We describe our entry to the LoViF 2026 AIGC Image Compression Challenge, a benchmark for ultra-low-bitrate coding of AI-generated images under a strict global rate budget of 0.025 bits per pixel (BPP). Generated imagery poses a distinct challenge for compression: it frequently contains rendered typography, synthetic edges, repeated motifs, UI-like layout, and stylized micro-texture that conventional distortion-oriented codecs erase at this rate, while unconstrained generative decoders can restore plausible-looking detail that no longer matches the source geometry or symbols. We treat this as a rate-perception allocation problem. Our system fine-tunes four rate-specialized checkpoints of the AEIC one-step diffusion codec, generates per-image candidates from all four, including one latent refined through encoder-side test-time optimization (TTO) with periodic entropy-conditioning refresh, entropy-codes every candidate with practical rANS coding, and selects exactly one bitstream per image with an exact multiple-choice knapsack solved over true coded file sizes. A fixed, zero-additional-bit residual restoration network is applied at decode time. Every submitted bitstream is independently decodable by the shipped decoder, which uses no source image or external side information. We report the full pipeline, an ablation history spanning 77 logged experiments, and a set of negative results, including why PSNR could not be pushed to parity with rate-distortion-oriented competitors under this architecture, useful to future participants. This is a challenge report: our entry scored 31.527739 (PSNR 27.02 dB, MS-SSIM 0.9176, LPIPS 0.0778, DISTS 0.0390 at 0.02495 BPP), the second-best DISTS on the leaderboard, ranking 5th on the final test-phase leaderboard announced August 4, 2026.
Figures & tables
Figure 1: Final compression pipeline. Dashed flow denotes encoder-only access to the source for candidate evaluation and rate allocation; the submitted decoder receives only the selected bitstream and reconstructs without any source access.
Checkpoint
Init
Steps
Corpus
Target rate
aigc8_5000
ft8 (zero-shot, λ8 )
5k
train-800 8× + initial synth.
0.020 BPP
r2_18000
aigc4_5000 ( λ4 )
16k+
train-800 8× + synth.
0.037 BPP
r3l4_8000
r2_18000
8k
train-800 8× + SD3.5/SDXL synth.
0.037 BPP
r3l8_8000
aigc8_5000
8k
train-800 8× + SD3.5 synth.
0.026 BPP
Table 1: Training configuration of the four shipped checkpoints. All use AEIC’s codec loss λRR+λ2∥x−x^∥22+λDLDISTS+λCLCLIP+λALadv+λLLLPIPS . Exact weights (λR,λ2,λD,λC,λA,λL) are (8,2,1,.1,.1,1) for aigc8_5000; (4,2,1.5,.1,.1,2) for r2_18000 and r3l4_8000; and (8,2,1.5,.1,.1,2) for r3l8_8000. These values are retained in the four released training configuration files. Target rate is the checkpoint’s measured natural operating region, not an imposed per-image constraint.
Raw checkpoint
Restorer change
Final
Checkpoint
BPP
PSNR
MS-SSIM
LPIPS
DISTS
Δ PSNR
Δ MS-SSIM
Δ LPIPS
Δ DISTS
selected
aigc8_5000
.019739
25.6427
.891992
.112686
.063923
+.1848
+.000272
-.001876
-.001634
20
r2_18000
.036242
27.0023
.921605
.082632
.049385
+.2166
+.000084
-.001581
-.001376
12
r3l4_8000
.036478
27.0471
.922338
.081227
.048786
+.2114
+.000063
-.001420
-.001200
40
r3l8_8000
.025387
26.3583
.907997
.091023
.052620
+.1948
+.000164
-.001377
-.001219
28
Table 2: Per-checkpoint full-test results before global selection, retained from the actual coded candidates. These are descriptive measurements of frozen, already-generated challenge candidates and were not used to update checkpoint weights. Restorer columns are restored-minus-raw metric changes (negative is better for LPIPS/DISTS); BPP is unchanged. “Final selected” counts the checkpoint-ID bytes in the 100 submitted streams, including streams whose latent was refined by TTO.
Checkpoint
Nominal rate
Score
BPP
ft8 (zero-shot)
low
106.19
0.017
ft4 (zero-shot)
mid
109.21
0.0239
ft2 (zero-shot)
high
111.28
0.031
Early zero-shot mix (mean-BPP solver)
–
109.55
0.0250
Table 3: Zero-shot AEIC-ME rate ladder on the challenge validation set, before any AIGC fine-tuning, TTO, or restoration ( S scale, i.e. leaderboard scale +80 ; see Sec. 4.3 ). A single checkpoint at the legal rate already trails a per-image mix of unrelated rate points, motivating the checkpoint-bank design used throughout the rest of the system. Under the same early (mean-BPP-constrained, later superseded) solver, adding three fine-tuned checkpoints to this pool raised the mix from 109.55 to 109.89 (+0.34); this same-methodology signal fine-tuning was worth adding, not comparable to Tab. 6 ’s later rows.
Quantity
Final test entry
Official score ( B=S−80 )
31.527739
PSNR
27.022356 dB
MS-SSIM
0.917594
LPIPS
0.077794
DISTS
0.038970
Average BPP
0.024946
Table 4: Official organizer-recalculated final-test metrics for the submitted entry.
Rank
Team
Score
BPP
PSNR
MS-SSIM
LPIPS
DISTS
1
Ashes
34.004381
0.025000
29.820286
0.930888
0.076335
0.051785
2
rrrrty
33.557593
0.024987
28.926965
0.934522
0.071049
0.046815
3
NeFIC++
33.271115
0.024153
28.601395
0.932167
0.071724
0.044575
4
Freedom
32.849392
0.025000
28.779010
0.933339
0.077461
0.054115
5
ZeroR
31.527739
0.024946
27.022356
0.917594
0.077794
0.038970
6
prophet1
30.349026
0.024962
26.303701
0.908165
0.087057
0.038851
Table 5: Official final test-phase leaderboard, organizer-recalculated and announced August 4, 2026, at full precision. Our DISTS (0.038970) is second-lowest overall (only prophet1’s 0.038851 is lower) and lowest among every entry ranked above us; our LPIPS is not a leadership position. PSNR trails the top four by 1.6–2.8 dB ( Sec. 5 ).
Figure 2: Subjective stage comparison on hand-selected test image 000009; bottom row enlarges the same poster region. The fine-tuned decode is 0.01774 BPP; TTO produces the selected 0.01885-BPP stream; restoration adds no bits, and the last column is the submitted reconstruction. The comparison isolates visible stage effects, not aggregate performance; see Sec. 4.6 for full-set evidence.
Candidate pool (cumulative)
S
BPP
Status
Early 11-way mixed pool (v3)
109.94
0.0250
confirmed
+ expanded fine-tuning, 300-step TTO (v17)
111.53
0.0250
confirmed
+ refresh, MILP pool (v20)
111.66
0.0250
confirmed
+ 500-step TTO (v21)
111.83
0.0250
dev-set est.
Matched pool without TTO
111.18
0.0250
dev-set ablation
Matched pool with TTO
111.51
0.0250
dev-set ablation
Table 6: Legal-rate ( ≤0.025 BPP) results. Progression and matched-ablation rows use validation; the last row is the official test set and is not directly comparable across datasets. In particular, v20 was not evaluated on test, so v20 → v21 transfer cannot be inferred from the official test score. The matched TTO comparison uses the same four checkpoints, restorer, and MILP allocation, differing only in the availability of TTO candidates.
Diffusion-based image compression has achieved strong perceptual quality at ultra-low bitrates. However, existing codecs are often tied to specific backbones and specialized components, making diverse, rapidly evolving generative models difficult to reuse. This raises a natural question: Can modern generative foundation models be connected to image compression through a simple and extensible interface? Two insights guide our design: stronger generative priors make a simpler codec interface viable, and generation and compression can be intrinsically linked through latent transport. We therefore propose Gen2-IC with two stages: (1) Latent Compression maps clean image latents to entropy-constrained latents; and (2) Latent Transport refines them with one near-terminal update based on the pretrained model. Gen2-IC requires neither auxiliary conditioning signals nor task-specific backbone modifications. With lightweight adaptation and no distillation, it supports fast encoding and one-step decoding across multiple bitrates. We validate Gen2-IC on SD-2.1, SANA-1.5, FLUX.1-dev, and Qwen-Image-2512, spanning U-Net and Transformer architectures as well as diffusion and flow-matching formulations. With stronger priors, Gen2-IC delivers gains below 0.05 bpp: the Qwen variant leads diffusion-based generative codecs in reconstruction fidelity (PSNR), perceptual similarity (LPIPS and DISTS), and recognizer-based semantic fidelity (OCR CER/WER and face-ROI similarity) across four benchmarks.
Pretrained diffusion models provide powerful generative priors for extreme image compression, but their stochastic, iterative generation processes are not designed to decode compressed representations directly. We present PrismCodec, which recasts a pretrained Diffusion Transformer (DiT) as a bitstream-grounded one-step generative decoder. PrismCodec uses the decoded representation in two complementary roles: a projected codec state defines the starting state for generation, while hierarchical visual priors derived from the same decoded latent provide semantic and structural guidance. To bridge the gap between stochastic, iterative generation and deterministic one-step decoding, we develop a progressive adaptation strategy. We first learn codec-conditioned transport from a noisy neighborhood centered on the codec state, distill it into a one-step mapping, and finally anneal the source perturbation to zero during end-to-end optimization. The resulting decoder reconstructs deterministically from the received representation without additional conditioning bits. Across Kodak, CLIC, and DIV2K, PrismCodec consistently improves rate-perception performance over prior generative codecs, achieving BD-rate reductions of 70-75% in LPIPS and 75-81% in DISTS over StableCodec. Code will be released upon acceptance.
Yunuo Chen, Jingnan Gao, Bing He +6
Shanghai Jiao Tong University · University of Sydney
Generative image compression produces visually plausible reconstructions at low bitrates, yet their behavior as the rate approaches zero remains largely unexplored. When pushed below normal operating rates, representative codecs undergo semantic collapse: rather than gracefully losing source-specific detail, they produce malformed or unrecognizable content. Our analysis identifies two factors. As the bitrate decreases, reconstruction losses increasingly conflict with semantic objectives on gradients and visual results, while pixel-space and reconstruction-oriented VAE diffusion models become less efficient on semantic preservation. Guided by these findings, we introduce RAE-CoD, a compression-oriented diffusion (CoD) built in a representation autoencoder (RAE) space with direct alignment between compressed and source representations, preserving recognizable, naturally structured content for a 256×256 image with as few as 16 bits. We evaluate this framework using five vision foundation models (VFM) and a blinded vision-language model protocol. On MSCOCO-30K, RAE-CoD stands out from all evaluation. At 0.001-0.008 bpp, it reduces relative VFM feature MSE and Fréchet Distance ratio by at least 25.7% and 69.1% over the best competitors. Meanwhile, semantic recognizability and quality of the reconstructions remain nearly constant while source consistency falls smoothly, replacing abrupt semantic collapse with a graceful transition toward unconditional generation. Code will be released at https://github.com/LuizScarlet/RAE-CoD.