Skip to content

Decode HDR Latents

Decodes an HDR latent, applies tone mapping, and optionally exports the raw linear EXR frame sequence.

Category: ModularDiffusion/Encode\Decode

TL;DR

  • Extends Decode Media Latent: same inputs and dynamic image/video output, plus tone mapping and EXR export for HDR video pipelines (e.g. LTX 2.3 HDR).
  • Output is always a tone-mapped SDR image or MP4 video — set exr_output_folder to also save the raw linear HDR frames as an OpenEXR sequence before tone mapping is applied.
  • Standard (non-HDR) pipelines behave identically to Decode Media Latent; tone mapping is not applied.
  • HDR output applies only to pipeline drivers that return linear np.ndarray frames (LTX 2.3 HDR); all other pipelines use the standard decode path.

Typical workflow position

Generate Media Latents → [Decode HDR Latents] → Save Image / Save Video

Node preview

Decode HDR Latents

Inputs

Name Type Required Notes
pipeline Pipeline Config Yes Must match the pipeline that produced the latent. Use an HDR-capable model (e.g. LTX 2.3 HDR) to get linear frame output.
latent_tensor LatentArtifact Yes Latent to decode.

Outputs

Name Type Notes
output_image ImageArtifact Tone-mapped image. Shown for image pipelines.
output_video VideoUrlArtifact Tone-mapped MP4 video. Shown for video pipelines.
logs str Per-frame EXR write log. Populated only when exr_output_folder is set.

Parameters

Name Type Default Notes
tone_mapping clip \| reinhard \| aces_filmic \| cv2_reinhard \| cv2_mantiuk aces_filmic Operator applied to linear HDR frames before encoding to SDR output. Ignored for non-HDR pipelines.
fps int (1–120) 25 Output frame rate. Only shown for video pipelines.
exr_output_folder path — Folder or .exr file path for saving the raw linear HDR EXR sequence. Leave empty to skip EXR export.
save_exr_as_half_float bool True Store EXR frames as float16 — 2.5× smaller files with negligible quality loss.

Tips & pitfalls

  • aces_filmic preserves bright highlights better than clip. Use clip only when chroma accuracy matters more than highlight rolloff, or when comparing SDR to HDR output numerically.
  • EXR filenames are derived from the path you supply — three patterns are supported:
    • output/frames/ (trailing slash, folder) → files saved as output/frames/frame_0000.exr, frame_0001.exr, …
    • output/frames (no trailing slash, no extension, existing or non-existing folder) → same as above, stem frame
    • output/frames/shot.exr (.exr extension) → files saved as output/frames/shot_0000.exr, shot_0001.exr, …
  • Use the same pipeline that produced the latent. Each pipeline carries the VAE it was trained with — decoding with a mismatched VAE produces corrupt output.
  • Large latents need more VRAM to decode. High-resolution or multi-frame latents require more memory during decode. Enable vae_slicing on the Pipeline Builder to decode in batches and keep peak VRAM usage lower.

See also