The forward map decodes PQ (the perceptual quantizer curve), applies the BT.2020 to BT.709 gamut matrix M709, encodes with the 709 curve, then quantizes and compresses (ε).
LumaFluxarXiv:2604.02787
SDR input next to LumaFlux output
Left of each white line is the SDR input, right is LumaFlux; the boxed regions are enlarged at right. HDR shown tone-mapped.
LumaFluxarXiv:2604.02787
Where earlier methods fall short
Analytic operators
Reinhard, BT.2446
One curve for every scene. Clipped highlights stay clipped.
Learned converters
HDRTVNet, FMNet, KUNet
Excellent in domain. They drift when the tone mapping changes.
Generative priors
LEDiff, X2HDR
A real structure prior, but color and tone drift.
We wanted that prior, pinned to what the SDR frame measures.
LumaFluxarXiv:2604.02787
Part 1 of 2
How LumaFlux works
What has to change in a frozen SDR image model before it can output calibrated HDR?
LumaFlux starts the flow at the SDR frame
Rectified flow straightens the paths between paired samples, so a few Euler steps follow them. Figure from Liu et al., Flow Straight and Fast.
z1 = zSDR + 0.05 ε
The encoded SDR frame plus a small perturbation, ε ~ N(0, I).
Train on the straight path from the HDR latent to z1.
Sample in 8 Euler steps, with no strength knob.
40 steps cost 4.7× the time and calibrate no better.
LumaFluxarXiv:2604.02787
LumaFlux in one picture
Left: adapting the MM-DiT directly, driven by a text prompt. Right: LumaFlux, with the MM-DiT and VAE frozen.
Frozen FLUX.1-dev MM-DiT, VAE and SigLIP. No text encoder.
PGA, Physically-Guided Adaptation, a LoRA gated by measurements.
PCM, Perceptual Cross-Modulation, and a residual coupler add scene context.
RQS, a monotone rational-quadratic spline, calibrates luma.
0.57%of 12.48B weights trained, 71.3M parameters
LumaFluxarXiv:2604.02787
Measurements of the SDR frame steer every frozen block
All 57 MM-DiT blocks are wrapped (19 dual-stream, 38 single-stream, d = 3072, 24 heads). Flames mark the trainable parts.
Per token luminance, gradient, saturation
Per frame luminance statistics, 8 FFT bands
Per block and step six gains from the conditioner Ψ(t, l)
LumaFluxarXiv:2604.02787
A monotone spline turns the VAE output into HDR luma
Ŷ = RQS(Yout; ξ, η, s)
A rational-quadratic spline (RQS) with 8 bins, 25 parameters per frame.
Widths, heights and slopes are positive, so luma keeps its ordering.
Removing it costs 9.69 dB. A linear spline costs 0.45 dB.
8 widths ξ, 8 heights η and 9 knot slopes s, from a 21.9K-parameter head that starts at the identity (dashed).
LumaFluxarXiv:2604.02787
Part 2 of 2
Data, benchmark, results
Does it hold up when every method gets the same inputs and the same scoring code?
A training corpus of user-generated and professional HDR
314,396SDR-HDR pairs
2,092source videos
171,384 / 143,012UGC / PGC examples
HDR source
What it adds
CHUG (Saini et al., ICIP 2025)
856 user-generated references, 846 for training and 10 reserved
LIVE-TMHDR
40 references with their 120 expert-graded SDR clips
HDRTV1K
1,235 training pairs
Sol Levante
HDR10 master, Netflix Open Content (CC BY)
Missing SDR sides are synthesized with eight tone-mapping operators at three compression levels.
A manifest regenerates everything in about four CPU hours.
(a) PCA of ten frame statistics. The corpus hull holds all 100 LIVE-TMHDR frames and 116 of the 117 HDRTV1K test frames. (b) CIE xy footprint.
LumaFluxarXiv:2604.02787
LumaEval scores every method the same way
Track
Pairs
What is scored
Checkpoint
Synthetic (primary)
100
20 held-out CHUG and LIVE-TMHDR contents, eight tone-mapping variants, 78 video-variant groups
mixed-corpus
Expert
100
LIVE-TMHDR pairs whose SDR is an expert grade
mixed-corpus
Native
117
HDRTV1K test pairs
mixed-corpus
Legacy HDRTV1K
117
same pairs, five-metric suite, published baseline values
HDRTV1K-specific
Fixed-QP
468
intra-only HEVC on the SDR input at QP 27, 32, 37, 42
both
Temporal
5 clips
Bonfire and four CHUG clips, PU21 flicker energy
mixed-corpus
Text-to-HDR
6 clips
LTX-Video clips, 121 frames each, intrinsic metrics only
mixed-corpus
Outputs are PNG16 PQ BT.2020 at 1,000 nit. Every row is scored with identical inputs, outputs and metric code.
Degradations
Eight tone-mapping variants, from BT.2390 to YouTube LogC.
Metrics
PU21-PSNR and SSIM, ΔEITP color error, HDR-LPIPS, HDR-VDP-3, FR-HIDROVQA.
FR-HIDROVQA is a full-reference distance on HIDRO-VQA features (Saini et al., WACV 2024).
Baselines
Two analytic operators, four CNNs, PromptIR, LEDiff, X2HDR, and a VAE + RQS control.
LumaFluxarXiv:2604.02787
Best in every column of the synthetic track
Method
PU21-PSNR ↑
PU21-SSIM ↑
ΔEITP ↓
HDR-LPIPS ↓
HDR-VDP-3 ↑
FR-HIDRO ↓
BT.2446c
23.74
0.8291
61.60
0.307
5.611
0.753
Reinhard operator
23.62
0.8286
61.76
0.295
5.726
0.753
HDRTVNet++
23.03
0.8192
66.67
0.328
5.417
0.703
FMNet
22.86
0.8176
67.92
0.330
5.412
0.724
ITM-LUT
22.92
0.8142
67.41
0.315
5.496
0.761
VAE + RQS (control)
18.54
0.7647
111.67
0.381
3.673
0.924
LumaFlux (ours)
24.23
0.8294
56.54
0.269
5.812
0.631
Macro-averages over the same 100 frame IDs and 78 video-variant groups. PU21-PSNR-Y is in the backup copy. Bold best, underline second best. FR-HIDRO is the HIDRO-VQA feature distance (Saini et al., WACV 2024).
24.23 dBPU21-PSNR, 0.49 dB above the strongest baseline
56.54ΔEITP, 5.06 lower than the strongest baseline
0.631FR-HIDROVQA, from 0.703 for HDRTVNet++
One checkpoint, no per-track tuning. Without diffusion transport the control collapses to 18.54 dB.
LumaFluxarXiv:2604.02787
On human-graded SDR, LumaFlux trails only on ΔEITP and HDR-VDP-3
Method
PU21-PSNR ↑
PU21-SSIM ↑
ΔEITP ↓
HDR-LPIPS ↓
HDR-VDP-3 ↑
FR-HIDRO ↓
BT.2446c
33.16
0.8442
37.83
0.148
7.397
0.660
Reinhard operator
31.48
0.8270
43.59
0.155
7.173
0.662
HDRTVNet++
31.53
0.8125
44.92
0.147
7.144
0.671
ITM-LUT
31.46
0.8085
45.49
0.142
7.090
0.639
LEDiff
18.30
0.6480
103.15
0.496
4.723
0.898
X2HDR
23.78
0.7578
51.28
0.366
5.532
0.758
LumaFlux (ours)
33.31
0.8480
40.19
0.138
6.919
0.544
The 100 LIVE-TMHDR pairs whose SDR side is an expert grade, with the preregistered per-content exposure calibration. Bold best, underline second best. The full table, with every baseline and PU21-PSNR-Y, is in the backup.
FR-HIDROVQA 0.544, 14.9% below the best baseline. BT.2446c keeps the lower color error, 37.83 against 40.19, and the higher HDR-VDP-3.
LumaFluxarXiv:2604.02787
Three scenes, every method
The widest gap is on Sunset, 28.22 dB against 24.67 for BT.2446c.
Full-frame PU21-PSNR above each crop, best in bold. LumaFlux 30.83 dB on Nature (BT.2446c 30.70) and 26.13 on Monument (HDRTVNet++ 25.68). LEDiff (7.90, Nature) and X2HDR (14.36, Monument) are lowest.
LumaFluxarXiv:2604.02787
One region of one frame, every method
Tower skyline, BT.2390 tone mapping at CRF 23. Same crop for every method, LumaFlux outlined. HDR shown tone-mapped.
LumaFlux keeps the facade structure and brings back the sky the curves and CNNs wash out.
LumaFluxarXiv:2604.02787
Where the color error sits
Each panel splits into the preview, upper left, and per-pixel ΔEITP, lower right, on one shared scale. Darker is lower error.
Night city: the other three light up across sky and towers; LumaFlux stays dark.
Flame: LumaFlux still overshoots chroma in the core.
LumaFluxarXiv:2604.02787
Color volume on one city frame
Frames above, CIE xyY point clouds below. Height is PQ-coded luminance; triangles are BT.709 and BT.2020. HDR shown tone-mapped.
BT.2446c and HDRTVNet++ under-fill the volume; LumaFlux recovers most of what the reference fills.
LumaFluxarXiv:2604.02787
Luminance along one row tracks the reference
City towers. The red line marks the sampled row; the plot shows luminance along it on a log nit axis.
0.058LumaFlux, mean absolute log deviation
0.166HDRTVNet++
0.190BT.2446c
LumaFluxarXiv:2604.02787
Removing the spline costs 9.69 dB
Row
PGA
Gate
PCM
Coup.
T-mod
RQS
PSNR (Δ vs g)
ΔEITP (Δ vs g)
(c)
•
•
•
28.57 (−0.62)
33.48 (+5.66)
(d)
•
•
•
•
29.56 (+0.37)
28.38 (+0.57)
(f) seed-0 repeat of (g)
•
•
•
•
•
•
28.67 (−0.52)
35.00 (+7.18)
(g) full design, ours
•
•
•
•
•
•
29.19 (ref.)
27.81 (ref.)
(h) linear spline
•
•
•
•
•
lin.
28.74 (−0.45)
32.16 (+4.35)
(i) no RQS
•
•
•
•
•
19.50 (−9.69)
261.78 (+233.97)
(j) no T-mod
•
•
•
•
•
28.38 (−0.81)
33.90 (+6.09)
Component ablations: 20k steps on a 55-video pool, the same 82 held-out frames. A dot marks a component that is on; Gate is the spectral gate, T-mod the timestep-and-layer modulation, lin. a linear spline. Higher PSNR and lower ΔEITP are better; deltas are relative to row (g). Rows (a), (b) and (e) are in the backup.
No RQS, row (i), falls 9.69 dB.
A linear spline, row (h), costs 0.45 dB.
The spectral gate, (c) to (d), lifts 28.57 to 29.56 dB.
No timestep-layer modulation, row (j), costs 0.81 dB.
Rows (f) and (g) are the same configuration, 0.52 dB apart, so smaller gaps are noise.
LumaFluxarXiv:2604.02787
Heavier compression barely moves LumaFlux
PU21-PSNR change from QP 27 (QP is the HEVC quantization parameter) on the 117 HDRTV1K pairs. Flatter is better.
PU21-PSNR (dB)
QP 27
QP 32
QP 37
QP 42
Δ
HDRTVNet++
35.90
35.30
34.34
33.13
−2.77
FMNet
35.56
34.97
34.04
32.85
−2.71
LumaFlux (mixed-corpus)
27.29
27.27
27.04
26.75
−0.55
LumaFlux (in-domain)
33.94
33.67
33.12
32.33
−1.61
Fixed-QP track, 117 HDRTV1K pairs. Δ is QP 42 minus QP 27; closer to zero is better. Mixed-corpus is the main checkpoint; in-domain is the HDRTV1K-specific one. Bold best, underline second per column. ΔEITP and HDR-VDP-3 are in the backup.
From QP 27 to 42 the mixed-corpus model loses 0.55 dB. HDRTVNet++ loses 2.77 and FMNet 2.71.
The CNNs still score higher at every QP on their home set.
LumaFluxarXiv:2604.02787
What it costs
5.28 sper 1080p frame, 8 steps, one GH200
27.06 GBpeak memory at 1080p
71.32Mtrainable of 12.48B parameters
FR-HIDROVQA (lower is better) on 30 expert-track frames. LumaFlux 0.431, 0.477 and 0.567 at 540p, 720p and 1080p, in 2.64, 3.08 and 5.29 s.
Method
Steps
Params (M)
Trainable (M)
s/frame
TFLOPs
Mem. (GB)
HDRTVNet++
1
0.59
0.59
0.857
0.5
2.31
ITM-LUT*
1
0.60
0.60
0.367
–
0.18
PromptIR + Reinhard
1
35.59
35.59
1.708
10.9
17.86
LEDiff (5122)
50
1370.23
859.53
4.449
149
6.37
X2HDR (5122)
30
17020.02
149.42
4.244
1,023
33.77
LumaFlux, 8 steps
8
12484.77
71.32
5.282
2,067
27.06
LumaFlux, 40 steps
40
12484.77
71.32
24.993
10,100
27.06
One GH200, batch 1, complete 1080p exports; LEDiff and X2HDR at their registered 5122 protocol (scaled by pixel count to 1080p: 35.19 and 33.57 s). *ITM-LUT runs a compiled kernel the tracer cannot observe. Lower is better for s/frame, TFLOPs and memory. Full table in the backup.
2.4 to 16.1× faster than X2HDR and LEDiff. HDRTVNet++ stays 173 to 201× faster than us.
LumaFluxarXiv:2604.02787
Video without temporal training
Per-frame mean PU21 luma, four clips, first 120 frames, relative to each reference mean. Only the stabilized run is drawn.
Method
Bonfire
02c3
055e
0392
0237
Mean
BT.2446c
7.813
1.411
0.450
0.063
0.124
1.972
HDRTVNet++
3.005
0.207
0.147
0.442
1.257
1.012
FMNet
3.853
0.202
0.074
0.579
1.124
1.166
LumaFlux
1.992
0.691
2.500
0.968
0.569
1.344
LumaFlux + stabilization
2.109
0.469
1.676
0.879
0.362
1.099
Temporal-energy deviation |flickerpred − flickerref| in 106 PU21-energy units on five clips (02c3 to 0237 are CHUG content IDs). Smaller is closer to the reference; bold marks the closest method per column. The reference row is zero by definition.
Reuse one bridge noise ε for every frame.
Smooth the spline over time, θ̄f = 0.8 θ̄f−1 + 0.2 θf.
Signed excess flicker falls 53.3%. Raw flicker stays within 3.27% of the reference.
Nothing is trained for video, so there are no temporal layers, optical flow or fine-tuning.
LumaFluxarXiv:2604.02787
Lifting generated video to HDR
Two of six LTX-Video clips. Right column: gray is at or below 100 nit, color is above it.
Text-to-video gives SDR; LumaFlux lifts it frame by frame.
Peaks of 739 to 957 nit, 7.4 to 9.6× the SDR ceiling.
5.82 s per frame, same checkpoint, no retraining.
LumaFluxarXiv:2604.02787
Limitations
The VAE is an SDR codec. The spline cannot restore what clipping removed.
Oversaturation. Mean ITP chroma is 60.65 against 49.80 for the references.
Native HDRTV1K. In-domain CNNs reach 38.11 dB where we reach 23.56.
Speed. 5.282 s per 1080p frame against 0.367 to 1.708 s single-pass.
Still ahead: a larger primary track and a controlled subjective study.
The LumaFlux sampler, with z1 = zSDR + σ ε, ε ~ N(0, I), σ = 0.05. Training interpolates zt = (1 − t) zHDR + t z1 and predicts the velocity z1 − zHDR. A text-to-image model starts at pure noise and needs many more steps.
LumaFluxarXiv:2604.02787
Physical conditioning and PGA in detail
What the model measures
Linear-light BT.2020 luminance Y = m2020T xlin, with m2020 = [0.2627, 0.6780, 0.0593].
Per token: Tphys = Conv3×3([Y, log(1 + |∇Y|), sat]).
Per frame: g = MLPg([μY, σY, p95, p99]) in ℝ16.
Spectrum: r = log mean energy of the FFT of Y in Kr = 8 radial bands.
LFM regresses the bridge velocity. The two L1 terms compare decoded output and target in linear light (inverse PQ at the 1,000-nit peak), on luminance and on RGB.
Lsmooth regularizes adjacent knot slopes of the spline.
Decoded losses run every fourth optimizer step on the one-step estimate ẑ0 = zt − t v̂t; the sampler is never unrolled.
About 36 hours on four GH200 GPUs, one per node, 144 GPU-hours.
A second, HDRTV1K-specific checkpoint trains for 50,000 steps, adapters only. It appears only on the legacy HDRTV1K table and the fixed-QP table, labeled in-domain.
HDR reconstruction loss against optimizer step. The full design converges to 29.2 dB; without the RQS decoder 19.5 dB; LoRA only 19.3 dB.
LumaFluxarXiv:2604.02787
Frame win rates on the synthetic track
Method
PSNR
SSIM
ΔEITP
LPIPS
VDP-3
HIDRO
BT.2446c
14.2
19.1
16.5
7.2
13.2
6.5
Reinhard operator
27.2
37.1
27.5
14.2
11.2
3.5
HDRTVNet++
6.2
11.1
11.0
1.2
6.0
17.0
FMNet
13.2
13.1
14.0
0.2
1.2
3.0
ITM-LUT
3.2
0.1
3.0
7.2
2.2
1.0
VAE + RQS (control)
7.0
4.1
2.0
1.0
4.0
1.0
LumaFlux (ours)
29.2
15.1
26.0
69.2
62.0
68.0
Share of aligned synthetic-track frames (%) on which each method is best, exact ties split equally. Higher is better. Bold best, underline second best.
LumaFlux wins the most PSNR frames, 29.2%.
The fixed Reinhard operator wins the most SSIM and ΔEITP frames, 37.1% and 27.5%, against 15.1% and 26.0% for LumaFlux.
On the perceptual metrics LumaFlux is best on 69.2% of frames for HDR-LPIPS, 62.0% for HDR-VDP-3 and 68.0% for FR-HIDROVQA.
LumaFluxarXiv:2604.02787
Native HDRTV1K pairs, where the in-domain CNNs win
Method
PU21-PSNR ↑
PU21-PSNR-Y ↑
PU21-SSIM ↑
ΔEITP ↓
HDR-LPIPS ↓
HDR-VDP-3 ↑
FR-HIDRO ↓
BT.2446c
25.00
24.89
0.9511
39.64
0.078
7.632
0.514
Reinhard operator
24.49
24.32
0.9342
43.97
0.083
7.637
0.485
HDRTVNet++
38.11
40.06
0.9824
8.13
0.043
8.224
0.338
FMNet
38.00
40.05
0.9822
8.10
0.047
8.188
0.358
KUNet
37.99
39.85
0.9820
7.84
0.038
8.235
0.258
LEDiff
10.99
11.28
0.5245
198.58
0.593
3.964
1.088
X2HDR
18.65
18.80
0.6584
92.20
0.445
5.071
0.874
PromptIR
23.18
23.02
0.8617
53.35
0.099
7.554
0.475
VAE + RQS (control)
21.47
23.30
0.9215
71.86
0.201
6.192
0.804
LumaFlux (ours, mixed-corpus)
23.56
27.71
0.9316
47.59
0.170
6.899
0.632
Native track: 117 HDRTV1K test pairs scored with the mixed-corpus checkpoint, which was never tuned on HDRTV1K. ITM-LUT has no result on this track.
CNNs trained on this set reach 38.11 dB. The mixed-corpus model, never tuned on HDRTV1K, reaches 23.56 dB, above X2HDR (18.65) and LEDiff (10.99). That is a distribution gap.
LumaFluxarXiv:2604.02787
The in-domain checkpoint on the legacy HDRTV1K suite
Method
PSNR
SSIM
SR-SIM
ΔEITP
HDR-VDP-3
HDRTVNet family
AGCM+LE+HG
37.21
0.9699
0.9968
9.11
8.569
AGCM-LE-HR++
38.36
0.9735
0.9975
8.28
8.751
AGCM-LE++, joint
38.60
0.9745
0.9973
7.67
8.696
HDCFM
38.42
0.9732
0.9974
7.83
8.572
FMNet
37.94
0.9747
0.9957
8.10
8.510
KUNet
37.78
0.9868
0.9971
7.80
8.393
Deep SR-ITM
37.10
0.9686
0.9950
9.24
8.233
ITM-LUT
36.69
0.9811
–
10.19
8.912
Beyond-FMG
35.06
0.9609
–
–
–
RealRep
29.12
0.9412
–
–
–
LumaFlux (in-domain, ours)
33.34
0.9427
0.9941
14.69
7.962
Baseline values as published under each author's protocol; "–" means not reported. The LumaFlux row is the HDRTV1K-specific checkpoint (50,000 steps, adapters only) under our 117-image PQ-code protocol. Higher is better except ΔEITP.
Two checkpoints, never in the same row
33.34 dB belongs to the in-domain checkpoint, with adapters retrained on HDRTV1K for 50,000 steps.
The main mixed-corpus model scores 27.52 dB zero-shot on the same 117 images under this protocol.
The published best is 38.60 dB, so even in domain we stay below the specialized CNNs.
Each baseline keeps its author's protocol, so cross-row ranking is protocol-sensitive. This table and the native-track table use different metric suites and are not interchangeable.
LumaFluxarXiv:2604.02787
Where the highlights go
Headroom maps of one LumaEval frame. Gray is at or below 100 nit; color marks pixels above the SDR ceiling. Percentages under the maps are for this frame.
Method
Share above 100 nit
IoU with reference highlights
Error inside highlights (stops)
Reference
32.3%
–
–
BT.2446c
19.1%
–
–
Reinhard
29.7%
0.563
–
HDRTVNet++
47.9%
0.538
0.81
LumaFlux
37.5%
0.579
0.87
100 synthetic-track frames. Share closer to the reference and higher IoU are better; lower error inside the reference highlights is better.
The analytic curve under-expands, HDRTVNet++ pushes too much of the frame bright, and LumaFlux puts highlights where the reference has them.
LumaFluxarXiv:2604.02787
Best in every column of the synthetic track (full table)
Method
PU21-PSNR ↑
PU21-PSNR-Y ↑
PU21-SSIM ↑
ΔEITP ↓
HDR-LPIPS ↓
HDR-VDP-3 ↑
FR-HIDRO ↓
BT.2446c
23.74
24.22
0.8291
61.60
0.307
5.611
0.753
Reinhard operator
23.62
24.25
0.8286
61.76
0.295
5.726
0.753
HDRTVNet++
23.03
23.23
0.8192
66.67
0.328
5.417
0.703
FMNet
22.86
22.97
0.8176
67.92
0.330
5.412
0.724
ITM-LUT
22.92
23.06
0.8142
67.41
0.315
5.496
0.761
VAE + RQS (control)
18.54
19.41
0.7647
111.67
0.381
3.673
0.924
LumaFlux (ours)
24.23
24.57
0.8294
56.54
0.269
5.812
0.631
Macro-averages over the same 100 frame IDs and 78 video-variant groups. Bold best, underline second best. FR-HIDRO is the HIDRO-VQA feature distance (Saini et al., WACV 2024).
24.23 dBPU21-PSNR, 0.49 dB above the strongest baseline
56.54ΔEITP, 5.06 lower than the strongest baseline
0.631FR-HIDROVQA, from 0.703 for HDRTVNet++
One checkpoint, no per-track tuning. Without diffusion transport the control collapses to 18.54 dB.
LumaFluxarXiv:2604.02787
On human-graded SDR, LumaFlux leads five of seven metrics (full table)
Method
PU21-PSNR ↑
PU21-PSNR-Y ↑
PU21-SSIM ↑
ΔEITP ↓
HDR-LPIPS ↓
HDR-VDP-3 ↑
FR-HIDRO ↓
BT.2446c
33.16
34.26
0.8442
37.83
0.148
7.397
0.660
Reinhard operator
31.48
32.24
0.8270
43.59
0.155
7.173
0.662
HDRTVNet++
31.53
32.05
0.8125
44.92
0.147
7.144
0.671
FMNet
31.64
32.11
0.8137
44.85
0.147
7.135
0.675
KUNet
31.51
31.96
0.8078
45.26
0.143
7.058
0.647
ITM-LUT
31.46
32.06
0.8085
45.49
0.142
7.090
0.639
LEDiff
18.30
18.99
0.6480
103.15
0.496
4.723
0.898
X2HDR
23.78
24.69
0.7578
51.28
0.366
5.532
0.758
PromptIR
29.21
29.67
0.6756
53.11
0.185
6.141
0.690
VAE + RQS (control)
26.05
27.33
0.7231
91.70
0.237
5.814
0.863
LumaFlux (ours)
33.31
35.17
0.8480
40.19
0.138
6.919
0.544
The 100 LIVE-TMHDR pairs whose SDR side is an expert grade, with the preregistered per-content exposure calibration. KUNet and PromptIR (86 of the 100 synthetic IDs) and LEDiff and X2HDR (five synthetic variants) are scored here and on the native track.
FR-HIDROVQA 0.544, 14.9% below the best baseline. BT.2446c keeps the lower color error, 37.83 against 40.19, and the higher HDR-VDP-3.
LumaFluxarXiv:2604.02787
Removing the spline costs 9.69 dB (full table)
Row
PGA
Gate
PCM
Coup.
T-mod
RQS
PSNR (Δ vs g)
ΔEITP (Δ vs g)
(a) LoRA only
•
19.29 (−9.90)
256.84 (+229.03)
(b) attention + MLP LoRA
•
•
28.36 (−0.83)
36.35 (+8.53)
(c)
•
•
•
28.57 (−0.62)
33.48 (+5.66)
(d)
•
•
•
•
29.56 (+0.37)
28.38 (+0.57)
(e)
•
•
•
•
•
29.20 (+0.01)
31.76 (+3.95)
(f) seed-0 repeat of (g)
•
•
•
•
•
•
28.67 (−0.52)
35.00 (+7.18)
(g) full design, ours
•
•
•
•
•
•
29.19 (ref.)
27.81 (ref.)
(h) linear spline
•
•
•
•
•
lin.
28.74 (−0.45)
32.16 (+4.35)
(i) no RQS
•
•
•
•
•
19.50 (−9.69)
261.78 (+233.97)
(j) no T-mod
•
•
•
•
•
28.38 (−0.81)
33.90 (+6.09)
Component ablations: 20k steps on a 55-video pool, the same 82 held-out frames. A dot marks a component that is on; Gate is the spectral gate, T-mod the timestep-and-layer modulation, lin. a linear spline. Higher PSNR and lower ΔEITP are better; deltas are relative to row (g).
No RQS, row (i), falls 9.69 dB.
A linear spline, row (h), costs 0.45 dB.
The spectral gate, (c) to (d), lifts 28.57 to 29.56 dB.
No timestep-layer modulation, row (j), costs 0.81 dB.
Rows (f) and (g) are the same configuration, 0.52 dB apart, so smaller gaps are noise.
PU21-PSNR change from QP 27 (QP is the HEVC quantization parameter) on the 117 HDRTV1K pairs. Flatter is better.
From QP 27 to 42 the mixed-corpus model loses 0.55 dB. HDRTVNet++ loses 2.77 and FMNet 2.71.
The CNNs still score higher at every QP on their home set.
Method
QP 27
QP 32
QP 37
QP 42
Δ
PU21-PSNR (dB), higher is better
HDRTVNet++
35.90
35.30
34.34
33.13
−2.77
FMNet
35.56
34.97
34.04
32.85
−2.71
LumaFlux (mixed-corpus)
27.29
27.27
27.04
26.75
−0.55
LumaFlux (in-domain)
33.94
33.67
33.12
32.33
−1.61
ΔEITP, lower is better
HDRTVNet++
10.41
11.18
12.47
13.98
3.58
FMNet
10.53
11.33
12.62
14.18
3.65
LumaFlux (mixed-corpus)
30.32
30.56
31.30
32.08
1.76
LumaFlux (in-domain)
13.64
14.09
15.01
16.20
2.56
HDR-VDP-3, higher is better
HDRTVNet++
8.409
8.167
7.759
7.198
−1.211
FMNet
8.329
8.097
7.704
7.156
−1.173
LumaFlux (mixed-corpus)
7.615
7.525
7.308
6.948
−0.667
LumaFlux (in-domain)
7.918
7.791
7.533
7.117
−0.802
Fixed-QP track, 117 HDRTV1K pairs. Δ is QP 42 minus QP 27; closer to zero is better. Mixed-corpus is the main checkpoint; in-domain is the HDRTV1K-specific one. Bold best, underline second per column and block.
LumaFluxarXiv:2604.02787
What it costs (full table)
5.28 sper 1080p frame, 8 steps, one GH200
27.06 GBpeak memory at 1080p
71.32Mtrainable of 12.48B parameters
FR-HIDROVQA (lower is better) on 30 expert-track frames. LumaFlux 0.431, 0.477 and 0.567 at 540p, 720p and 1080p, in 2.64, 3.08 and 5.29 s.
Method
Steps
Params (M)
Trainable (M)
s/frame
TFLOPs
Mem. (GB)
BT.2446c
1
0.00
0.00
0.415
–
–
HDRTVNet++
1
0.59
0.59
0.857
0.5
2.31
FMNet
1
1.30
1.30
0.640
1.5
2.16
KUNet
1
2.16
2.16
0.557
3.7
3.76
ITM-LUT*
1
0.60
0.60
0.367
–
0.18
PromptIR + Reinhard
1
35.59
35.59
1.708
10.9
17.86
LEDiff (5122)
50
1370.23
859.53
4.449
149
6.37
X2HDR (5122)
30
17020.02
149.42
4.244
1,023
33.77
LumaFlux, 8 steps
8
12484.77
71.32
5.282
2,067
27.06
LumaFlux, 40 steps
40
12484.77
71.32
24.993
10,100
27.06
One GH200, batch 1, complete 1080p exports; LEDiff and X2HDR at their registered 5122 protocol (scaled by pixel count to 1080p: 35.19 and 33.57 s). *ITM-LUT runs a compiled kernel the tracer cannot observe. Lower is better for s/frame, TFLOPs and memory.
2.4 to 16.1× faster than X2HDR and LEDiff. HDRTVNet++ stays 173 to 201× faster than us.
LumaFluxarXiv:2604.02787
Video without temporal training (full table)
Per-frame mean PU21 luma, four clips, first 120 frames, relative to each reference mean. Only the stabilized run is drawn.
Method
Bonfire
02c3
055e
0392
0237
Mean
Reference
0.000
0.000
0.000
0.000
0.000
0.000
BT.2446c
7.813
1.411
0.450
0.063
0.124
1.972
HDRTVNet++
3.005
0.207
0.147
0.442
1.257
1.012
FMNet
3.853
0.202
0.074
0.579
1.124
1.166
LumaFlux
1.992
0.691
2.500
0.968
0.569
1.344
LumaFlux + stabilization
2.109
0.469
1.676
0.879
0.362
1.099
Temporal-energy deviation |flickerpred − flickerref| in 106 PU21-energy units on five clips (02c3 to 0237 are CHUG content IDs). Smaller is closer to the reference; bold marks the closest method per column.
Reuse one bridge noise ε for every frame.
Smooth the spline over time, θ̄f = 0.8 θ̄f−1 + 0.2 θf.
Signed excess flicker falls 53.3%. Raw flicker stays within 3.27% of the reference.
Nothing is trained for video, so there are no temporal layers, optical flow or fine-tuning.
LumaFluxarXiv:2604.02787
Three scenes enlarged, every HDR method
The red box on each SDR frame marks the enlarged region, all under one fixed display transform. HDR shown tone-mapped.
The flame row shows the chroma overshoot that LumaFlux still has.
LumaFluxarXiv:2604.02787
The same regions in log luminance
Every method on one log scale from 0.1 to 1,000 nit. SDR uses a 100-nit display white; LumaFlux outlined.
Selected examples, not an aggregate ranking.
LumaFluxarXiv:2604.02787
Luminance along one row, two scenes
Mean absolute log deviation from the reference: BT.2446c, HDRTVNet++, LumaFlux = 0.190, 0.166, 0.058 (City towers); 0.272, 0.171, 0.078 (Sunset forest).
LumaFluxarXiv:2604.02787
Text-to-HDR keyframes over time
Each panel splits into the frame and its log-luminance map, 0.1 to 1,000 nit. SDR maps saturate at the 100-nit ceiling.
LumaFluxarXiv:2604.02787
Three more text-to-HDR lifts
Each lift at its keyframe with the largest area above 100 nit. Gray is at or below 100 nit, color above.
Of the pixels above 100 nit, 90.4% to 98.3% lie among the brightest source pixels.
LumaFluxarXiv:2604.02787
Training evolution
Reconstruction at 5k to 50k steps; per-pixel ΔEITP against the HDR reference falls as training runs.
LumaFluxarXiv:2604.02787
Frozen-VAE round trips, PQ against PU21
Six LIVE-TMHDR crops through the frozen VAE. Lower right of each round trip is its per-pixel ΔEITP.
LumaFluxarXiv:2604.02787
One crop at three display exposures
One stored crop at −4, 0 and +4 EV of display exposure, with one shared display mapping.
Dropping exposure shows what each method put above the SDR range.
LumaFluxarXiv:2604.02787
Tone variants of one content stay close inside the frozen model
Cosine similarity within one content's tone variants, against between contents, for VAE tokens and velocity at t = 0.25, 0.5, 0.75.
The within-content gap is positive in all four panels.