Generative Vision Atlas

Which approach is actually better?

The honest answer is that the published numbers do not support a clean ranking, and the reason is worth understanding before reading any table below.

What you can conclude

What you cannot conclude

ImageNet 256, with guidance

Lower is better. The guidance column is why this table is not a leaderboard.

gFIDPaperLineGuidanceBudgetParams
1.03PAESemantic-plus-detail hybridsnot stated in abstract13x faster convergence claimednot stated
1.06RAEv2Representation-space generationREPA-Guidance80 epochs839M
1.12REPA-ERepresentation-aligned latentsCFG (scale not stated)800 epochs675M
1.13RAERepresentation-space generationAutoGuidance800 epochs839M
1.35VA-VAE / LightningDiTRepresentation-aligned latentscfg-interval + timestep shift (scale not stated)800 epochs675M
1.42REPARepresentation-aligned latentsguidance-interval800 epochs (~4M iters)675M
1.55MARContinuous-token autoregressionCFG (scale swept, value not stated)800 epochs943M
1.67MAETokRepresentation-aligned latentscfg=2.0, 250 steps4M steps675M generator + 176M tokenizer
1.73VARDiscrete-token autoregressionguidance ratio 2.0 + rejection sampling350 epochs2.0B
1.8FlatDINOSemantic-plus-detail hybridsnot stated in abstractnot stated675M (DiT-XL)
1.82JiTSingle-stage pixel transformersCFG-interval600 epochs~2B
1.92SVGRepresentation-space generationCFG (scale unresolved in source)1400 epochs675M
2.06SiTFlow matching and rectified flowcfg=1.57M steps675M
2.27DiTVAE-latent diffusioncfg=1.57M steps675M

PAE: Taken from the abstract; the guidance method and budget behind the headline are not stated there. Needs a PDF-level check before being treated as directly comparable.

RAEv2: The paper's own Table 4 shows the identical model at 1.65 (no guidance), 1.49 (CFG), 1.14 (AutoGuidance) and 1.06 (REPA-Guidance). Guidance method alone moves this number by more than 0.5 FID, which is larger than most claimed architectural wins in this table.

RAE: AutoGuidance is a stronger guidance method than the plain CFG used by the VAE baselines in most cross-paper comparisons, and contributes a large share of the gap.

VA-VAE / LightningDiT: The widely quoted '21x faster than DiT' refers to the 64-epoch checkpoint (gFID 2.11), not to this 800-epoch headline number. The speed claim and the quality claim come from different rows.

REPA: Headline uses guidance-interval, not the plain CFG its DiT/SiT baselines use. With plain CFG (w=1.35) the same model reports 1.80.

VAR: arXiv v1 reported 1.80 for this configuration and the current version reports 1.73; cite which version you mean. Uses rejection sampling, which the diffusion baselines do not.

FlatDINO: Headline claim is efficiency (8x fewer FLOPs per forward pass), not raw FID.

JiT: Uses roughly 2B parameters against the 675-839M of the latent-space rows, and a guidance-interval technique. No unguided number is reported, so this cannot be compared against the unguided column. Any claim that pixel-space now beats latent-space is confounded by scale and guidance simultaneously.

SVG: Trained far longer than most rows here (1400 epochs). At a comparable 500 epochs it reports 2.10. The source gives conflicting CFG scales (4.0 in a figure caption, 1.5 in an appendix table).

ImageNet 256, without guidance

The cleaner comparison, because it removes the single largest confound. Fewer papers report it, which is itself telling.

gFIDPaperLineBudgetParams
1.51RAERepresentation-space generation800 epochs839M
1.65RAEv2Representation-space generation80 epochs839M
1.69REPA-ERepresentation-aligned latents800 epochs675M
2.17VA-VAE / LightningDiTRepresentation-aligned latents800 epochs675M
2.31MAETokRepresentation-aligned latents4M steps675M
2.35MARContinuous-token autoregression800 epochs943M
3.36SVGRepresentation-space generation1400 epochs675M
5.9REPARepresentation-aligned latents4M iters675M
8.3SiTFlow matching and rectified flow7M steps675M
9.62DiTVAE-latent diffusion7M steps675M

Note what happens here. The gap between a plain VAE-latent transformer and a representation latent is enormous without guidance and much smaller with it. Guidance is doing a large share of the work that headline numbers attribute to the latent.

One-step and few-step

A separate contest. These models sample in one or a few passes rather than dozens, so they are not comparable to the tables above.

gFIDNFEPaperLineNote
1.481Distilling Drifting TransformersNatively few-step objectivesDiT^DH-XL backbone in RAE latent, one-step distilled
1.541Drifting ModelsNatively few-step objectivesImageNet 256x256, latent space, one-step
1.611Drifting ModelsNatively few-step objectivesImageNet 256x256, pixel space, one-step
1.998IMMNatively few-step objectivesImageNet 256x256, 8 steps
3.431MeanFlowNatively few-step objectivesImageNet 256x256, one-step, trained from scratch

The result worth pausing on: one method reports 1.54 in latent space and 1.61 in pixel space with everything else held fixed. A 0.07 gap is weak support for the latent being essential.

Reconstruction quality of the latent itself

No generator involved. This is the ceiling the tokenizer imposes, and the point is that it does not predict the generation numbers above.

rFIDTokenizerPaper
0.16RAE with frozen MAE-B encoderRAE
0.28VA-VAE f16d32VA-VAE / LightningDiT
0.48MAETok, 128 tokensMAETok
0.49RAE with frozen DINOv2-B encoderRAE
0.62SD-VAE f8d4, as re-measured by MAETokLDM / Stable Diffusion
0.65Frozen DINOv3 + residual detail branchSVG
0.69DC-AE f32c32DC-AE

RAE: MAE-B gives RAE's best reconstruction but not its best generation — a compact illustration that rFID does not predict gFID.

VA-VAE / LightningDiT: The strongest cross-paper agreement in this table: three independent papers report the same value.

DC-AE: DC-AE reports no ImageNet 256x256 gFID at all; its generation results are 512x512 only, and its headline 'beats SD-VAE' result there uses 4x the training iterations.

Text-to-image

Higher is better. Read these as coarse bands, not rankings: the same checkpoint scores differently across papers, and the benchmark itself has measured drift.

GenEvalSystemLineParams
0.91Qwen-Image Qwen-Image after RL tuningVAE-latent diffusion20B
0.87Qwen-Image Qwen-Image base modelVAE-latent diffusion20B MMDiT + Qwen2.5-VL 7B encoder
0.75SVG-T2I SVG-T2I 2.6B DiT, frozen DINOv3 latentRepresentation-space generation2.6B
0.74SD3 / MMDiT SD3 8B (depth 38) at 1024px, with DPOFlow matching and rectified flow8B + 4.7B T5-XXL
0.73Lumina-Image 2.0 Lumina-Image 2.0VAE-latent diffusion2.6B
0.495Scale-RAE Best ablation config, deliberately budget-limited (~30K iterations)Representation-space generation0.5B-9.8B swept

Qwen-Image: Post-RL. Illustrates that RL alignment moves GenEval by ~0.04, comparable to the gap between several distinct systems in this table.

SVG-T2I: Reported with an LLM prompt rewriter, which is not comparable to scores measured on raw prompts. The same paper reports 0.82 for FLUX.1-dev under the same rewriter, where other papers report 0.66-0.67 for that identical checkpoint.

SD3 / MMDiT: SD3's own paper reports neither DPG-Bench nor MJHQ-30K; any such number attributed to SD3 is a third-party re-measurement.

Scale-RAE: Scale-RAE is a controlled scaling study, not a system release. Reading this as a state-of-the-art claim is a category error. The legitimate comparison is its within-paper ablation at matched budget: RAE 49.5 versus a VAE baseline at 39.6.

DPG-BenchSystemParams
88.32Qwen-Image20B
87.2Lumina-Image 2.02.6B
85.78SVG-T2I2.6B
76.9Scale-RAE

How to read a claim in this field

When a paper says it beats a baseline, check three columns before believing the headline: the guidance method, the training budget, and the parameter count. If any of the three differs from the baseline, the gap is not attributable to the idea being sold. The within-paper ablation, where those are held fixed, is almost always the more informative number, and it is almost always smaller than the abstract implies.