Generative Vision Atlas

Research line · superseded

Cascaded and multiscale pixel diffusion

Keep generation in pixels, but handle resolution progressively: generate small, then refine upward.

What defines membership

Pixels were never the problem. Resolution was. Manage resolution with a cascade or a shifted schedule and pixel diffusion matches latent diffusion, with no tokenizer needed.

The bet

This line agrees with single-stage pixel transformers that no learned tokenizer is necessary, and disagrees about why pixel generation was hard. The diagnosis here is not the prediction target but the resolution: model a small image well, then refine upward, and pixel diffusion comes within a fraction of a FID point of latent diffusion with no VAE anywhere in the pipeline.

The arc

Cascaded Diffusion Models sets the pattern in 2021 — a chain of pixel-space models, each super-resolving the last, with conditioning augmentation to stop errors compounding along the chain. simple diffusion then asks whether the chain is avoidable: with a shifted noise schedule and capacity concentrated at low resolution, can one end-to-end model be comparable to a cascade and to latent diffusion? It argues comparability. It does not claim cascades are unnecessary, and this atlas previously overstated it as doing so.

HDiT supplies the architecture that makes single-stage pixel training practical, a hierarchical transformer whose cost scales as O(n) rather than the O(n²) of a standard DiT, by restricting global attention to the coarsest level and using neighbourhood attention elsewhere. It trains natively at 1024×1024 on FFHQ; its ImageNet results are at 256×256 and are competitive rather than state of the art.

SiD2 is the result that should change how the current wave is read, though it is usually reported less carefully than it deserves. In October 2024 it reached FID 1.48 on ImageNet 512 in pixel space, at 512 steps with guidance scale 1.0 over the interval (-3, 5). It is often cited as matching latent diffusion. It does not: in the paper’s own matched-setting table EDM2-XXL with a guidance interval reaches 1.40, and the authors concede that latent diffusion still has slightly better scaling properties. What is true, and sufficient, is that pixel-space generation was within 0.08 FID of the best latent model a year before the tokenizer-free papers argued the case — and that no pixel-space paper in this atlas has beaten 1.48 at 512 since.

Edify Image shipped industrial-scale pixel-space generation the following month using Laplacian frequency bands, though it reports no quantitative benchmark, so it cannot be placed against these numbers. PixelFlow carried the multiscale bet across to flow matching in 2025.

Why this line is marked superseded

Not because anything in it was refuted. Nothing here was. It is superseded in attention: the field moved to single-stage transformers, and the cascaded literature stopped accumulating. The honest reading is that a cascade is more moving parts than the tokenizer it was meant to replace, that errors compound between stages and need explicit machinery to control, and that most of this work predates or sits outside the transformer scaling story the field now expects a generative method to have.

Why it still matters

It is the strongest available evidence that the 2025–26 tokenizer-free wave rediscovered something rather than discovered it. The argument that pixels were never the real obstacle, and that a well-designed pixel model comes within a fraction of a FID point of the best latent model, was made here with numbers years earlier. Note the careful wording: this line got close and said so honestly. It is the current wave, not this one, that reaches for the word parity.

That is worth stating plainly, because it is the kind of thing an atlas organised around papers rather than ideas would miss entirely. These papers do not cite the current wave and the current wave does not always cite them, so nothing in the citation graph connects the two. The connection is in the claim, not the references.

How the line developed

Read top to bottom: what came before, the idea itself, the evidence for it, what improved, and where it breaks.

The idea

CDM · 2021-05landmark

A chain of pixel-space diffusion models with conditioning augmentation to keep errors from compounding between stages.

Evidence

SiD2 · 2024-10landmark

FID 1.48 on ImageNet 512 in pixel space, at 512 steps with guidance 1.0 over the interval (-3, 5). Narrowly behind EDM2-XXL's 1.40 in the paper's own matched table, and the paper concedes latent diffusion scales slightly better, but a year ahead of the tokenizer-free wave and still the best pixel-space number at 512 in this atlas.

Improvement

simple diffusion · 2023-01core

Asks whether one end-to-end model, with a shifted noise schedule and capacity concentrated at low resolution, can be comparable to cascades and to latent diffusion. It argues comparability rather than claiming cascades are unnecessary.

HDiT · 2024-01core

A hierarchical transformer with linear rather than quadratic cost in pixels, making native megapixel pixel training feasible. The architectural bridge to the single-stage line.

PixelFlow · 2025-04core

Carries the multiscale bet to flow matching, ImageNet 256 FID 1.98 with no VAE.

At scale

Edify Image · 2024-11core

Laplacian frequency-band pixel diffusion shipped at industrial scale while the open literature treated latents as settled. Reports no quantitative benchmark.

What it gets right

  • Got pixel-space within 0.08 FID of the best latent model at ImageNet 512 in 2024, years before the current wave, and was largely ignored
  • Progressive resolution sidesteps the attention cost that constrains single-stage pixel transformers
  • Sustained by well-resourced industrial groups rather than by a single result

Where it is weak

  • A cascade is more moving parts than a tokenizer, which is what it was meant to avoid
  • Errors compound between stages, needing conditioning augmentation to control
  • Largely UNet-era work, so it did not inherit the transformer scaling story the field now expects
  • Superseded by the single-stage transformer line in attention rather than in evidence: no paper here was refuted, and SiD2 still holds one of the best pixel-space numbers

Competing answers

Open problems it has not solved