LTX just dropped an HDR IC-LoRA beta: EXR output, built for production pipelines
First AI video model to output 16-bit half-float EXR frames for production grading.
Lightricks just dropped an HDR IC-LoRA beta for their LTX-2.3 model, marking the first model-level solution for generating true high-dynamic-range output from an AI video model. This upgrade transforms SDR footage into 16-bit half-float EXR frames via video-to-video and image-to-video pipelines, outputting Linear sRGB unbounded. The result drops directly into DaVinci Resolve and standard EXR-compatible compositing tools, making it production-ready. Previous AI video models were capped at 8-bit SDR, which caused highlights to clip and shadows to crush when grading, and failed to composite cleanly against higher-bit-depth CGI. This fix addresses dynamic range as the real bottleneck, not resolution.
The IC-LoRA was trained on top of LTX-2.3 with exposure variations, high/low luminance blurring, contrast augmentation, and MP4 compression artifact injection, ensuring it handles real-world compressed source footage rather than clean lab inputs. The beta release includes per-frame .exr files and an 8-bit SDR .mp4 preview. Lightricks is actively improving it and collecting feedback. Available via HuggingFace, GitHub Python pipeline, ComfyUI workflow, documentation, and the LTX API, this open-source tool is a significant leap for integrating AI video into professional production pipelines.
- Upgrades SDR footage to 16-bit half-float EXR frames via video-to-video and image-to-video pipelines.
- Outputs Linear sRGB unbounded, compatible with DaVinci Resolve and standard EXR compositing tools.
- Trained with exposure variations, blurring, contrast augmentation, and MP4 compression artifact injection for real-world footage.
Why It Matters
Enables AI video to integrate into professional production pipelines with proper HDR grading and compositing.