How to Add Native HDR Output to Video for VFX Workflows

This tutorial is based on a hands-on walkthrough by VFX creator Doug Hogan, who tested a new HDR IC-LoRA and shared his results on YouTube. Watch the video above, then follow the steps below to run it yourself.

Why this matters: AI video has been stuck at 8-bit

Every AI video model released so far, including earlier versions of LTX, has generated standard 8-bit SDR footage. That means each color channel is limited to 256 possible values. For casual viewing on YouTube or a phone screen, that's fine. But the moment that footage enters a real production pipeline for color grading, compositing, or final conform, the limits show up fast: banding, clipped highlights, and no headroom to grade.

The IC-LoRA covered in this tutorial closes that gap by converting standard 8-bit output into scene-linear 16-bit EXR, the same file format used throughout the VFX and film industry for decades. Sixteen-bit color gives roughly 65,000 possible values per channel instead of 256, a jump in precision that translates directly into usable range for grading in tools like DaVinci Resolve or Nuke.

Under the hood, the model generates in a log-encoded, compressed color space, then applies a post-decode inversion transform that expands the result back out to linear floating-point values. That's what lets the extended dynamic range travel through generation inside a normal numerical range, then unpack afterward into a full 16-bit EXR. You don't need to manage any of that manually; it happens as part of the workflow below.

What you'll need

• ComfyUI with the ComfyUI-LTX-Video custom node installed
• The HDR IC-LoRA and its example workflow, both on the LTX Hugging Face page
• A viewer capable of displaying scene-linear EXR correctly, such as Nuke or DaVinci Resolve (a standard image viewer will render the raw output looking washed out or strange, since it isn't meant to be viewed at face value)

Step 1: Download the IC-LoRA and workflow

Go to the LTX Hugging Face page and download the HDR IC-LoRA file along with the example workflow JSON. Drop the LoRA file into your standard ComfyUI LoRA folder.

Step 2: Load the workflow into ComfyUI

Open the example workflow JSON in ComfyUI. Treat this step as another post-process stage in your pipeline: place it after your upscaling step, so the IC-LoRA is working on footage that's already at your target resolution rather than a low-res intermediate.

Step 3: Feed in your finished SDR clip

Load your final AI-generated clip, the version you'd otherwise be sending off for delivery, into the workflow's video input node.

Step 4: Set your output path and run

Point the output node to wherever you want the EXR sequence saved, then run the workflow. The IC-LoRA will transform your clip through the log-to-linear decode step and output a 16-bit scene-linear EXR.

Step 5: View the result correctly

Open the output in Nuke or Resolve rather than a standard media player. Because you're working with linear EXRs, the footage will look wrong or overexposed until you apply an exposure adjustment; that's expected. Pull the exposure down and you should see substantially more retained highlight and shadow detail than the same adjustment applied to the original 8-bit source, which will clip almost immediately by comparison.

Optional: generate a synthetic HDRI for CG lighting

The same pipeline can also produce usable HDRI light probes, the panoramic HDR images used to light CG scenes, using a still image instead of a video clip:

1. Start with a standard SDR image (a real photo or an AI-generated frame).
2. Feed it into an image-to-video model to generate a short single-frame clip, using a duration setting of 1 frame.
3. Run that frame through the same HDR IC-LoRA workflow described above.
4. Export the result as an .hdr file and load it into your lighting package of choice (Blender, for example) as an environment light.

Results won't be a perfect one-to-one match with a true captured HDRI, but they hold up well enough to give CG lighting work a usable starting point, including directional color bleed from the image's light sources.

The bottom line

Native 16-bit HDR EXR output is a meaningful step for teams trying to bring AI-generated elements into real production pipelines that need to meet broadcast or theatrical delivery standards. Results will vary by footage, but going from an 8-bit generation ceiling to a workable scene-linear EXR is a real unlock for VFX teams evaluating AI tools for actual delivery work, not just previz.

Get started: grab the HDR IC-LoRA and example workflow from the LTX Hugging Face page and the ComfyUI-LTX-Video GitHub repo.