Pith. sign in

REVIEW 3 major objections 6 minor 4 cited by

LayerFusion: Harmonized Multi-Layer Text-to-Image Generation with Generative Priors

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a single training-free diffusion pipeline can generate an RGBA foreground, an RGB background, and a harmonized blend simultaneously by blending attention outputs between two pretrained models.

desk verdict Training-free attention-level blending of two diffusion streams is a genuinely new trick, but the central mask-formulation text is mathematically inverted and the quantitative evaluation leans on circular references. read the letter →

arxiv 2412.04460 v1 pith:MPZG3A65 submitted 2024-12-05 cs.CV

Yusuf Dalva , Yijun Li , Qing Liu , Nanxuan Zhao , Jianming Zhang , Zhe Lin , Pinar Yanardag This is my paper · ORCID
classification cs.CV
keywords latentdiffusionmodelstext-to-imagegenerationlayeredcontentRGBAtransparencyattentionblendingharmonizationtraining-freepipelinegenerativepriors
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

LayerFusion aims to generate a layered image triplet--a foreground layer with transparency (RGBA), a background layer (RGB), and a blended composite--from two text prompts in one diffusion process, with no finetuning. The key claim is that harmony between layers should be enforced inside the attention mechanism rather than by generating layers sequentially or composing them afterward. If correct, creative workflows that need editable cutouts, such as graphic design, animation, and digital art, would get a foreground with an alpha channel, a background, and a natural composition that respond to each other.

What carries the argument

The central objects are attention-level blending masks: a soft mask $\text{masksoft} = \text{normalize}(s \cdot c)$ built from the structure prior $s$ and the content confidence prior $c$, and a hard mask $\text{maskhard} = \sigma(d(\text{masksoft}-0.5))$. The soft mask injects foreground content into the blended image, while the hard mask transfers the blended result back into the foreground, creating a bidirectional harmonization loop that requires no training updates.

What would settle it

Generate a foreground prompt outside the foreground generator's familiar territory (for example, a rare or compound object such as 'a glass chess piece with a metal pawn inside') against a busy background; if the extracted self-attention sparsity map or the <EOS> cross-attention map does not cover the object's full silhouette, background texture will leak into the alpha channel or the object will be clipped, showing that the generative priors are not sufficient for harmonized blending on new subjects.

Watch

Extended reading notes

Core claim

The paper proposes that attention probability maps of the foreground generator encode two complementary generative priors: a structure prior from the last self-attention layer, computed as an inverted per-token sparsity score, and a content confidence prior from the <EOS> cross-attention map. Multiplying these gives a soft blending mask, and a sigmoid binarization produces a hard mask. During denoising, the blended attention output is replaced by a weighted mix of foreground and blended attention (Eq. 1), and the foreground attention is updated by the blended output inside the hard mask (Eq. 2). This two-way attention exchange is what the paper claims yields a coherent transparent foreground, a clean background, and a harmonized blend simultaneously.

Load-bearing premise

The method assumes the attention maps of the foreground generator--the sparsity of the last self-attention layer and the <EOS> cross-attention map--reliably mark where the foreground object and its transparency live in the latent image.

Editorial extensions

If this is right

  • If the claim holds, layered images for editing arrive in one generation step: a clean foreground cutout, a clean background, and a composed result that share consistent lighting and geometry.
  • Background context can steer foreground appearance without losing subject identity, so the same prompt can change outfit or weather details when the background prompt changes.
  • Because no training is needed, the pipeline can be applied to any compatible pair of pretrained diffusion models, not only the exact ones used in the experiments.
  • Harmonization happens at generation time, so grounding and shadow effects are handled by the blend itself rather than by a separate post-processing harmonization step.
  • Spatial edits become straightforward: the generated foreground can be moved, replaced, or removed over the background without the background-completion artifacts seen in sequential generation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same attention-level blending recurrence could, in principle, be stacked for three or more layers, but stability across many layers is untested and would need a new study.
  • Because the masks come from potentially noisy attention maps, averaging the structure and content priors over several diffusion timesteps before binarizing could sharpen boundaries at no training cost; the paper does not report this experiment.
  • Since the content prior is read from the <EOS> token, prompt phrasing may act as a dial for how strongly the background influences the foreground; testing whether rephrasing the foreground prompt with a fixed seed changes the final alpha channel would directly probe that lever.
  • If the extracted attention masks remain temporally coherent, the same harmonized-blending idea could transfer to video layers, but per-frame mask consistency would need to be verified to avoid flicker.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes LayerFusion, a training-free method for simultaneous generation of a foreground RGBA layer, a background RGB layer, and a blended RGB image using pre-trained latent diffusion models. The method extracts structure and content confidence masks from the self-attention and cross-attention maps of the foreground generator (LayerDiffuse), then uses these masks to blend attention outputs between the foreground and background generators (Eqs. 1 and 2). The paper claims that this attention-level blending yields harmonized layers with dynamic interaction between foreground and background, without fine-tuning either model. Qualitative results and a comparison with LayerDiffuse, PP-Matting, MattingAnything, and harmonization methods are presented, along with quantitative CLIP, KID, FID, and user-study results.

Significance. If the central mechanism is correct, the method is a useful contribution: it addresses layered content generation in a training-free way and enables simultaneous, harmonized foreground/background/blended generation, which could benefit creative workflows. The attention-level blending idea using generative priors is interesting, and the qualitative examples are visually suggestive. However, the current empirical support is weakened by two load-bearing issues: the structure prior definition appears mathematically inverted relative to its stated interpretation, and the distributional quantitative evaluation (FID/KID) is circular, comparing outputs against reference distributions produced by the same models used in the pipeline. The user study also lacks significance testing. These issues need to be addressed before the claims can be considered substantiated.

major comments (3)
  1. [3.2, Eqs. (1)-(2)] The structure prior definition is internally inconsistent. For an attention probability row m_i, the quantity Σ_j m_{i,j}^2 is maximized by a one-hot (sparse) row and minimized by a uniform (dense) row; hence s_i = 1/Σ_j m_{i,j}^2 is large for dense rows and small for sparse rows. Calling s_i a 'sparsity score' is therefore backwards, and s'_i = 1 - normalize(s_i) is large for sparse rows, contradicting the claim that s'_i 'favors dense probability distributions over sparse ones.' Since Eqs. (1) and (2) use this mask directly, the inconsistency is load-bearing. The authors should correct the notation/derivation and provide a quantitative validation of the extracted masks against the actual foreground alpha (e.g., IoU with the generated alpha channel) to demonstrate that the masks localize the foreground as claimed.
  2. [4.2 (Foreground & Background Quality)] The FID and KID reference distributions are generated by the very models used in the proposed pipeline: the foreground reference is the output of the LayerDiffuse foreground generator (ϵθ,FG) and the background reference is non-finetuned SDXL (ϵθ). Because both models are part of the pipeline, the reported FID/KID scores measure self-consistency rather than fidelity to real images, and they cannot support the claim of 'closeness to the real imaging distribution.' The authors should report FID/KID against a real-image dataset (e.g., COCO or OpenImages), or at minimum against reference distributions produced by independent models not used in the pipeline, and include confidence intervals.
  3. [4.2 (User Study)] The user study reports mean ratings of 2.960 ± 0.692 for LayerDiffuse and 3.233 ± 0.566 for the proposed method, but no significance test is reported. With overlapping standard deviations, the claim that 'our results receive higher ratings for more satisfying results' is not statistically established. Please add a paired significance test (e.g., the Wilcoxon signed-rank test) over participant/triplet pairs and report the test statistic and p-value.
minor comments (6)
  1. [3.2] The equation defining s_i is typeset incorrectly in the text ('si = 1PM j=1 m2 i,j'); it should be s_i = 1 / Σ_{j=1}^M m_{i,j}^2.
  2. [4.2] Using the first pooling layer of Inception-V3 for FID is non-standard; FID is typically computed on the final pooling layer. Please clarify which feature layer is used and justify the choice.
  3. [4.2] The phrase 'real imaging distribution' in the quantitative section is misleading because the references are model-generated; please rephrase to 'reference generator distributions' or equivalent.
  4. [4.2] The user study description in the supplementary material would benefit from details on how individual participant ratings were aggregated (e.g., per-triplet averages) and whether the order of methods was randomized.
  5. [4.1.4] The ablations for the soft decision boundary coefficient, mask extraction timestep, and attention layer are qualitative only; quantitative results (e.g., CLIP score, mask IoU) would strengthen the justification of these hyperparameter choices.
  6. [1] There is a typo in 'Profilic' (should be 'Prolific') in Section 4.2.

Circularity Check

2 steps flagged · score 6.0 of 10

Quantitative FID/KID comparison is self-referential (the reference distribution is generated by the paper's own input models), and the <EOS> content prior is imported from overlapping-author prior work; the core attention-blending derivation is itself not circular.

  1. self definitional [Section 4.2, 'Foreground & Background Quality' paragraph]
    "Using the images generated by the foreground generator of [25] and backgrounds generated by non-finetuned SDXL as the real imaging distributions, we quantitatively compare our generations in terms of prompt alignment with the CLIP score ([14]), and the closeness to the real imaging distribution with KID ([1]) and FID ([7]) scores."

    The FID/KID reference distribution is defined as the output of the two exact models that generate the method's outputs: the foreground generator of [25] (epsilon_theta_FG) and non-finetuned SDXL (epsilon_theta). Therefore the measured 'closeness to the real imaging distribution' is a self-consistency score, not a comparison to an external real-image distribution. The paper's own conclusion that the background 'aligns better with the RGB diffusion model' is partly true by construction, because the background and blended images are produced by that same RGB diffusion model with attention-level blending.

  2. ansatz smuggled in via citation [Section 3.2, 'Retrieving Content Confidence Priors']
    "Utilizing the unidirectional nature of CLIP Text Encoder, we extract the content confidence map from <EOS> attention probability map, to accumulate all information related to the foreground, following the observations presented in [24]."

    The claim that the <EOS> cross-attention map accumulates all foreground information is load-bearing, because the soft and hard blending masks in Eqs. (1)-(2) are built from the content confidence prior c extracted from that <EOS> map. The only cited support for this premise is [24], whose author list overlaps with the present paper (Yesiltepe, Dalva, Yanardag). The present paper provides no external, independent validation of the <EOS> observation beyond its own Fig. 3, so the central mask construction inherits an ansatz through a self-citation rather than from an independent theorem or benchmark. This is a secondary circularity: the core blending equation is not circular, but one of its key generative priors is imported from the authors' own prior work on faith.

full rationale

The central inference-time mechanism of the paper--Eqs. (1) and (2)--is not circular: it starts from fixed pretrained generators (the LayerDiffuse foreground model and non-finetuned SDXL), extracts attention-derived masks, and defines a blending rule whose outputs are not identical to the mask inputs. The qualitative comparisons and the user study provide evidence that is independent of the method's own equations. The main circularity is in the quantitative evaluation. Section 4.2 sets the FID/KID reference distribution to be images generated by the same foreground generator of [25] and the same non-finetuned SDXL that the method uses as its generators, so the reported 'closeness to the real imaging distribution' is a self-consistency score. The conclusion that the background distribution 'aligns better with the RGB diffusion model' is partly true by construction, since the background and blended images are generated with that very model. A second, weaker circularity is the import of the <EOS> content-prior observation from [24], which shares authors with the present paper and is not independently validated here. Separately, Sec. 3.2's sparsity score is internally inconsistent (s_i = 1/sum_j m_ij^2 is large for dense probability rows and small for sparse rows, so the label 'sparsity' and the subsequent negation are inverted); this is a correctness risk rather than a circularity. The supplementary limitation A.2 concedes the framework 'relies heavily on the quality of the cross-attention and self-attention masks,' which is consistent with the mask-inversion concern. Overall, the core harmonization method is not a derivation that collapses into its inputs, but one of the headline quantitative claims (distribution alignment with the base models) is evaluated against the paper's own input models, meriting a partial-circularity score of 6.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The method adds no new learned parameters but depends on two pretrained models (LayerDiffuse and SDXL) and on several hand-chosen attention heuristics: the structure prior layer, the EOS token, the decision boundary d, and the 0.8T timestep. The quantitative evaluation partially runs in a circle by using model-generated images as reference distributions.

free parameters (3)
  • soft decision boundary coefficient d = 10
    Chosen via qualitative ablation in Sec 4.1.4; controls the hardness of mask_hard in Eq. (2).
  • mask extraction timestep = t = 0.8T
    Masks are retrieved at 80% of the denoising schedule, stated in the Fig. 3 caption but not swept.
  • structure prior attention layer = up.1.attns.2.block.1
    Selected in supplementary A.3 based on visual inspection; affects the boundary quality.
assumptions (5)
  • domain assumption LayerDiffuse's latent transparency model yields a valid alpha channel for a single centered foreground object
    Inherited from reference [25] and used as the foreground generator without modification.
  • domain assumption Self-attention probability sparsity identifies foreground structure in the latent diffusion model
    Sec 3.2: 'we interpret the density of the distribution ... as a vote on whether that variable is a foreground.'
  • domain assumption The <EOS> token cross-attention map accumulates all foreground prompt information
    Sec 3.2, citing the observations in reference [24].
  • domain assumption Attention-output blending preserves both base models' output distributions
    Sec 3.3 claims this without proof; the method assumes swapping attention vectors does not break the learned manifold.
  • domain assumption Attention outputs from the foreground model can be mixed into the RGB model without architectural alignment issues
    Eq. (1) and Eq. (2) mix attention outputs from two different models; the latent spaces are assumed compatible.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LayerFusion: Harmonized Multi-Layer Text-to-Image Generation with Generative Priors." pith.science (2026). https://pith.science/paper/MPZG3A65

@misc{pith2026241204460,
  author       = {Pith},
  title        = {Pith review of: LayerFusion: Harmonized Multi-Layer Text-to-Image Generation with Generative Priors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MPZG3A65}},
  note         = {Machine review of arXiv:2412.04460}
}
read the original abstract

Large-scale diffusion models have achieved remarkable success in generating high-quality images from textual descriptions, gaining popularity across various applications. However, the generation of layered content, such as transparent images with foreground and background layers, remains an under-explored area. Layered content generation is crucial for creative workflows in fields like graphic design, animation, and digital art, where layer-based approaches are fundamental for flexible editing and composition. In this paper, we propose a novel image generation pipeline based on Latent Diffusion Models (LDMs) that generates images with two layers: a foreground layer (RGBA) with transparency information and a background layer (RGB). Unlike existing methods that generate these layers sequentially, our approach introduces a harmonized generation mechanism that enables dynamic interactions between the layers for more coherent outputs. We demonstrate the effectiveness of our method through extensive qualitative and quantitative experiments, showing significant improvements in visual coherence, image quality, and layer consistency compared to baseline methods.

Figures

Figures reproduced from arXiv: 2412.04460 by the authors.

Figure 1
Figure 1. LayerFusion. We propose a framework for generating a foreground (RGBA), background (RGB) and blended (RGB) image simultaneously from an input text prompt. By introducing an optimization-free blending approach that targets the attention layers, we introduce an interaction mechanism between the image layers (i.e., foreground and background) to achieve harmonization during blending. Furthermore, as our framework benefi… view at source ↗
Figure 2
Figure 2. LayerFusion Framework. By making use of the gen￾erative priors extracted from transparent generation model ϵθ,F G, LayerFusion is able to generate image triplets consisting a fore￾ground (RGBA), a background, and a blended image. Our frame￾work involves three fundamental components that are connected with each other. First we introduce a prior pass on ϵθ,F G (a) for extracting the structure prior, and then introduce… view at source ↗
Figure 3
Figure 3. Visualization of the masks extracted as generative priors. Throughout the generation process, we extract a structure prior s and a content confidence prior c. To combine the structure and content information, we construct masksoft and maskhard during the blending process. As visible from the provided maps (as priors), We can both capture the overall object structure with the structure prior s and incorporate the con… view at source ↗
Figures from the paper (19 more)
Figure 4
Figure 4. Figure 4: Qualitative Results. We present qualitative results on multi-layer generation over different visual concepts. In each column, we show the high-quality results of foreground layer, background layer and their generative blending respectively, in terms of text-image align…
Figure 5
Figure 5. Figure 5: We perform extensive ablation studies on the effect of (a) [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Qualitative Comparisons on Layered Generation. We compare our proposed framework with [25] to evaluate the performance in terms of layered image generation (e.g. foreground, background, blended). It clearly shows that [25] propagates the background completion issues ob…
Figure 7
Figure 7. Figure 7: Comparisons with Foreground Extraction Methods To illustrate the advantage of our method over the task of foreground extraction given a blended image, we qualitatively compare our approach with LayerDiffuse ([25]), Matting Anything ([10]), and PPMatting ([2]). As also …
Figure 8
Figure 8. Figure 8: Comparisons on Image Harmonization. We qualitatively evaluate our methods blending capabilities by comparing with image harmonization methods Harmonizer ([9]), INR-Harmonization ([3]), and PCT-Net ([6]). Our proposed generative blending approach results in adaptation o…
Figure 9
Figure 9. Figure 9: As it can also be observed visually, the structure prior extracted from the last self attention layer provides a more precise estimate of the shape of the foreground being “a man, standing” High down.1.attns.1.block.1 Low Foreground down.2.attns.0.block.9 down.2.attns.…
Figure 10
Figure 10. Figure 10: Example Question from the User Study. To evaluate the effectiveness our method perceptually, we conduct a user study over 40 generated image triplets. We provide an example question from this study for clarity. The users are shown an image triplet in the order of fore…
Figure 11
Figure 11. Figure 11: Supplementary Generation Results with animal subjects. Supplementary results with image resolution 896x1152. The foreground & background prompt pairs from left to right are: “a lynx”, “a snowy forest”), (“a crab”, “a rocky tide pool”), (“a duck”, “a village pond”), (“…
Figure 12
Figure 12. Figure 12: Supplementary Generation Results with animal subjects. Supplementary results with image resolution 1024x1024. The foreground & background prompt pairs from left to right are: (“a monkey”, “a vibrant tropical rainforest”), (“a rabbit”, “a backyard garden”), (“a hedgeho…
Figure 13
Figure 13. Figure 13: Supplementary Generation Results with stylization prompts. We provide additional examples with stylization prompts to demonstrate the harmonization capabilities of our method. For each image triplet, we generate the examples with the prompt set (“a man, standing”, “a …
Figure 14
Figure 14. Figure 14: Supplementary Generation Results for “comics” style. To demonstrate the stylization capabilities of our layer harmonization approach, we provide additional results with the background prompt “a street, comics style”. The resolution is 896x1152 for all of the images. 1…
Figure 15
Figure 15. Figure 15: Supplementary Generation Results with human subjects. We provide additional examples with human subjects with different background prompts. The background prompts used are “a rainy jungle”, “a forest in fire”, “a street, winter time”, “a street, daytime”. Note that de…
Figure 16
Figure 16. Figure 16: Supplementary Generation Results for the background “a rainy forest”. For each of the images, the background prompt ”a rainy forest” is used to generate the background image. As it can also be observed from the provided examples, the background creates an influence ov…
Figure 17
Figure 17. Figure 17: Supplementary Generation Results for subjects with transparency property. To demonstrate that our framework is able to preserve the transparency properties of layered image representations, we provide additional results here. With the background prompt ”a table” we us…
Figure 18
Figure 18. Figure 18: Supplementary Generation Results for the subject ”a campfire”. We provide additional generation results for the foreground prompt “a campfire” and background prompt “a beach, night time.” The image resolution is 896x1152 for all examples. 20 [PITH_FULL_IMAGE:figures/…
Figure 19
Figure 19. Figure 19: Supplementary Generation Results for the subject “a book”. We provide additional generation results for the foreground prompt “a book” and background prompt “a table”. The image resolution is 896x1152 for all examples. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_19.png]
Figure 20
Figure 20. Figure 20: Supplementary Generation Results for the subject “a candle”. We provide additional generation results for the foreground prompt “a candle” and background prompt “a dark cave”. The image resolution is 896x1152 for all examples. 22 [PITH_FULL_IMAGE:figures/full_fig_p02…
Figure 21
Figure 21. Figure 21: Supplementary Generation Results for Grounding and Shadowing Effects. We provide additional generation examples to demonstrate the grounding and shadowing capabilities of our framework. Our approach succeeds in both appropriate lighting compared to alpha blending (see…
Figure 22
Figure 22. Figure 22: Supplementary Generation Results Demonstrating Harmonization Capabilities. We provide additional generation exam￾ples to demonstrate the harmonization capabilities of our approach. In each row, we provide triplets that are generated with the same initial seed, which t…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 4 Pith papers

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

  1. LaRender: Training-Free Occlusion Control in Image Generation via Latent Rendering

    cs.CV 2025-08 conditional novelty 7.0 of 10

    LaRender replaces cross-attention layers in a pretrained diffusion model with a latent alpha-compositing operation that renders object features in occlusion order, giving training-free occlusion control.

  2. LoRAShop: Training-Free Multi-Concept Image Generation and Editing with Rectified Flow Transformers

    cs.CV 2025-05 conditional novelty 7.0 of 10

    LoRAShop localizes each LoRA's effect to attention-derived spatial masks inside a Flux transformer, enabling training-free multi-concept image generation and editing.

  3. TKG-DM: Training-free Chroma Key Content Generation Diffusion Model

    cs.CV 2024-11 conditional novelty 6.0 of 10

    Adjusting the mean of specific channels in the initial noise of Stable Diffusion produces foreground objects on a uniform, user-selected chroma key background without any fine-tuning.

  4. Text-Conditioned Background Generation for Editable Multi-Layer Documents

    cs.CV 2025-12 conditional novelty 5.0 of 10

    A training-free system combines soft latent masking, WCAG-contrast-optimized semi-transparent text backings, and recursive LLM summaries to generate readable, style-consistent backgrounds for multi-page documents.

Reference graph

Works this paper leans on

25 extracted references · 13 canonical work pages · cited by 4 Pith papers

  1. [1]

    Demystifying mmd gans

    Mikołaj Bi ´nkowski, Danica J Sutherland, Michael Arbel, and Arthur Gretton. Demystifying mmd gans. In International Conference on Learning Representations, 2018. 9

    work page 2018

  2. [2]

    Pp-matting: High-accuracy natural image matting

    Guowei Chen, Yi Liu, Jian Wang, Juncai Peng, Yuying Hao, Lutao Chu, Shiyu Tang, Zewu Wu, Zeyu Chen, Zhiliang Yu, et al. Pp-matting: High-accuracy natural image matting. arXiv preprint arXiv:2204.09433, 2022. 2, 5, 8 9

    arXiv 2022

  3. [3]

    Dense pixel-to-pixel harmonization via con- tinuous image representation

    Jianqi Chen, Yilan Zhang, Zhengxia Zou, Keyan Chen, and Zhenwei Shi. Dense pixel-to-pixel harmonization via con- tinuous image representation. IEEE Transactions on Circuits and Systems for Video Technology, pages 1–1, 2023. 5, 9

    work page 2023

  4. [4]

    Scaling recti- fied flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas M ¨uller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling recti- fied flow transformers for high-resolution image synthesis. In Forty-first International Conference on Machine Learn- ing, 2023. 2

    work page 2023

  5. [5]

    Implicit style-content separation using b-lora

    Yarden Frenkel, Yael Vinker, Ariel Shamir, and Daniel Cohen-Or. Implicit style-content separation using b-lora. arXiv preprint arXiv:2403.14572, 2024. 11

    arXiv 2024

  6. [6]

    Pct-net: Full resolution image harmonization using pixel-wise color transformations

    Julian Jorge Andrade Guerreiro, Mitsuru Nakazawa, and Bj¨orn Stenger. Pct-net: Full resolution image harmonization using pixel-wise color transformations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5917–5926, 2023. 5, 9

    work page 2023

  7. [7]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium

    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. 9

    work page 2017

  8. [8]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. 2

    2020

Show all 25 references
  1. [9]

    Zhanghan Ke, Chunyi Sun, Lei Zhu, Ke Xu, and Ryn- son W.H. Lau. Harmonizer: Learning to perform white-box image and video harmonization. In European Conference on Computer Vision (ECCV), 2022. 5, 9

    2022

  2. [10]

    Matting anything

    Jiachen Li, Jitesh Jain, and Humphrey Shi. Matting anything. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1775–1785, 2024. 5, 8

    2024

  3. [11]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision, pages 4195–4205,

  4. [12]

    Sdxl: Improving latent diffusion models for high-resolution image synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis. In The Twelfth Inter- national Conference on Learning Representations, 2023. 2, 5

    2023

  5. [13]

    Alfie: Democratising rgba image generation with no $$$

    Fabio Quattrini, Vittorio Pippi, Silvia Cascianelli, and Rita Cucchiara. Alfie: Democratising rgba image generation with no $$$. arXiv preprint arXiv:2408.14826, 2024. 2

    2024 arXiv

  6. [14]

    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. In International conference on machine learning, ...

    2021

  7. [15]

    Zero-shot text-to-image generation

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea V oss, Alec Radford, Mark Chen, and Ilya Sutskever. Zero-shot text-to-image generation. In International confer- ence on machine learning, pages 8821–8831. Pmlr, 2021. 2

    2021

  8. [16]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 2

    2022

  9. [17]

    Photorealistic text-to-image diffusion models with deep language understanding

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding. Advances in neural information...

    2022

  10. [18]

    Denois- ing diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denois- ing diffusion implicit models. In International Conference on Learning Representations, 2020. 2

    2020

  11. [19]

    Score-based generative modeling through stochastic differential equa- tions

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions. In International Conference on Learning Represen- tations, 2020. 2

    2020

  12. [20]

    Rethinking the inception archi- tecture for computer vision

    Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the inception archi- tecture for computer vision. In Proceedings of the IEEE con- ference on computer vision and pattern recognition , pages 2818–2826, 2016. 9

    2016

  13. [21]

    Mulan: A multi layer anno- tated dataset for controllable text-to-image generation

    Petru-Daniel Tudosiu, Yongxin Yang, Shifeng Zhang, Fei Chen, Steven McDonagh, Gerasimos Lampouras, Ignacio Iacobacci, and Sarah Parisot. Mulan: A multi layer anno- tated dataset for controllable text-to-image generation. In Proceedings of the IEEE/CVF Conference on Computer Vi...

    2024

  14. [22]

    Attention is all you need

    A Vaswani. Attention is all you need. Advances in Neural Information Processing Systems, 2017. 2

    2017

  15. [23]

    Diffusers: State-of-the-art diffu- sion models

    Patrick von Platen, Suraj Patil, Anton Lozhkov, Pedro Cuenca, Nathan Lambert, Kashif Rasul, Mishig Davaadorj, Dhruv Nair, Sayak Paul, William Berman, Yiyi Xu, Steven Liu, and Thomas Wolf. Diffusers: State-of-the-art diffu- sion models. https://github.com/huggingface/ diffusers...

    2022

  16. [24]

    The curi- ous case of end token: A zero-shot disentangled image edit- ing using clip

    Hidir Yesiltepe, Yusuf Dalva, and Pinar Yanardag. The curi- ous case of end token: A zero-shot disentangled image edit- ing using clip. arXiv preprint arXiv:2406.00457, 2024. 4

    2024 arXiv

  17. [25]

    a man, standing

    Lvmin Zhang et al. Transparent image layer diffusion using latent transparency. arXiv preprint arXiv:2402.17113, 2024. Last revised 23 Jun 2024. 2, 3, 5, 7, 8, 9, 11 10 LayerFusion: Harmonized Multi-Layer Text-to-Image Generation with Generative Priors Supplementary Material A...

    2024 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.