Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

A 0.4B diffusion transformer matches 20x larger models and runs on a phone

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 10:51 UTC pith:6YZX535J

load-bearing objection Solid on-device DiT systems paper; the 'surpasses 20×' claim overstates its own Table 2. the 3 major comments →

arxiv 2601.08303 v3 pith:6YZX535J submitted 2026-01-13 cs.CV

SnapGen++: Unleashing Diffusion Transformers for Efficient High-Fidelity Image Generation on Edge Devices

Dongting Hu , Aarush Gupta , Magzhan Gabidolla , Arpit Sahni , Huseyin Coskun , Yanyu Li , Yerlan Idelbayev , Ahsan Mahmood
show 7 more authors
Aleksei Lebedev Dishani Lahiri Anujraaj Goyal Ju Hu Mingming Gong Sergey Tulyakov Anil Kag
This is my paper
classification cs.CV
keywords diffusion transformerson-device image generationsparse attentionelastic networksstep distillationmobile deploymenttext-to-image
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper claims that transformer-based diffusion models, not just U-Nets, can be made to run efficiently on phones without losing the quality advantage transformers bring. It proposes a compact three-stage DiT with adaptive sparse attention, an elastic training scheme that lets one model serve several hardware tiers, and a step-distillation method that cuts sampling to four steps. If the claims hold, a 0.4-billion-parameter model can generate 1024-square images in about 1.8 seconds on a phone, matching or beating models up to twenty times larger on standard text-to-image benchmarks.

Core claim

The paper's central claim is that a 0.4B-parameter diffusion transformer, running entirely on a mobile device, can match the generation quality of server-scale models with up to 20× more parameters. This is achieved by combining three components: an architecture that replaces full self-attention with an adaptive mix of compressed global attention and blockwise neighborhood attention; an elastic training framework that shares weights across sub-networks of different widths; and a knowledge-guided distribution matching distillation that compresses the sampling process to four steps. The authors support this with benchmark scores (DPG-Bench, GenEval, T2I-CompBench, CLIP), a user study, and on-d

What carries the argument

The load-bearing object is the Adaptive Sparse Self-Attention (ASSA) layer, which splits attention into a coarse global branch (key/value features compressed by a strided 2×2 convolution) and a fine local branch (blockwise neighborhood attention with a small number of blocks and radius), adaptively interpolating between the two per head. Around it sits a three-stage transformer (down/middle/up) with token downsampling in the middle, elastic width-sliced supernetwork training, and K-DMD, a distillation objective that adds output-level and feature-level supervision from a few-step teacher to the standard DMD loss.

Load-bearing premise

The architecture is tuned against ImageNet validation loss at 256 resolution as a proxy for perceptual quality and human preference, and the authors assume this proxy is reliable and that the conclusions transfer to 1024-resolution text-to-image generation.

What would settle it

Run a controlled user study with the same prompts used in the paper, pitting the 0.4B on-device model against a server-scale model (e.g., 12B) and check whether the claimed win in realism, fidelity, or alignment holds; alternatively, evaluate the 4-step distilled 0.4B model on a larger, more diverse prompt set and see whether its GenEval and DPG scores collapse.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • On-device generation at this quality and latency would make text-to-image features practical in consumer apps without cloud round-trips.
  • The elastic training result implies one trained model can serve phones, tablets, and servers, replacing per-device fine-tuning.
  • The 4-step K-DMD result suggests few-step distillation can be applied successfully to small models, not just large ones.
  • The architectural ablations indicate that sparse attention with global and local branches can recover most of the quality of full attention while cutting latency and memory.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the validation-loss proxy transfers to human preference, the ablation-driven design choices (ASSA, layer distribution, FFN expansion) likely survive scale-up to larger datasets and resolutions; however, that transfer is not demonstrated here.
  • The comparison to server models relies on benchmarks at 1024 resolution; a direct side-by-side human study with more prompts and diverse devices would show how far the quality claim generalizes.
  • The elastic framework could be extended to also slice depth and attention heads, not just width, potentially yielding even finer cost-quality trade-offs.
  • Because the K-DMD teacher is already a few-step model, the method may extend to future one-step or few-step models, possibly enabling real-time generation with fewer than four steps.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes SnapGen++, a family of efficient diffusion transformers for mobile/edge deployment. Three contributions are claimed: (i) a three-stage DiT architecture with Adaptive Sparse Self-Attention (ASSA), combining compressed global attention and blockwise neighborhood attention; (ii) an elastic training framework that jointly optimizes subnetworks of different widths from one supernetwork; and (iii) Knowledge-guided Distribution Matching Distillation (K-DMD), a step-distillation pipeline that adds a few-step teacher to the DMD objective. The headline claim is that a 0.4B, 4-step variant 'surpasses models up to 20× larger' while running in about 1.8 s on an iPhone 16 Pro Max, and that a 1.6B full variant approaches server-level T2I quality. Experiments include ImageNet-1K validation-loss ablations, T2I benchmarks (DPG-Bench, GenEval, T2I-CompBench, CLIP-Score), latency measurements, a human preference study, and qualitative comparisons.

Significance. The paper addresses a practically important problem — high-fidelity image generation on edge devices — and the proposed components are well-motivated and plausible. The architectural ablations are quantified, and the reported on-device latencies are concrete and specific. If the central comparative claim is correct, the work would be a meaningful step toward making DiT-based T2I generation practical on phones. The elastic training framework and the few-step distillation pipeline also offer useful recipes beyond the specific model. However, the headline comparison is currently undermined by an evaluation-protocol inconsistency in Table 2, and the architecture ablations rest on a single proxy metric whose correlation with perceptual quality is asserted rather than demonstrated. The paper does not release code or evaluation scripts, which would substantially increase confidence in the numerical claims.

major comments (3)
  1. [Table 2 and Sec. 4.2] The footnote to Table 2 states that '† indicates out-of-memory (OOM) at 1024×1024 resolution,' yet most baseline rows carrying that mark (PixArt-Σ, SANA, SD3-Medium, SD3.5-Large, Flux.1-dev, etc.) report DPG-Bench, GenEval, T2I-CompBench, and CLIP scores. If those models cannot run at 1024×1024 in the stated measurement setup, the numbers must come from a different evaluation protocol (e.g., lower resolution, different sampling steps/CFG, or original papers). T2I benchmarks are resolution-sensitive, so mixing protocols makes the comparison uncontrolled. This matters directly for the central claim 'surpasses models up to 20× larger': the margin over SD3.5-Large is 0.4 DPG points, 0.01 GenEval, and 0.001 T2I-CompBench, so a small protocol mismatch could flip the result. Please specify exactly how each OOM row was evaluated, or separate the 'as-measured-by-us' scores from literature scores
  2. [Sec. 3.1, Fig. 3] Every architecture decision — ASSA, three-stage layout, FFN expansion, layer redistribution, GQA — is selected using ImageNet-1K validation loss at 256×256 plus iPhone latency. The paper asserts that validation loss 'shows stronger correlation with perceptual quality and human preference than conventional image metrics such as FID,' but no evidence is provided for this correlation in the context of architecture ablations, and the cited finding [18] does not directly validate this proxy for the architectural variants under consideration. If the proxy is unreliable, the claimed support for the architectural choices collapses. Please provide either (a) FID or human-preference measurements for the ablated variants, or (b) a defensible citation or analysis that validation-loss differences of the observed magnitude (e.g., 0.5130 vs. 0.5090) are perceptually meaningful.
  3. [Table 2 vs. supplementary Tables 2–4] The main paper's headline table omits Qwen-Image (20B) and HiDream-I1 (17B), even though both appear in the supplementary detailed tables. Since Qwen-Image is the actual KD teacher and is itself a '20× larger' model, the claim that the 0.4B variant 'surpasses models up to 20× larger' is not tested against the model that defines the teacher quality ceiling. The supplementary numbers show the 0.4B model is far below Qwen-Image on DPG (85.2 vs. 88.3), so the current wording overstates the result relative to the largest relevant baseline. Please include Qwen-Image (and ideally HiDream-I1) in the main comparison table, or qualify the claim to the specific baselines listed.
minor comments (5)
  1. [Sec. 3.3, Eqs. (9)–(10)] The K-DMD objective is not fully specified. In Eq. (9), the mapping F and the sampling distribution of τ are not defined; Eq. (10) reintroduces ξ' but the definitions of L^{ξ'}_out and L^{ξ'}_feat are only inferred from Eqs. (6)–(7); and the timestep-aware scaling operator S in Eq. (8) is never given explicitly. Please complete the notation so the method is reproducible.
  2. [Table 1] The caption says 'Comparison between Standalone and Elastic training for 0.4B and 2B models,' but the table columns are '0.4B' and '1.6B'. Please align the text and the table.
  3. [Sec. 4.1 and supplementary Sec. A] The main paper says the total on-device runtime is 'around 1.7 s' (Sec. 4.1), while Figure 1 and the supplementary latency table report 1.8 s for the 0.4B model. This inconsistency should be resolved.
  4. [Fig. 7] The human preference study reports percentages but no error bars, number of participants, or statistical significance tests. Given that some reported differences are small, please add confidence intervals or at least the number of ratings.
  5. [General] The paper does not state whether evaluation code or model weights will be released. Given that the main comparison depends on exact evaluation protocols (and the OOM issue above), releasing the evaluation harness would greatly improve trust in the numbers.

Circularity Check

0 steps flagged

No circularity: the system is empirically trained and benchmarked; self-citations are pipeline reuse, not load-bearing reductions.

full rationale

The paper makes no derivation-from-first-principles claim that reduces to its own inputs. Architectural choices (Sec. 3.1) are selected by ImageNet-1K validation loss and iPhone latency; validation loss is used as a training/ablation signal, not fitted to the T2I benchmark numbers reported in Table 2. The headline claim that the small variant (0.4B) surpasses models up to 20x larger is an empirical comparison on DPG-Bench, GenEval, T2I-CompBench, and CLIP, not a consequence of the equations. K-DMD (Eq. 10) combines DMD with knowledge distillation from an external few-step teacher [51] and external DMD references [78, 79]; the elastic framework is a standard Once-for-All/Slimmable slicing idea applied to a DiT and is acknowledged as such. Self-citations to SnapGen [28] describe a training pipeline and baseline, but the central results are not proved by citing [28]; no uniqueness theorem or ansatz is imported via self-citation. The Table 2 footnote ('† indicates out-of-memory (OOM) at 1024×1024 resolution') combined with reported benchmark scores for those rows is an internal-consistency/measurement-protocol concern, not a circularity: it does not show that any number was produced by fitting the paper's equations. No step qualifies under any circularity pattern.

Axiom & Free-Parameter Ledger

5 free parameters · 3 axioms · 0 invented entities

The central claims rest mostly on training and evaluation choices rather than new theoretical postulates. There are no invented physical entities. The largest unproved load-bearing assumption is the validation-loss proxy; the remaining entries are standard hyperparameters and domain assumptions about distillation and elastic training.

free parameters (5)
  • BNA block count B and neighborhood radius r = B=16, r=1
    Controls the sparsity and locality of attention; selected by ablation and described in Supp. Sec. D. At 1024×1024 this yields 9 neighboring tokens.
  • Elastic width multipliers for subnetworks = 0.375x, 0.5x, 1.0x
    Defines the 0.3B tiny, 0.4B small, and 1.6B full model variants; chosen heuristically to match hardware tiers in Sec. 3.2.
  • K-DMD LoRA rank and alpha = rank=64, alpha=128
    Hyperparameters for the student and critic LoRA in the step-distillation stage, reported in Supp. Sec. H.
  • DMD teacher guidance scale and time shift = cfg=4, time shift=3
    Fixed following Qwen-Image and Qwen-Image-Lightning settings, reported in Supp. Sec. H.
  • Distillation loss weighting / timestep-aware scaling S
    The relative weights and exact form of S in Eq. (8) and the balance among L_DMD, L_out, L_feat in Eq. (10) are load-bearing for convergence but not specified in the paper.
axioms (3)
  • domain assumption Validation loss on ImageNet correlates with perceptual quality and human preference.
    Sec. 3.1 uses Val Loss as the main architecture selection metric and claims it is more aligned with perceptual quality than FID; if false, the ablations do not support the design.
  • domain assumption A width-sliced super-network trained with uniform sampling, gradient scaling, and self-distillation remains stable and transfers across subnetworks.
    Sec. 3.2 asserts stable convergence and robust generalization, but provides no analysis and only small-scale ImageNet evidence in Table 1.
  • domain assumption Distillation from Qwen-Image and Qwen-Image-Lightning transfers enough generative ability to 0.3B/0.4B students.
    Sec. 3.3 and 4.1 rely on this; the near-lossless 4-step claim is only supported by the paper's own DPG/GenEval tables, not by an independent evaluation.

pith-pipeline@v1.3.0-alltime-deepseek · 34669 in / 13656 out tokens · 133022 ms · 2026-08-03T10:51:26.638716+00:00 · methodology

0 comments
read the original abstract

Recent advances in diffusion transformers (DiTs) have set new standards in image generation, yet remain impractical for on-device deployment due to their high computational and memory costs. In this work, we present an efficient DiT framework tailored for mobile and edge devices that achieves transformer-level generation quality under strict resource constraints. Our design combines three key components. First, we propose a compact DiT architecture with an adaptive global-local sparse attention mechanism that balances global context modeling and local detail preservation. Second, we propose an elastic training framework that jointly optimizes sub-DiTs of varying capacities within a unified supernetwork, allowing a single model to dynamically adjust for efficient inference across different hardware. Finally, we develop Knowledge-Guided Distribution Matching Distillation, a step-distillation pipeline that integrates the DMD objective with knowledge transfer from few-step teacher models, producing high-fidelity and low-latency generation (e.g., 4-step) suitable for real-time on-device use. Together, these contributions enable scalable, efficient, and high-quality diffusion models for deployment on diverse hardware.

Figures

Figures reproduced from arXiv: 2601.08303 by Aarush Gupta, Ahsan Mahmood, Aleksei Lebedev, Anil Kag, Anujraaj Goyal, Arpit Sahni, Dishani Lahiri, Dongting Hu, Huseyin Coskun, Ju Hu, Magzhan Gabidolla, Mingming Gong, Sergey Tulyakov, Yanyu Li, Yerlan Idelbayev.

Figure 1
Figure 1. Figure 1: Top: Our text-to-image Diffusion Transformer (0.4B parameters) generates diverse, high-fidelity 1K images in just 1.8 s on a mobile device. All examples are produced by this on-device model at a resolution of approximately 10242 . Bottom: Comparison across various text-to-image models. Both our on-device (small) and server-side (full) versions achieve competitive visual quality. 1 arXiv:2601.08303v2 [cs.CV… view at source ↗
Figure 2
Figure 2. Figure 2: Efficient DiT Overview. Left: Our model consists of three stages: Down, Middle and Up. Down and Up blocks operate on high-resolution latent while using our novel Adaptive Sparse Self-Attention (ASSA) layers. Middle blocks operate at latents downsampled by 2 × 2 window and use standard Self-Attention (SA) layers. Other layers in the blocks are Cross-Attention (CA) for modulating with input text conditioning… view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of Blockwise Neighborhood Attention. (a) Naive Neighborhood Attention, where each query attends to its local window of 3 neighboring tokens. (b) Corresponding self￾attention mask showing the limited receptive field for each query. (c) Blockwise Neighborhood Attention extends this concept by grouping tokens into 8 local blocks, enabling efficient attention computation while preserving locality.… view at source ↗
Figure 5
Figure 5. Figure 5: Elastic Training Framework. Given a supernetwork, we define sub-networks as different granularities of the hidden di￾mension. During training, we sample sub-networks uniformly and supervise them using the output from the supernetwork. In addi￾tion, we use standard diffusion loss on all granularities. This leads to more stable training and imparts knowledge to sub-networks. key and value projections are not… view at source ↗
Figure 6
Figure 6. Figure 6: Knowledge-guided Distribution Matching Distilla￾tion (K-DMD). Our step distillation method combines distribution matching with knowledge transfer from a few-step teacher. Step Distillation. Following recent one-step distillation methods [78, 79], we adopt Distribution Matching Distil￾lation (DMD) for step distillation. However, DMD requires careful tuning of hyperparameters such as teacher guidance scale a… view at source ↗
Figure 7
Figure 7. Figure 7: Human Evaluation. We conduct a user study comparing our small (0.4B) and full (1.6B) variants with three baselines—SANA (1.6B), SD3-Medium (2B), and Flux.1-dev (12B)—across three key attributes: realism, visual fidelity, and text–image alignment [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Few-step Generation. Comparison of images produced by the tiny (0.3B), small (0.4B), and full (1.6B) models under 28- step (w/o K-DMD) and 4-step (w/ K-DMD) settings. Numbers in the corners denote DPG / GenEval scores. 5. Conclusion In this work, we presented an Efficient Diffusion Transformer that brings transformer-based image gener￾ation to mobile and edge devices. Through adaptive global–local sparse a… view at source ↗
Figure 1
Figure 1. Figure 1: On-device Image Generation Demo. Screenshots from our on-device application running on an iPhone 16 Pro Max. The left panel shows results from the small (0.4B) model, and the right panel shows results from the full variant with 4-bit quantization. on mobile hardware. Finally, the model is exported via CoreML to generate a computation graph for deployment. To deploy the full model (1.6B) on deivce, we quant… view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of Blockwise Neighborhood Attention (BNA). Visualization of BNA under different hyperparameter settings of block number (b) and neighborhood radius (r), showing the corresponding spatial neighbor coverage and attention sparsity. D. Additional Illustration of Blockwise Neigh￾borhood Attention In [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison on ImageNet-1K. Visual comparison between on-device models SnapGen [28] (0.4B, left in each pair, validation loss = 0.5131) and our small model (0.4B, right in each pair, validation loss = 0.5090). Our model produces sharper textures, more consistent colors, and improved structural fidelity across diverse categories. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Additional Qualitative Comparison. Our models demonstrate competitive visual quality and superior prompt-following ability. Input text prompts are shown above each image grid; all images are generated at 10242 resolution. Zoom in for details. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Additional Qualitative Comparison. Our models demonstrate competitive visual quality and superior prompt-following ability. Input text prompts are shown above each image grid; all images are generated at 10242 resolution. Zoom in for details. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_5.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Cross-Space Distillation: Teaching One-Step Students with Modern Diffusion Teachers

    cs.CV 2026-06 unverdicted novelty 7.0

    Introduces a Bridge latent interface that maps mismatched student latents into teacher space, enabling distillation from modern diffusion teachers to compact one-step students and raising SD 1.5 HPSv3 from 5.4 to 9.4 ...

Reference graph

Works this paper leans on

86 extracted references · 26 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Stable diffusion 3.5

    Stability AI. Stable diffusion 3.5. https://github.com/Stability-AI/sd3.5, 2024. 7, 8, 15, 16

    2024

  2. [2]

    GQA: Training generalized multi-query transformer models from multi-head checkpoints

    Joshua Ainslie, James Lee-Thorp, Michiel de Jong, Yury Zemlyanskiy, Federico Lebron, and Sumit Sanghai. GQA: Training generalized multi-query transformer models from multi-head checkpoints. InProceedings of the 2023 Confer- ence on Empirical Methods in Natural Language Processing, pages 4895–4901, Singapore, 2023. Association for Compu- tational Linguistics. 5

    2023

  3. [3]

    Sd3.5- flash: Distribution-guided distillation of generative flows,

    Hmrishav Bandyopadhyay, Rahim Entezari, Jim Scott, Reshinth Adithyan, Yi-Zhe Song, and Varun Jampani. Sd3.5- flash: Distribution-guided distillation of generative flows,

  4. [4]

    All are worth words: A vit backbone for diffusion models

    Fan Bao, Shen Nie, Kaiwen Xue, Yue Cao, Chongxuan Li, Hang Su, and Jun Zhu. All are worth words: A vit backbone for diffusion models. InCVPR, 2023. 3

    2023

  5. [5]

    All are worth words: A vit backbone for diffusion models

    Fan Bao, Shen Nie, Kaiwen Xue, Yue Cao, Chongxuan Li, Hang Su, and Jun Zhu. All are worth words: A vit backbone for diffusion models. InCVPR, 2023. 5

    2023

  6. [6]

    Large scale gan training for high fidelity natural image synthe- sis

    Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale gan training for high fidelity natural image synthe- sis. InInternational Conference on Learning Representa- tions (ICLR), 2019. 3

    2019

  7. [7]

    Once-for-all: Train one network and specialize it for efficient deployment

    Han Cai, Chuang Gan, Tianzhe Wang, Zhekai Zhang, and Song Han. Once-for-all: Train one network and specialize it for efficient deployment. InInternational Conference on Learning Representations (ICLR), 2020. 3

    2020

  8. [8]

    Hidream-i1: A high-efficient image generative foundation model with sparse diffusion transformer.arXiv preprint arXiv:2505.22705, 2025

    Qi Cai, Jingwen Chen, Yang Chen, Yehao Li, Fuchen Long, Yingwei Pan, Zhaofan Qiu, Yiheng Zhang, Feng- bin Gao, Peihan Xu, Yimeng Wang, Kai Yu, Wenxuan Chen, Ziwei Feng, Zijian Gong, Jianzhuang Pan, Yi Peng, Rui Tian, Siyu Wang, Bo Zhao, Ting Yao, and Tao Mei. Hidream-i1: A high-efficient image generative foundation model with sparse diffusion transformer....

    Pith/arXiv arXiv 2025

  9. [9]

    EdgeFusion: On-Device Text-to-Image Generation.arXiv preprint arXiv:2404.11925, 2024

    Thibault Castells, Hyoung-Kyu Song, Tairen Piao, Shinkook Choi, Bo-Kyeong Kim, Hanyoung Yim, Changgwun Lee, Jae Gon Kim, and Tae-Ho Kim. EdgeFusion: On-Device Text-to-Image Generation.arXiv preprint arXiv:2404.11925, 2024. 3

    Pith/arXiv arXiv 2024

  10. [10]

    Pixart-σ: Weak-to-strong training of dif- fusion transformer for 4k text-to-image generation.arXiv preprint arXiv:2403.04692, 2024

    Junsong Chen, Chongjian Ge, Enze Xie, Yue Wu, Lewei Yao, Xiaozhe Ren, Zhongdao Wang, Ping Luo, Huchuan Lu, and Zhenguo Li. Pixart-σ: Weak-to-strong training of dif- fusion transformer for 4k text-to-image generation.arXiv preprint arXiv:2403.04692, 2024. 3, 8, 15, 16

    Pith/arXiv arXiv 2024

  11. [11]

    Pixart-α: Fast training of dif- fusion transformer for photorealistic text-to-image synthesis

    Junsong Chen, Jincheng YU, Chongjian GE, Lewei Yao, Enze Xie, Zhongdao Wang, James Kwok, Ping Luo, Huchuan Lu, and Zhenguo Li. Pixart-α: Fast training of dif- fusion transformer for photorealistic text-to-image synthesis. InThe Twelfth International Conference on Learning Repre- sentations, 2024. 2, 3, 8, 15, 16

    2024

  12. [12]

    Sana-sprint: One-step diffusion with continuous-time con- sistency distillation, 2025

    Junsong Chen, Shuchen Xue, Yuyang Zhao, Jincheng Yu, Sayak Paul, Junyu Chen, Han Cai, Song Han, and Enze Xie. Sana-sprint: One-step diffusion with continuous-time con- sistency distillation, 2025. 3

    2025

  13. [13]

    Scalable high-resolution pixel-space image syn- thesis with hourglass diffusion transformers

    Katherine Crowson, Stefan Andreas Baumann, Alex Birch, Tanishq Mathew Abraham, Daniel Z Kaplan, and Enrico Shippole. Scalable high-resolution pixel-space image syn- thesis with hourglass diffusion transformers. InProceedings of the 41st International Conference on Machine Learning, pages 9550–9575. PMLR, 2024. 3, 4

    2024

  14. [14]

    Transformers are SSMs: Gen- eralized models and efficient algorithms through structured state space duality

    Tri Dao and Albert Gu. Transformers are SSMs: Gen- eralized models and efficient algorithms through structured state space duality. InInternational Conference on Machine Learning (ICML), 2024. 3

    2024

  15. [15]

    Deepfloyd.https://github.com/deep-floyd/IF,

    DeepFloyd. Deepfloyd.https://github.com/deep-floyd/IF,

  16. [16]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. 3

    2009

  17. [17]

    Kakade, Ali Farhadi, and Prateek Jain

    Khatri Devvrit, Sneha Kudugunta, Aditya Kusupati, Tim Dettmers, Kaifeng Chen, Inderjit S Dhillon, Yulia Tsvetkov, Hannaneh Hajishirzi, Sham M. Kakade, Ali Farhadi, and Prateek Jain. Matformer: Nested transformer for elastic in- ference. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024. 3, 5

    2024

  18. [18]

    Scaling rectified flow trans- formers for high-resolution image synthesis, 2024

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas M ¨uller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, Dustin Podell, Tim Dockhorn, Zion English, Kyle Lacey, Alex Goodwin, Yan- nik Marek, and Robin Rombach. Scaling rectified flow trans- formers for high-resolution image synthesis, 2024. 2, 3, 7, 8, 15, 16

    2024

  19. [19]

    Geneval: An object-focused framework for evaluating text- to-image alignment.Advances in Neural Information Pro- cessing Systems, 36, 2024

    Dhruba Ghosh, Hannaneh Hajishirzi, and Ludwig Schmidt. Geneval: An object-focused framework for evaluating text- to-image alignment.Advances in Neural Information Pro- cessing Systems, 36, 2024. 7

    2024

  20. [20]

    Generative adversarial nets

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. InAdvances in Neural Information Processing Systems (NeurIPS), 2014. 3

    2014

  21. [21]

    Mamba: Linear-time sequence modeling with selective state spaces.arXiv preprint arXiv:2312.00752, 2023

    Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces.arXiv preprint arXiv:2312.00752, 2023. 3

    Pith/arXiv arXiv 2023

  22. [22]

    Generalized neighborhood attention: Multi-dimensional sparse attention at the speed of light.arXiv preprint arXiv:2504.16922, 2025

    Ali Hassani, Steven Walton, Humphrey Shi, et al. Generalized neighborhood attention: Multi-dimensional sparse attention at the speed of light.arXiv preprint arXiv:2504.16922, 2025. 3

    Pith/arXiv arXiv 2025

  23. [23]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium.Advances in neural information processing systems, 30, 2017

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium.Advances in neural information processing systems, 30, 2017. 3

    2017

  24. [24]

    Denoising diffu- sion probabilistic models.Advances in Neural Information Processing Systems (NeurIPS), 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models.Advances in Neural Information Processing Systems (NeurIPS), 2020. 3

    2020

  25. [25]

    sim- ple diffusion: End-to-end diffusion for high resolution im- ages

    Emiel Hoogeboom, Jonathan Heek, and Tim Salimans. sim- ple diffusion: End-to-end diffusion for high resolution im- ages. InProceedings of the 40th International Conference on Machine Learning, pages 13213–13232. PMLR, 2023. 3 9

    2023

  26. [26]

    Simpler diffu- sion: 1.5 fid on imagenet512 with pixel-space diffusion

    Emiel Hoogeboom, Thomas Mensink, Jonathan Heek, Kay Lamerigts, Ruiqi Gao, and Tim Salimans. Simpler diffu- sion: 1.5 fid on imagenet512 with pixel-space diffusion. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 18062–18071, 2025. 3, 4

    2025

  27. [27]

    Dynabert: Dynamic bert with adaptive width and depth

    Lu Hou, Zhiqi Huang, Lifeng Shang, Xin Jiang, Xiao Chen, and Qun Liu. Dynabert: Dynamic bert with adaptive width and depth. InAdvances in Neural Information Processing Systems, pages 9782–9793. Curran Associates, Inc., 2020. 3

    2020

  28. [28]

    Chan, Mingming Gong, Sergey Tulyakov, Anil Kag, Yanwu Xu, and Jian Ren

    Dongting Hu, Jierun Chen, Xijie Huang, Huseyin Coskun, Arpit Sahni, Aarush Gupta, Anujraaj Goyal, Dishani Lahiri, Rajesh Singh, Yerlan Idelbayev, Junli Cao, Yanyu Li, Kwang-Ting Cheng, S.-H. Chan, Mingming Gong, Sergey Tulyakov, Anil Kag, Yanwu Xu, and Jian Ren. Snap- gen: Taming high-resolution text-to-image models for mobile devices with efficient archi...

    Pith/arXiv arXiv 2024

  29. [29]

    LoRA: Low-rank adaptation of large language models

    Edward J Hu, yelong shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-rank adaptation of large language models. InIn- ternational Conference on Learning Representations, 2022. 7

    2022

  30. [30]

    Ella: Equip diffusion models with llm for enhanced semantic alignment.arXiv preprint arXiv:2403.05135, 2024

    Xiwei Hu, Rui Wang, Yixiao Fang, Bin Fu, Pei Cheng, and Gang Yu. Ella: Equip diffusion models with llm for enhanced semantic alignment.arXiv preprint arXiv:2403.05135, 2024. 7

    Pith/arXiv arXiv 2024

  31. [31]

    T2I-CompBench++: An Enhanced and Comprehensive Benchmark for Compositional Text-to- Image Generation .IEEE Transactions on Pattern Analysis Machine Intelligence, (01):1–17, 5555

    Kaiyi Huang, Chengqi Duan, Kaiyue Sun, Enze Xie, Zhen- guo Li, and Xihui Liu. T2I-CompBench++: An Enhanced and Comprehensive Benchmark for Compositional Text-to- Image Generation .IEEE Transactions on Pattern Analysis Machine Intelligence, (01):1–17, 5555. 7

  32. [32]

    Self forcing: Bridging the train- test gap in autoregressive video diffusion.arXiv preprint arXiv:2506.08009, 2025

    Xun Huang, Zhengqi Li, Guande He, Mingyuan Zhou, and Eli Shechtman. Self forcing: Bridging the train- test gap in autoregressive video diffusion.arXiv preprint arXiv:2506.08009, 2025. 7

    Pith/arXiv arXiv 2025

  33. [33]

    TP-blend: Textual-prompt attention pairing for precise object-style blending in diffusion models.Transactions on Machine Learning Research, 2025

    Xin Jin, Yichuan Zhong, and Yapeng Tian. TP-blend: Textual-prompt attention pairing for precise object-style blending in diffusion models.Transactions on Machine Learning Research, 2025. 3

    2025

  34. [34]

    Ascan: Asymmetric convolution-attention networks for efficient recognition and generation.arXiv preprint arXiv:2411.04967, 2024

    Anil Kag, Huseyin Coskun, Jierun Chen, Junli Cao, Willi Menapace, Aliaksandr Siarohin, Sergey Tulyakov, and Jian Ren. Ascan: Asymmetric convolution-attention networks for efficient recognition and generation.arXiv preprint arXiv:2411.04967, 2024. 3

    Pith/arXiv arXiv 2024

  35. [35]

    Bk-sdm: Architecturally Compressed Sta- ble Diffusion for Efficient Text-to-Image Generation

    Bo-Kyeong Kim, Hyoung-Kyu Song, Thibault Castells, and Shinkook Choi. Bk-sdm: Architecturally Compressed Sta- ble Diffusion for Efficient Text-to-Image Generation. In Workshop on Efficient Systems for Foundation Models@ ICML2023, 2023. 3

    2023

  36. [36]

    Flux: A generative model by black for- est labs.https://github.com/black- forest- labs/flux, 2024

    Black Forest Labs. Flux: A generative model by black for- est labs.https://github.com/black- forest- labs/flux, 2024. Accessed: 2025-05-14. 2, 3, 7, 8, 15, 16

    2024

  37. [37]

    Playground v1,

    Daiqing Li, Aleks Kamko, Ali Sabet, Ehsan Akhgari, Lin- miao Xu, and Suhail Doshi. Playground v1, . 3

  38. [38]

    Playground v2,

    Daiqing Li, Aleks Kamko, Ali Sabet, Ehsan Akhgari, Lin- miao Xu, and Suhail Doshi. Playground v2,

  39. [39]

    Playground V2

    Daiqing Li, Aleks Kamko, Ehsan Akhgari, Ali Sabet, Lin- miao Xu, and Suhail Doshi. Playground V2. 5: Three Insights towards Enhancing Aesthetic Quality in Text-to- Image Generation.arXiv preprint arXiv:2402.17245, 2024. 3, 8, 15, 16

    Pith/arXiv arXiv 2024

  40. [40]

    Svdquant: Absorbing outliers by low-rank components for 4-bit diffusion models

    Muyang Li*, Yujun Lin*, Zhekai Zhang*, Tianle Cai, Xiuyu Li, Junxian Guo, Enze Xie, Chenlin Meng, Jun-Yan Zhu, and Song Han. Svdquant: Absorbing outliers by low-rank components for 4-bit diffusion models. InThe Thirteenth In- ternational Conference on Learning Representations, 2025. 2, 3

    2025

  41. [41]

    Snap- fusion: Text-to-Image Diffusion Model on Mobile Devices within Two Seconds.Advances in Neural Information Pro- cessing Systems, 36, 2024

    Yanyu Li, Huan Wang, Qing Jin, Ju Hu, Pavlo Chemerys, Yun Fu, Yanzhi Wang, Sergey Tulyakov, and Jian Ren. Snap- fusion: Text-to-Image Diffusion Model on Mobile Devices within Two Seconds.Advances in Neural Information Pro- cessing Systems, 36, 2024. 2, 3, 13

    2024

  42. [42]

    SDXL- Lightning: Progressive Adversarial Diffusion Distillation,

    Shanchuan Lin, Anran Wang, and Xiao Yang. SDXL- Lightning: Progressive Adversarial Diffusion Distillation,

  43. [43]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13, pages 740–755. Springer, 2014. 7

    2014

  44. [44]

    Flow matching for generative mod- eling.arXiv preprint arXiv:2210.02747, 2022

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximil- ian Nickel, and Matt Le. Flow matching for generative mod- eling.arXiv preprint arXiv:2210.02747, 2022. 3

    Pith/arXiv arXiv 2022

  45. [45]

    Playground v3: Improving text-to- image alignment with deep-fusion large language models

    Bingchen Liu, Ehsan Akhgari, Alexander Visheratin, Aleks Kamko, Linmiao Xu, Shivam Shrirao, Joao Souza, Suhail Doshi, and Daiqing Li. Playground v3: Improving text-to- image alignment with deep-fusion large language models. arXiv preprint arXiv:2409.10695, 2024. 3

    Pith/arXiv arXiv 2024

  46. [46]

    Magicedit: High-fidelity and temporally coherent video edit- ing.arXiv preprint arXiv:2303.08354, 2023

    Renjing Liu, Jiatao Li, William Peebles, and Saining Xie. Magicedit: High-fidelity and temporally coherent video edit- ing.arXiv preprint arXiv:2303.08354, 2023. 3

    Pith/arXiv arXiv 2023

  47. [47]

    Linfusion: 1 gpu, 1 minute, 16k image.arXiv preprint arXiv:2409.02097, 2024

    Songhua Liu, Weihao Yu, Zhenxiong Tan, and Xinchao Wang. Linfusion: 1 gpu, 1 minute, 16k image.arXiv preprint arXiv:2409.02097, 2024. 3

    Pith/arXiv arXiv 2024

  48. [48]

    Flow straight and fast: Learning to generate and transfer data with rectified flow.arXiv preprint arXiv:2209.03003, 2022

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow.arXiv preprint arXiv:2209.03003, 2022. 3, 7

    Pith/arXiv arXiv 2022

  49. [49]

    On distillation of guided diffusion models

    Chenlin Meng, Ruiqi Gao, Diederik P Kingma, Stefano Er- mon, Jonathan Ho, and Tim Salimans. On distillation of guided diffusion models. InNeurIPS 2022 Workshop on Score-Based Methods, 2022. 3

    2022

  50. [50]

    SDEdit: Guided image synthesis and editing with stochastic differential equa- tions

    Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jia- jun Wu, Jun-Yan Zhu, and Stefano Ermon. SDEdit: Guided image synthesis and editing with stochastic differential equa- tions. InInternational Conference on Learning Representa- tions, 2022. 3

    2022

  51. [51]

    Qwen-image-lightning: Distilled qwen-image models for fast, high-fidelity text-to-image generation

    ModelTC. Qwen-image-lightning: Distilled qwen-image models for fast, high-fidelity text-to-image generation. https://github.com/ModelTC/Qwen- Image- Lightning, 2025. Version V1.x/ V2.x available; Apache- 2.0 license. 7, 15

    2025

  52. [52]

    Kim, Aliaksandr Siarohin, 10 and Anil Kag

    Dogyun Park, Moayed Haji-Ali, Yanyu Li, Willi Menapace, Sergey Tulyakov, Hyunwoo J. Kim, Aliaksandr Siarohin, 10 and Anil Kag. Sprint: Sparse-dense residual fusion for ef- ficient diffusion transformers, 2025. 3

    2025

  53. [53]

    Scalable diffusion models with transformers.arXiv preprint arXiv:2212.09748, 2023

    William Peebles and Saining Xie. Scalable diffusion models with transformers.arXiv preprint arXiv:2212.09748, 2023. 2, 3

    Pith/arXiv arXiv 2023

  54. [54]

    Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis.arXiv preprint arXiv:2307.01952, 2023

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis.arXiv preprint arXiv:2307.01952, 2023. 2, 3, 8, 15, 16

    Pith/arXiv arXiv 2023

  55. [55]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. InInternational conference on machine learning, pages 8748–8763. PMLR, 2021. 3, 7

    2021

  56. [56]

    High-resolution image syn- thesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj¨orn Ommer. High-resolution image syn- thesis with latent diffusion models. InIEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 2, 3

    2022

  57. [57]

    Photorealistic text-to-image diffusion models with deep lan- guage understanding

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Sali- mans, Jonathan Ho, David J Fleet, and Mohammad Norouzi. Photorealistic text-to-image diffusion models with deep lan- guage understanding. InAdvances in Neural Information Processing Systems, pages 3647...

    2022

  58. [58]

    Progressive distillation for fast sampling of diffusion models

    Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models. InInternational Confer- ence on Learning Representations, 2022. 3

    2022

  59. [59]

    Fast transformer decoding: One write-head is all you need.arXiv preprint arXiv:1911.02150, 2019

    Noam Shazeer. Fast transformer decoding: One write-head is all you need.arXiv preprint arXiv:1911.02150, 2019. 3

    Pith/arXiv arXiv 1911

  60. [60]

    Score-based generative modeling through stochastic differential equa- tions.International Conference on Learning Representations (ICLR), 2021

    Yang Song, Jascha Sohl-Dickstein, Durk P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions.International Conference on Learning Representations (ICLR), 2021. 3

    2021

  61. [61]

    Consis- tency models.International Conference on Machine Learn- ing (ICML), 2023

    Yang Song, Chenlin Meng, and Stefano Ermon. Consis- tency models.International Conference on Machine Learn- ing (ICML), 2023. 3

    2023

  62. [62]

    Bitsfusion: 1.99 bits weight quantization of diffusion model

    Yang Sui, Yanyu Li, Anil Kag, Yerlan Idelbayev, Junli Cao, Ju Hu, Dhritiman Sagar, Bo Yuan, Sergey Tulyakov, and Jian Ren. Bitsfusion: 1.99 bits weight quantization of diffusion model. InAdvances in Neural Information Processing Sys- tems, pages 76775–76818. Curran Associates, Inc., 2024. 3

    2024

  63. [63]

    Gemma 3n

    Gemma Team. Gemma 3n. 2025. 5

    2025

  64. [64]

    Gemma Team, Aishwarya Kamath, Johan Ferret, Shreya Pathak, Nino Vieillard, Ramona Merhej, Sarah Perrin, Ta- tiana Matejovicova, Alexandre Ram ´e, Morgane Rivi `ere, Louis Rouillard, Thomas Mesnard, Geoffrey Cideron, Jean bastien Grill, Sabela Ramos, Edouard Yvinec, Michelle Cas- bon, Etienne Pot, Ivo Penchev, Ga ¨el Liu, Francesco Visin, Kathleen Kenealy,...

    2025

  65. [65]

    U-dits: Downsample tokens in u-shaped diffusion transformers, 2024

    Yuchuan Tian, Zhijun Tu, Hanting Chen, Jie Hu, Chao Xu, and Yunhe Wang. U-dits: Downsample tokens in u-shaped diffusion transformers, 2024. 3, 4

    2024

  66. [66]

    Sortednet: A scalable and generalized framework for training modular deep neural networks, 2024

    Mojtaba Valipour, Mehdi Rezagholizadeh, Hossein Ra- jabzadeh, Parsa Kavehzadeh, Marzieh Tahaei, Boxing Chen, and Ali Ghodsi. Sortednet: A scalable and generalized framework for training modular deep neural networks, 2024. 11 3

    2024

  67. [67]

    Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314, 2025

    Team Wan, Ang Wang, Baole Ai, Bin Wen, Chaojie Mao, Chen-Wei Xie, Di Chen, Feiwu Yu, Haiming Zhao, Jianx- iao Yang, Jianyuan Zeng, Jiayu Wang, Jingfeng Zhang, Jin- gren Zhou, Jinkai Wang, Jixuan Chen, Kai Zhu, Kang Zhao, Keyu Yan, Lianghua Huang, Mengyang Feng, Ningyi Zhang, Pandeng Li, Pingyu Wu, Ruihang Chu, Ruili Feng, Shiwei Zhang, Siyang Sun, Tao Fan...

    Pith/arXiv arXiv 2025

  68. [68]

    Phased consistency model.arXiv preprint arXiv:2405.18407, 2024

    Fu-Yun Wang, Zhaoyang Huang, Alexander William Bergman, Dazhong Shen, Peng Gao, Michael Lingel- bach, Keqiang Sun, Weikang Bian, Guanglu Song, Yu Liu, et al. Phased consistency model.arXiv preprint arXiv:2405.18407, 2024. 3

    Pith/arXiv arXiv 2024

  69. [69]

    Hat: Hardware-aware transformers for efficient natural language processing

    Hanrui Wang, Zhanghao Wu, Zhijian Liu, Han Cai, Ligeng Zhu, Chuang Gan, and Song Han. Hat: Hardware-aware transformers for efficient natural language processing. InAn- nual Conference of the Association for Computational Lin- guistics, 2020. 3

    2020

  70. [70]

    Qwen-image technical report,

    Chenfei Wu, Jiahao Li, Jingren Zhou, Junyang Lin, Kaiyuan Gao, Kun Yan, Sheng ming Yin, Shuai Bai, Xiao Xu, Yilei Chen, Yuxiang Chen, Zecheng Tang, Zekai Zhang, Zhengyi Wang, An Yang, Bowen Yu, Chen Cheng, Dayiheng Liu, De- qing Li, Hang Zhang, Hao Meng, Hu Wei, Jingyuan Ni, Kai Chen, Kuan Cao, Liang Peng, Lin Qu, Minggang Wu, Peng Wang, Shuting Yu, Tingk...

  71. [71]

    Tinyclip: Clip distillation via affinity mimicking and weight inheritance

    Kan Wu, Houwen Peng, Zhenghong Zhou, Bin Xiao, Mengchen Liu, Lu Yuan, Hong Xuan, Michael Valenzuela, Xi (Stephen) Chen, Xinggang Wang, Hongyang Chao, and Han Hu. Tinyclip: Clip distillation via affinity mimicking and weight inheritance. InProceedings of the IEEE/CVF In- ternational Conference on Computer Vision (ICCV), pages 21970–21980, 2023. 7

    2023

  72. [72]

    Taming diffusion transformer for efficient mobile video gen- eration in seconds, 2025

    Yushu Wu, Yanyu Li, Anil Kag, Ivan Skorokhodov, Willi Menapace, Ke Ma, Arpit Sahni, Ju Hu, Aliaksandr Siaro- hin, Dhritiman Sagar, Yanzhi Wang, and Sergey Tulyakov. Taming diffusion transformer for efficient mobile video gen- eration in seconds, 2025. 3

    2025

  73. [73]

    Metaxas, Yanzhi Wang, Sergey Tulyakov, and Jian Ren

    Yushu Wu, Zhixing Zhang, Yanyu Li, Yanwu Xu, Anil Kag, Yang Sui, Huseyin Coskun, Ke Ma, Aleksei Lebedev, Ju Hu, Dimitris N. Metaxas, Yanzhi Wang, Sergey Tulyakov, and Jian Ren. Snapgen-v: Generating a five-second video within five seconds on a mobile device. InProceedings of the Com- puter Vision and Pattern Recognition Conference (CVPR), pages 2479–2490, 2025. 3

    2025

  74. [74]

    Sparse videogen: Accelerating video diffusion transformers with spatial-temporal sparsity.arXiv preprint arXiv:2502.01776,

    Ruijie Xi, Qingxiong Zhang, Hongyu Gao, et al. Sparse videogen: Accelerating video diffusion transformers with spatial-temporal sparsity.arXiv preprint arXiv:2502.01776,

    Pith/arXiv arXiv

  75. [75]

    Training-free and adaptive sparse attention for efficient long video generation.arXiv preprint arXiv:2502.21079, 2025

    Yifei Xia, Suhan Ling, Fangcheng Fu, Yujie Wang, Huixia Li, Xuefeng Xiao, and Bin Cui. Training-free and adaptive sparse attention for efficient long video generation.arXiv preprint arXiv:2502.21079, 2025. 3

    Pith/arXiv arXiv 2025

  76. [76]

    SANA: Efficient high-resolution text-to-image synthesis with linear diffusion transformers

    Enze Xie, Junsong Chen, Junyu Chen, Han Cai, Haotian Tang, Yujun Lin, Zhekai Zhang, Muyang Li, Ligeng Zhu, Yao Lu, and Song Han. SANA: Efficient high-resolution text-to-image synthesis with linear diffusion transformers. In The Thirteenth International Conference on Learning Repre- sentations, 2025. 3, 8, 15, 16

    2025

  77. [77]

    Ufogen: You forward once large scale text-to-image gener- ation via diffusion gans

    Yanwu Xu, Yang Zhao, Zhisheng Xiao, and Tingbo Hou. Ufogen: You forward once large scale text-to-image gener- ation via diffusion gans. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8196–8206, 2024. 3

    2024

  78. [78]

    Im- proved distribution matching distillation for fast image syn- thesis.arXiv preprint arXiv:2405.14867, 2024

    Tianwei Yin, Micha ¨el Gharbi, Taesung Park, Richard Zhang, Eli Shechtman, Fredo Durand, and William T Freeman. Im- proved distribution matching distillation for fast image syn- thesis.arXiv preprint arXiv:2405.14867, 2024. 2, 3, 7

    Pith/arXiv arXiv 2024

  79. [79]

    One-step diffusion with distribution matching distillation

    Tianwei Yin, Micha ¨el Gharbi, Richard Zhang, Eli Shecht- man, Fr´edo Durand, William T Freeman, and Taesung Park. One-step diffusion with distribution matching distillation. In CVPR, 2024. 2, 3, 7

    2024

  80. [80]

    From slow bidirectional to fast autoregressive video diffusion mod- els

    Tianwei Yin, Qiang Zhang, Richard Zhang, William T Free- man, Fredo Durand, Eli Shechtman, and Xun Huang. From slow bidirectional to fast autoregressive video diffusion mod- els. InCVPR, 2025. 7

    2025

Showing first 80 references.