High Dynamic Range (HDR) user-generated (UGC) videos are rapidly proliferating across social platforms, yet most perceptual video quality assessment (VQA) systems remain tailored to Standard Dynamic Range (SDR). HDR has a higher bit depth, wide color gamut, and elevated luminance range, exposing distortions such as near-black crushing, highlight clipping, banding, and exposure flicker that amplify UGC artifacts and challenge SDR models. To catalyze progress, we curate Beyond8Bits, a large-scale subjective dataset of 44K videos from 6.5K sources with over 1.5M crowd ratings, spanning diverse scenes, capture conditions, and compression settings. We further introduce HDR-Q, the first Multimodal Large Language Model (MLLM) for HDR-UGC VQA. We propose (i) a novel HDR-aware vision encoder to produce HDR-sensitive embeddings, and (ii) HDR-Aware Policy Optimization (HAPO), an RL finetuning framework that anchors reasoning to HDR cues. HAPO augments GRPO via an HDR-SDR contrastive KL that encourages token reliance on HDR inputs and a Gaussian weighted regression reward for fine-grained MOS calibration. Across Beyond8Bits and public HDR-VQA benchmarks, HDR-Q delivers state-of-the-art performance.
Figures & tables
Figure 1: Overview of our dataset and performance evaluation. Top: Example comparisons between HDR and SDR frames, illustrating differences in brightness range, color depth, and visual detail across diverse scenes. Bottom-left: The distribution of video categories in the Beyond8Bits dataset, covering human-centered content, nature & outdoor scenes, and various other real-world scenarios. Bottom-right: Performance comparison between our proposed HDR-Q model and baseline methods on three datasets, where HDR-Q achieves significant improvements in PLCC.
Figure 2: Typical challenges in HDR-UGC videos, including HDR-specific issues (e.g., highlight clipping, color blooming, dark grain) and UGC/compression-related artifacts (e.g., color distortion, blocking, ringing, blurring).
Figure 3: Pipeline of Beyond8Bits construction.
Figure 4: Overview of HDR-Q with HAPO . Left: HAPO compares rollouts under HDR inputs (text + SDR + HDR tokens) versus an HDR-deprived pathway (text + SDR only), maximizing their KL divergence to enforce HDR grounding and applying dual-entropy regularization to prevent reward hacking. Group-wise rewards include MOS/attribute accuracy, reasoning quality, and self-rewarding. Right: a LoRA-tuned LLM decodes the HDR-aware reasoning; visual inputs originate from both a standard encoder and our HDR-aware adapter.
Figure 5: HDR-aware vision encoder finetuning. We adapt SigLIP-2 ( Tschannen et al., 2025 ) using HDR–SDR frame–caption pairs with captions generated by Qwen2.5-VL-72B, promoting perceptually aligned HDR embeddings.
Figure 6: Given the same HDR video, OVIS 2.5 produces multiple incorrect judgments. In contrast, our HAPO-enhanced HDR-Q provides HDR-grounded reasoning. (Best viewed zoomed in)
Model
LIVE-HDR
SFV+HDR
SROCC( ↑ )
PLCC( ↑ )
RMSE( ↓ )
KRCC( ↑ )
SROCC( ↑ )
PLCC( ↑ )
RMSE( ↓ )
KRCC( ↑ )
DL models
BRISQUE ( Mittal et al., 2012 )
0.7251
0.7139
12.6404
0.3424
0.4664
0.4186
0.3811
0.3165
CONTRIQUE ( Madhusudana et al., 2022 )
0.8170
0.7875
11.2514
0.5876
0.5901
0.5959
0.3368
0.4204
RE-IQA ( Saha et al., 2023 )
0.7196
0.6883
15.1653
0.5197
0.5822
0.5998
0.3072
0.4145
VBLIINDS ( Saad et al., 2014 )
0.7483
0.7193
12.7794
0.2541
0.3335
0.2713
0.3988
0.2300
Table 2: Cross-dataset performance Comparison on LIVE-HDR ( Shang et al., 2023 ) and SFV+HDR ( Wang et al., 2024 ) Datasets.
Variant
HDR-Enc.
HAPO
HDR–SDR KL
Dual Ent.
HEW
Self-R.
PLCC
SRCC
RMSE
KRCC
CoT len
Tok. H
GRPO baseline
✗
✗
✗
✗
✗
✗
0.79
0.81
10.73
0.56
168
0.20
GRPO + HDR-Enc.
✓
✗
✗
✗
✗
✗
0.81
0.83
8.96
0.61
161
0.24
HAPO w/o HDR–SDR KL
✓
✓
✗
✓
✓
✓
0.84
0.86
7.10
0.64
142
0.29
HAPO w/o Dual Ent.
✓
✓
✓
✗
✓
✓
0.89
0.91
5.82
0.71
148
0.26
HAPO w/o HEW
✓
✓
✓
✓
✗
✓
0.87
0.88
6.11
0.68
155
0.27
HAPO w/o Self-Reward
✓
✓
✓
✓
✓
✗
0.90
0.92
5.22
0.71
140
0.31
Table 3: Component ablation on Beyond8Bits. ✓=enabled, ✗=disabled. CoT len: CoT length and Tok. H: mean token entropy.
Figure 7: Analysis of Chain-of-Thought (CoT) Length and Token Entropy over training iterations. (a) shows the decrease in CoT length, while (b) shows the corresponding increase in token entropy.
The high dynamic range (HDR) video ecosystem is approaching maturity, but the problem of upconverting legacy standard dynamic range (SDR) videos persists without a convincing solution. We propose a framework for HDR video synthesis from casual SDR footage by leveraging large-scale generative video models. We introduce a Multi-Exposure Video Model (MEVM) that can predict exposure-bracketed linear SDR video sequences from a single nonlinear SDR video input. We further propose a learnable Video Merging Model (VMM) that merges the predicted exposure-bracketed video into a high-quality HDR sequence while preserving detail in both shadows and highlights. Extensive experiments, quantitative and qualitative evaluation, and a user study demonstrate that our approach enables robust HDR conversion for in-the-wild examples from casual consumer videos and even iconic films. Finally, our model can support HDR synthesis pipelines built upon existing SDR generative video models. Output HDR videos can be viewed on our supplementary webpage: sdr2hdrvideo.github.io
SaiKiran Tedla, Francesco Banterle, Trevor Canham +6
Sony AI, Japan · York University, Canada · ISTI-CNR, Italy +1
High dynamic range (HDR) imagery offers a rich and faithful representation of scene radiance, but remains challenging for generative models due to its mismatch with the bounded, perceptually compressed data on which these models are trained. A natural solution is to learn new representations for HDR, which introduces additional complexity and data requirements. In this work, we show that HDR generation can be achieved in a much simpler way by leveraging the strong visual priors already captured by pretrained generative models. We observe that a logarithmic encoding widely used in cinematic pipelines maps HDR imagery into a distribution that is naturally aligned with the latent space of these models, enabling direct adaptation via lightweight fine-tuning without retraining an encoder. To recover details that are not directly observable in the input, we further introduce a training strategy based on camera-mimicking degradations that encourages the model to infer missing high dynamic range content from its learned priors. Combining these insights, we demonstrate high-quality HDR video generation using a pretrained video model with minimal adaptation, achieving strong results across diverse scenes and challenging lighting conditions. Our results indicate that HDR, despite representing a fundamentally different image formation regime, can be handled effectively without redesigning generative models, provided that the representation is chosen to align with their learned priors.
Naomi Ken Korem, Mohamed Oumoumad, Omer Hagage +8
Lightricks · Gear Productions · Tel Aviv University
Most digital videos are stored in 8-bit low dynamic range (LDR) formats, where much of the original high dynamic range (HDR) scene radiance is lost due to saturation and quantization. This loss of highlight and shadow detail precludes mapping accurate luminance to HDR displays and limits meaningful re-exposure in post-production workflows. Although techniques have been proposed to convert LDR images to HDR through dynamic range expansion, they struggle to restore realistic detail in the over- and underexposed regions. To address this, we present DiffHDR, a framework that formulates LDR-to-HDR conversion as a generative radiance inpainting task within the latent space of a video diffusion model. By operating in Log-Gamma color space, DiffHDR leverages spatio-temporal generative priors from a pretrained video diffusion model to synthesize plausible HDR radiance in over- and underexposed regions while recovering the continuous scene radiance of the quantized pixels. Our framework further enables controllable LDR-to-HDR video conversion guided by text prompts or reference images. To address the scarcity of paired HDR video data, we develop a pipeline that synthesizes high-quality HDR video training data from static HDRI maps. Extensive experiments demonstrate that DiffHDR significantly outperforms state-of-the-art approaches in radiance fidelity and temporal stability, producing realistic HDR videos with considerable latitude for re-exposure.