The original source of this model is Felldude/QWEN_2.1_HDR_VAE ; All I have done a conversion of the VAE to ComfyUI format. As this work is derived from the official Qwen Image 2.1 materials, the Qwen Research License Agreement applies; I do not impose any additional terms. Please consult the source link for full information on the model, and refer to Qwen materials for any license-related questions. In summary (Information below is partial information from the source README):
Qwen Image 2.1 HDR VAE
Fine-tuned Qwen Image 2.1 autoencoder for RGBA / HDR-oriented reconstruction.
The goal is stronger dynamic range, color volume, and detail energy for HDR-style workflows—not perfect fidelity to the source frame.
Intended use
HDR-oriented encode/decode in Qwen Image pipelines
Image editing / generation stacks that benefit from punchier latents
Research on RGBA + high dynamic range autoencoders
Not intended for: archival reconstruction, metrics-maximizing ablations, or any setting where matching the input pixel-for-pixel matters more than look.
Limitations
Will diverge from the source on sharpness, saturation, and color volume by design
Alpha is supported; behavior on unusual alpha mattes is not heavily tuned
Very large resolutions are memory-heavy (16× latent downscale)
Description
Original conversion to ComfyUI format.
FAQ
Comments (5)
is it me or the samples are both same
The differences are pretty subtle; I've added another one that I think makes them more visible, using two different formats that should also preserve the differences better.
say me the difference? little saturation?
Subtle differences in light intensity (but not at the extremes, because of clipping) and saturation are probably the most noticeable differences in normal non-HDR workflows. If you have an HDR VAE decoder that doesn't clamp the extremes of intensity (which normal VAE decode does), you'll have more differences at the extremes that are preserved in formats used for HDR like EXR. For ComfyUI, an extension like ComfyUI-OCIO -- https://github.com/SlavaSexton/ComfyUI-OCIO -- provides nodes for decoding without clamping the range and then saving to formats like EXR.
thank. this is GooOod