ACES.
  • Where to get it: Available on HuggingFace.
  • Integration: Out-of-the-box support for ComfyUI and via API (video2video).

Why This Matters for Professional Filmmaking

Cinematic AI: Lightricks Unveils LTX SDR-to-HDR LoRAFor those looking for a deeper dive, this release marks a significant shift in how we handle GenAI video. Most current generative models output in 8-bit sRGB. While this is perfectly fine for social media reels, it is a disaster for professional cinema and visual effects.

In an 8-bit environment, you are limited to only 256 values per color channel. This leads to "banding" (ugly color stripes in gradients) and a total loss of data when trying to adjust exposure or perform heavy color grading in post-production.

Professional VFX requires thousands of values per channel to manipulate lighting and compositing without degrading the image — the difference between a quick smartphone snap and raw cinematic footage.

The Genius of LumiVid: Latent Alignment

Cinematic AI: Lightricks Unveils LTX SDR-to-HDR LoRAThis model, referred to as LumiVid, is based on research from April 2026 titled "HDR Video Generation via Latent Alignment with Logarithmic Encoding".

What makes it truly fascinating is that the model was never explicitly trained on massive amounts of HDR material.

Instead of a costly ground-up rebuild, the researchers utilized a key insight: LogC3 logarithmic encoding.

This encoding, which is already an industry standard for ARRI Alexa cameras, maps HDR images into a distribution that naturally aligns with the latent space of pre-trained video diffusion models. 

Cinematic AI: Lightricks Unveils LTX SDR-to-HDR LoRABy using a lightweight LoRA (Low-Rank Adaptation) that affects less than 1% of the model’s parameters, Lightricks achieved incredible results with minimal resources:

  • Training Data: Only ~300 clips.
  • Training Time: Roughly 10,000 steps over 8 hours.
  • Hardware: Just a single GPU.

Also read:

Technical Superiority and Stability

Cinematic AI: Lightricks Unveils LTX SDR-to-HDR LoRAThe output of this workflow is a float16 EXR file that is scene-linear. This means the footage is ready for professional post-production workflows, including re-exposure and complex color timing.

One of the most impressive feats here is the temporal consistency. Because the frames are generated jointly by the underlying diffusion base, the flickering issues that typically plague SDR-to-HDR conversion methods are non-existent. In benchmarking tests, this LoRA outperformed specialized HDR base models by +3.7 dB on rare ARRI video footage.

In short: this is "fire" for the industry—a small, efficient bridge that finally brings AI video into the professional cinematic pipeline.

Share:
0