◎ Radiance HDR Exposure Blend

Photographers solve the "too much dynamic range" problem by shooting three copies of the same frame at different exposures and merging them. HDRExposureBlend is that merge, turned into a ComfyUI node: it takes a dark exposure (good highlights), a bright exposure (good shadows), optionally a middle one, and fuses them into a single HDR image that keeps the best of each.

In ComfyUI terms, the realistic source is a generation run at multiple exposure offsets - say the same seed at −2 EV, 0 EV, and +2 EV - or a batch where you varied the exposure. The node's whole philosophy is in its description: "Takes highlights from low exposure and shadows from high exposure for optimal color grading."

How it works

The default Mertens Fusion is the classic exposure-fusion algorithm, and the source implements it properly: for each exposure it computes a weight map from contrast (Laplacian energy), saturation, and well-exposedness, then blends with those weights. No tone-mapping step is needed - the fusion is the merge. The other methods (Luminance Weighted, Shadow/Highlight Mask, Exposure Weighted, Laplacian Pyramid) are alternative weighting strategies; Mertens is the one to start with, Laplacian Pyramid the one to try when you see halos.

Before fusion, the exposures are normalized to the same scale using exposure_offset_low/high/mid - these tell the node the EV of each input relative to 0 so it can compensate. Defaults are −2/+2/0, which matches exposures you bracketed symmetrically.

The inputs

• low_exposure (dark) and high_exposure (bright) - required. mid_exposure optional.
• blend_method - the five-way choice above.
• shadow_weight / highlight_weight (0–2) - bias the merge toward shadows or highlights.
• transition_smoothness - feathering between exposure regions.
• exposure_offset_low / high / mid - the EV of each input; get these right or the merge fights itself.
• ghost_removal (default off) - for when the frames aren't perfectly aligned (motion between shots).

Outputs: blended_hdr (the merged float image), blend_mask (the weight map - great for debugging where each exposure won), and blend_info (a STRING summary).

The traps

• This wants exposures of the same frame. If your "bracketed" images are actually different generations, ghosting and motion artifacts are guaranteed - that's what ghost_removal exists for, but it only does so much.
• Mertens fusion expects roughly aligned, normal-range inputs. Feeding it wild float HDR data before normalizing will make the well-exposedness term meaningless.
• The output is a single merged HDR float image. From there it's the usual route: grade in linear, tone-map to SDR, or save as EXR. The pack's community framing - HDR/EXR work inside ComfyUI - is exactly the workflow this node serves.

Install

ComfyUI Manager → search Radiance, or:

cd ComfyUI/custom_nodes
git clone https://github.com/fxtdstudios/radiance.git
cd radiance
pip install -r requirements.txt

Restart. Under FXTD Studios/Radiance/Color.