Pith. sign in

REVIEW 3 major objections 3 minor 39 references

Making Images from Images: Interleaving Denoising and Transformation

T0 review · 3 major / 3 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read The paper claims that interleaving diffusion denoising with a dynamic tile-matching step lets any fixed source image, such as the Mona Lisa, be rearranged into a new image matching a text prompt.

desk verdict A genuinely new dynamic-matching twist on Visual Anagrams, with strong qualitative results, but the latent-space 2% rollout mixing leaves the central interleaving claim under-supported. read the letter →

arxiv 2411.15925 v1 pith:JPOKLIZ2 submitted 2024-11-24 cs.CV cs.AIcs.GRcs.LGcs.NE

classification cs.CVcs.AIcs.GRcs.LGcs.NE
keywords diffusionmodelsimagetransformationtilepermutationdynamicmatchingopticalillusionstext-to-imagelatentHungarianalgorithm
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

This paper tries to show that a completely new image can be made from an existing one by only rearranging its tiles, and that the rearrangement rule can be discovered at the same time as the image is generated. The proposed method interleaves diffusion denoising, which produces image content from a text prompt, with an energy-minimization step that re-solves the tile-to-tile assignment at every iteration. If it works, any pre-existing artwork can be repurposed into any subject described by a prompt, and finer tile grids make the problem easier rather than harder, the reverse of previous illusion-generation systems. The authors demonstrate the recipe in pixel space and in latent diffusion, and extend it to rotations of concentric rings, flips, multiple copies of the source, and multiple source images.

What carries the argument

The load-bearing object is the interleaved dynamic matching loop: at each diffusion step, a rollout lookahead produces an idealized image for each prompt, and a Kuhn-Munkres assignment solves $\min_P \operatorname{tr}(PD)$ to pair tiles by L2 pixel distance, yielding the next permutation $\psi_{t-1}$. This turns a previously static, hand-chosen transform into a learned, iteration-dependent one; for a fixed source image $\beta$, rollout mixing $w^{\text{image}}_t \Psi^i_t(\beta) + w^{\text{noise}}_t \epsilon$ converts the noise-free source into a valid diffusion input.

What would settle it

Run the fixed-source pipeline on the same prompts and seeds while replacing the rollout with the current noisy image or with random permutations; if the resulting images keep matching prompts as well on CLIP scores, the rollout is not doing the claimed work. Alternatively, sweep the rollout length from 1 to 50 and check whether quality is flat; a flat curve would contradict the claim that lookahead is the mechanism.

Watch

Extended reading notes

Core claim

The central claim is that the transform between images need not be fixed in advance: a permutation of image tiles can be learned jointly with the image content, enabling a static source image to become a prompt-specified subject. The paper formulates this as a constrained optimization and solves it by alternating diffusion steps with dynamic matching: a short rollout diffusion run predicts an idealized denoised image, the idealized image is matched to the source tiles by the Hungarian algorithm, and the resulting permutation is used in the next diffusion step. For a fixed source, the mainline image is simply the permuted source, so the only free variables are the permutations themselves. In latent diffusion, rollouts run the full 50-step process from a 2% mixture of image and noise, and matching is done after decoding to pixels. The authors report that larger tile counts improve results and that the same interleaving works for concentric-ring rotations and flips, not only tile permutations.

Load-bearing premise

The whole procedure stands on the assumption that a short rollout from the current image reliably predicts what the final denoised image will look like; the lookahead length of 5 steps and the latent mixing ratio of 2% are empirical choices with no guarantee of convergence.

Editorial extensions

If this is right

  • Any fixed image can serve as the palette: the Mona Lisa, Starry Night, or a user's own photo can be turned into a new subject using only its exact tiles.
  • Finer tile grids give more degrees of freedom, so quality improves as the number of tiles grows from 8x8 to 64x64, opposite to static-transform baselines.
  • The same alternating denoising and assignment procedure extends beyond tile permutations to independent rotations of concentric rings and to flips and rotations at multiple resolutions.
  • With c copies of the source, possibly infinite, tiles may be reused, and multiple different source images can be mixed, enabling finer control over which source tiles appear.
  • In latent diffusion, dynamic matching plus full-length rollouts yields lower CLIP distance to prompts than pixel-space DeepFloyd results, meaning the method carries over to modern systems.

Reading between the lines

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

  • The rollout lookahead is effectively a planning step: it guesses where each prompt would go before committing a permutation. The same pattern could be applied to other discrete generative constraints, such as collage layout or puzzle assembly, wherever a cheap forward model exists.
  • Because more tiles improve quality, the method's limit may be set by the matching optimizer rather than image content; very large tile counts would test whether Hungarian matching remains tractable and whether perceptual distances beat L2.
  • The constraint-propagation effect visible with multiple prompts suggests the method could be used to co-generate a set of images that share hidden structure, potentially for visual cryptography or adversarial illusions for vision models.
  • A direct ablation replacing rollouts with the current noisy image would isolate whether the lookahead or the matching itself is responsible for the gains; the paper's design suggests lookahead is essential but does not prove it.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 3 minor

Summary. The paper proposes an iterative method for generating images by rearranging tiles of a fixed source image. It builds on Visual Anagrams and interleaves diffusion denoising with dynamic matching (Hungarian algorithm) on rollout images to update tile permutations. The method is demonstrated in both pixel space (DeepFloyd) and latent space (Stable Diffusion 2.1), with extensions to multiple copies of the source, multiple source images, and alternative transformations such as concentric rings and flips. The main claims are that dynamic matching improves over static matching, that a pre-specified source image can be transformed into a novel subject, and that increasing the number of tiles makes the problem easier.

Significance. If the central claims hold, the paper is a useful contribution to controlled diffusion-based image generation and optical-illusion synthesis. The general idea of interleaving diffusion steps with an energy-minimization matching step is interesting and goes beyond previous tile-permutation work by removing the need for pre-specified transformations. The paper includes a large quantitative study (1200 CLIP-distance trials with Wilcoxon tests) and a broad set of qualitative demonstrations, which are strengths. However, the quantitative evidence is thin for the headline fixed-source latent-diffusion claim, and the iterative feedback mechanism is not directly validated in that setting.

major comments (3)
  1. [Section 4, Eq. (2), Figure 10] The central fixed-source latent claim depends on the 2% rollout mixing ratio, yet the paper provides no measurement that the current source permutation Ψ_t(β) influences the idealized images Q_i^t. At 2% image weight, the rollout input is dominated by fresh noise and the prompt, so Q_i^t may be effectively independent of Ψ_t(β); in that case the iterative dynamic matching is a one-shot approximation of β by a prompt-conditioned diffusion sample, and the interleaving loop is not doing the claimed work. Please add a direct ablation: vary the mixing ratio, compare final outputs after one iteration versus many iterations, and/or measure how much source information survives (for example, by correlating final tiles with the source permutation). This is load-bearing because the headline result is that a fixed source image can be rearranged into a novel subject.
  2. [Section 2.1] The quantitative evaluation reports statistically significant but extremely small CLIP-distance improvements (22.39 vs 22.43 for 4×4 tiles and 22.36 vs 22.40 for 8×8 tiles) without effect sizes, variances, or a discussion of practical significance. More importantly, this evaluation is for the two-prompt dynamic-matching case, not for the fixed-source latent-diffusion setting that is the paper's main claim. The fixed-source results in Section 4.1 are supported only by qualitative figures, with no comparison against a one-shot baseline (for example, generate a prompt-conditioned image and directly match β's tiles to it). Please add quantitative evaluation for the fixed-source latent case, including such a baseline.
  3. [Section 3, penultimate paragraph] The assertion that with a fixed β, N prompts need not be solved simultaneously and that independent solves yield 'equivalent results' is made without proof or demonstration. This is logically important: if true, the parallel constraint propagation shown in Figure 13 is not needed for fixed-source tasks; if false, the multi-prompt fixed-source results are not supported. Please provide a derivation or a controlled comparison between simultaneous and independent solves.
minor comments (3)
  1. [Section 4.1] The final sentence says the final outputs are the images Ψ_0(x_0), but Section 3 eliminated x_t for the fixed-source formulation; this should be clarified or corrected.
  2. [Figure 12] The comparison uses different prompt prefixes ('a painting' for Visual Anagrams, 'a photograph of' for the proposed method), which confounds the qualitative comparison.
  3. [Throughout] There are typos such as 'the easier it is our for system' (Section 1) and 'Analagous' (Section 4.1), and the abstract's claim that any existing image can be transformed is stronger than Section 3's acknowledged limitation for low-color sources.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the interleaving algorithm iteratively optimizes a permutation against diffusion rollouts; no prediction reduces by construction to a fitted parameter.

full rationale

The paper's derivation chain is an iterative optimization, not a closed-form derivation. The permutation ψ_t is updated by dynamic matching between tiles of the source β and the rollout output Q_t, where Q_t is produced by a diffusion process seeded with the current Ψ_t(β) and conditioned on the external prompt. The source image and prompt are inputs; the CLIP distance used for quantitative evaluation is an external metric against the prompts, not the fitted matching cost. No equation defines the predicted output as the fitted value by construction: the lookahead length l=5 and the 2% rollout mixing ratio are empirical hyperparameters, not fitted parameters renamed as predictions. There are no self-citations or author-imported uniqueness theorems; the only cited prior framework (Visual Anagrams, [12]) is external and used as a baseline. The 'chicken-and-egg' passage in Section 2.1 is an explicit statement of the joint optimization that the alternating algorithm addresses; it is not a circular definition. Concerns that the 2% mixing ratio may make latent rollouts nearly independent of the current source permutation, or that convergence is only empirically observed, are correctness and robustness issues rather than circularity.

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

The method rests on pre-trained diffusion models (DeepFloyd, SD 2.1) and standard assignment optimization; no new physical entities are proposed. The main free parameters are the rollout lookahead length, the latent rollout mixing ratio, and the number of mainline steps, all selected empirically and stated in the text.

free parameters (3)
  • rollout lookahead steps l = 5
    Number of forward diffusion steps used in the rollout lookahead; chosen experimentally (Section 2.1: 'Numerous values were tested for l and, experimentally, we found a lookahead of 5 steps was sufficient').
  • rollout mixing ratio = 2%
    Weight of the image vs noise when initializing latent-diffusion rollouts; the paper says ranges of 1-4% were most often successful and chose 2% (Section 4).
  • mainline steps T = 15
    Number of mainline diffusion steps for latent diffusion; set to 15 as empirically sufficient (Section 4).
assumptions (3)
  • domain assumption Diffusion models (DDPM) with classifier-free guidance produce images matching text prompts
    The method relies on DeepFloyd and SD 2.1 pre-trained models as an oracle for generating idealized images during rollouts (Sections 2, 4).
  • domain assumption L2 distance between image tiles is an adequate measure of tile similarity
    Dynamic matching uses L2 pixel distance to assign tiles; the authors note perceptual metrics could be used but do not (Section 2.1).
  • ad hoc to paper The rollout lookahead approximates the final denoised image well enough for matching
    Introduced and tuned for this paper; no theoretical justification, only empirical choice of l=5 and full 50-step rollouts (Sections 2.1, 4).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Making Images from Images: Interleaving Denoising and Transformation." pith.science (2026). https://pith.science/paper/JPOKLIZ2

@misc{pith2026241115925,
  author       = {Pith},
  title        = {Pith review of: Making Images from Images: Interleaving Denoising and Transformation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JPOKLIZ2}},
  note         = {Machine review of arXiv:2411.15925}
}
read the original abstract

Simply by rearranging the regions of an image, we can create a new image of any subject matter. The definition of regions is user definable, ranging from regularly and irregularly-shaped blocks, concentric rings, or even individual pixels. Our method extends and improves recent work in the generation of optical illusions by simultaneously learning not only the content of the images, but also the parameterized transformations required to transform the desired images into each other. By learning the image transforms, we allow any source image to be pre-specified; any existing image (e.g. the Mona Lisa) can be transformed to a novel subject. We formulate this process as a constrained optimization problem and address it through interleaving the steps of image diffusion with an energy minimization step. Unlike previous methods, increasing the number of regions actually makes the problem easier and improves results. We demonstrate our approach in both pixel and latent spaces. Creative extensions, such as using infinite copies of the source image and employing multiple source images, are also given.

Figures

Figures reproduced from arXiv: 2411.15925 by the authors.

Figure 1
Figure 1. Classic examples of optical illusions. (A) G.Arcimboldo’s Fruit Basket (1590) that shows a face when upright, and a fruit basket when upside-down. (B) Depending on the orientation, this image appears either as a duck or a rabbit. depending on the viewing angle [18, 24] and generating the illusion of motion without movement [11]. Recently, a wide variety of novel work has been conducted in algorithmically creating ph… view at source ↗
Figure 2
Figure 2. Through simple tile permutations, a source image can be converted to a new image of any subject matter. Both the permutation and the content are learned simultaneously; the images created are suited to the tiles available for the composition. Examples with three famous paintings are shown. Each is converted into 3 different subjects (two results are shown for each). The number of tiles that the source is divided int… view at source ↗
Figure 3
Figure 3. A description of a Visual Anagrams step, adapted from [12], using N = 2. A single image that appears as a bowl of fruit can be subdivided into 4x4 square tiles and rearranged into a “smiley face” emoji. To create this, two transforms of the same image (here ψ 1=identity and ψ 2 is a permutation of 4 × 4 tiles) are denoised simultaneously. Two diffusion processes use different prompts to create their per-pixel classi… view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: In each pair, each image is 192×192 pixels. Each image can be subdivided into a grid of 8 × 8 (64 total blocks) of 24×24 pixels. For each pair of images, there is a permutation of the 64 blocks that transforms the left to right and vice-versa. As in [12], the permutati…
Figure 5
Figure 5. Figure 5: Incorporating dynamic matching between each diffusion step. In the green rectangle, the input image is permuted and the rollout occurs. In the orange rectangle, the images created at the end of the rollout are used to compute the new permutations through dynamic matchi…
Figure 6
Figure 6. Figure 6: As the diffusion process continues, changes between successive permutations is reduced. As the images become more consistent (as t → 0) each ψt−1 looks more like ψt. By iteration 15, the 8 × 8 tiles have ceased movement. Movement ends by iteration 9 for the 4 × 4 tiles…
Figure 7
Figure 7. Figure 7: The transformation is dynamically determined as the image is created. Note that the same prompt (Edgar Allan Poe), when paired with different prompts, yields different images because of the constraints imposed by the pairing. Qualitative results with dynamic matching a…
Figure 8
Figure 8. Figure 8: With a fixed image, β, the rollout procedure with an additional noising step is used to replace both the mainline and rollout procedures described in the previous section. In this example, an image of a man with a mustache is created by moving tiles around from Van Gog…
Figure 9
Figure 9. Figure 9: Examples using fixed source images. Top Row is the original image (β), the bottom 6 rows shows the images created by permuting the tiles in β. Examples are shown for 32 × 32 to 8 × 8 tiles; the exact same pixels as the original are used. Results with square crops of fo…
Figure 10
Figure 10. Figure 10: Changes for latent diffusion. Latent representations, with a dashed outline, are used for diffusion and pixel images are used for dynamic matching and calculating Ψ. Latents are decoded with D for dynamic matching then re-encoded with E; D&E are pretrained components …
Figure 11
Figure 11. Figure 11: Examples created using latent diffusion. The output images are 768 × 768 pixels. The left of each image pair is divided into an 8×8 grid of tiles that are permuted to get the image on the right. We repeat the same prompt with different pairs to show the effects of the…
Figure 12
Figure 12. Figure 12: Multiple prompts comparison. The images in a row use the same tiles. The images are 768 × 768 pixels and the tiles are 48 × 48, 24 × 24 or 12 × 12 pixels in image rows 1, 2 and 3, respectively. “No matching” skips the dynamic matching. “Vis. Anagrams” uses the impleme…
Figure 13
Figure 13. Figure 13: A visual demonstrations of how constraints are propagated through the parallel diffusion processes. In these three examples that use 5 prompts, 3 prompts are kept constant, while the last two are changed. The last 2 prompts have strong colors associated with them. Not…
Figure 14
Figure 14. Figure 14: Any number of image copies can be used as the basis. In each of the examples, we look at 1,5 and 10 copies of the original to create the prompt. Top Table: 16×16 tiles and prompts: of “a tea cup” and “a young boy” created with the Mona Lisa and Starry Night. Bottom Ta…
Figure 15
Figure 15. Figure 15: Two transforms are shown here. The first is the Concentric Rings transformation. Here, concentric circles are rotated independently; the original, shown in (a), is rotated with 2 concentric circles, shown in (b). Second, the Flips transformation, allows for horizontal…
Figure 16
Figure 16. Figure 16: Results for the Permutation, Concentric Rings and Flips transformations using similar prompts. Horizontal pairs are transforms of one another using the transformation in the left column [PITH_FULL_IMAGE:figures/full_fig_p019_16.png]
Figure 17
Figure 17. Figure 17: Rollout visualization. The mainline progress is shown on the left, progressing vertically downwards. For each mainline step, a rollout that begins with the current mainline image computes the idealized image for the prompt, “A Man with a Mustache”. Once the idealized …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 33 canonical work pages

  1. [1]

    In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Amir, D., Weiss, Y.: Understanding and simplifying perceptual distances. In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 12226–12235 (2021) 5

  2. [2]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Avrahami, O., Lischinski, D., Fried, O.: Blended diffusion for text-driven editing of natural images. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 18208–18218 (2022) 2

  3. [3]

    Bar-Tal, O., Yariv, L., Lipman, Y., Dekel, T.: Multidiffusion: Fusing diffusion paths for controlled image generation (2023) 3

  4. [4]

    Burgert, R., Li, X., Leite, A., Ranasinghe, K., Ryoo, M.S.: Diffusion illusions: Hiding images in plain sight (2023) 2

  5. [5]

    Chen, Z., Geng, D., Owens, A.: Images that sound: Composing images and sounds on a single canvas (2024) 2

  6. [6]

    IEEE Transactions on Pattern Analysis and Machine Intelligence (2023) 2

    Croitoru, F.A., Hondru, V., Ionescu, R.T., Shah, M.: Diffusion models in vision: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence (2023) 2

  7. [7]

    J Psychol Cognition3(2) (2018) 1

    Das, K.K.: Role of pre-embedded associated memory in generation of optical illusions. J Psychol Cognition3(2) (2018) 1

  8. [8]

    Advances in neural information processing systems34, 8780–8794 (2021) 2

    Dhariwal, P., Nichol, A.: Diffusion models beat gans on image synthesis. Advances in neural information processing systems34, 8780–8794 (2021) 2

Show all 39 references
  1. [9]

    In: Bengio, S., Wallach, H., Larochelle, H., Grauman, K., Cesa- Bianchi, N., Garnett, R

    Elsayed, G., Shankar, S., Cheung, B., Papernot, N., Kurakin, A., Goodfellow, I., Sohl-Dickstein, J.: Adversarial examples that fool both computer vision and time-limited humans. In: Bengio, S., Wallach, H., Larochelle, H., Grauman, K., Cesa- Bianchi, N., Garnett, R. (eds.) Adv...

  2. [10]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    Everaert, M.N., Bocchio, M., Arpa, S., Süsstrunk, S., Achanta, R.: Diffusion in style. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 2251–2261 (October 2023) 2

  3. [11]

    ACM Siggraph Computer Graphics25(4), 27–30 (1991) 2

    Freeman, W.T., Adelson, E.H., Heeger, D.J.: Motion without movement. ACM Siggraph Computer Graphics25(4), 27–30 (1991) 2

  4. [12]

    arXiv:2311.17919 (November 2023),https://arxiv

    Geng, D., Park, I., Owens, A.: Visual anagrams: Generating multi-view optical illusions with diffusion models. arXiv:2311.17919 (November 2023),https://arxiv. org/abs/2311.17919 2, 4, 5, 6, 11, 14

  5. [13]

    Geng, D., Park, I., Owens, A.: Factorized diffusion: Perceptual illusions by noise decomposition (2024) 2, 3

  6. [14]

    arXiv preprint arXiv:1412.6572 (2014) 1

    Goodfellow, I.J., Shlens, J., Szegedy, C.: Explaining and harnessing adversarial examples. arXiv preprint arXiv:1412.6572 (2014) 1

  7. [15]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Hendrycks, D., Zhao, K., Basart, S., Steinhardt, J., Song, D.: Natural adversarial examples. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 15262–15271 (June 2021) 1

  8. [16]

    Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. In: Ad- vances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020, NeurIPS 2020, December 6-12, 2020, virtual (2020), https : / / proceedings . neu...

  9. [17]

    arXiv:2207.12598 (2022) 4

    Ho, J., Salimans, T.: Classifier-free diffusion guidance. arXiv:2207.12598 (2022) 4

  10. [18]

    ACM Trans

    Hsiao, K.W., Huang, J.B., Chu, H.K.: Multi-view wire art. ACM Trans. Graph. 37(6), 242 (2018) 2 Making Images from Images: Interleaving Denoising and Transformation 23

  11. [19]

    Jaini, P., Clark, K., Geirhos, R.: Intriguing properties of generative classifiers (2024) 1

  12. [20]

    Naval research logistics quarterly2(1-2), 83–97 (1955) 4, 6

    Kuhn, H.W.: The hungarian method for the assignment problem. Naval research logistics quarterly2(1-2), 83–97 (1955) 4, 6

  13. [21]

    Liu, N., Li, S., Du, Y., Torralba, A., Tenenbaum, J.B.: Compositional visual generation with composable diffusion models (2023) 2

  14. [22]

    UCLA USJ24 (2011) 1

    Lo, C., Dinov, I.: Investigation of optical illusions on the aspects of gender and age. UCLA USJ24 (2011) 1

  15. [23]

    https://midjourney.com (2024), accessed: 2024-5-22 2

    Midjourney: Midjourney. https://midjourney.com (2024), accessed: 2024-5-22 2

  16. [24]

    ACM Transactions on Graphics28(5), 156–1 (2009) 2

    Mitra, N.J., Pauly, M.: Shadow art. ACM Transactions on Graphics28(5), 156–1 (2009) 2

  17. [25]

    Journal of the society for industrial and applied mathematics5(1), 32–38 (1957) 4, 6

    Munkres, J.: Algorithms for the assignment and transportation problems. Journal of the society for industrial and applied mathematics5(1), 32–38 (1957) 4, 6

  18. [26]

    Ngo, J., Sankaranarayanan, S., Isola, P.: Is CLIP fooled by optical illusions? (2023), https://openreview.net/forum?id=YdGkE4Ugg2C 1

  19. [27]

    arXiv preprint arXiv:2112.10741 (2021) 2

    Nichol,A.,Dhariwal,P.,Ramesh,A.,Shyam,P.,Mishkin,P.,McGrew,B.,Sutskever, I., Chen, M.: Glide: Towards photorealistic image generation and editing with text- guided diffusion models. arXiv preprint arXiv:2112.10741 (2021) 2

  20. [28]

    Oliva,A.,Torralba,A.,Schyns,P.G.:Hybridimages.ACMTransactionsonGraphics (TOG) 25(3), 527–532 (2006) 1, 2

  21. [29]

    In: International Conference on Machine Learning

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from natural language supervision. In: International Conference on Machine Learning. pp. 8748–8763. PMLR (2021) 6, 8

  22. [30]

    In: International Conference on Machine Learning

    Ramesh, A., Pavlov, M., Goh, G., Gray, S., Voss, C., Radford, A., Chen, M., Sutskever, I.: Zero-shot text-to-image generation. In: International Conference on Machine Learning. pp. 8821–8831. PMLR (2021) 2

  23. [31]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10684–10695 (2022) 11

  24. [32]

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models (2021) 2, 9

  25. [33]

    https://doi.org/10.48550/ARXIV.2205.11487 , https: //arxiv.org/abs/2205.11487 2

    Saharia, C., Chan, W., Saxena, S., Li, L., Whang, J., Denton, E., Ghasemipour, S.K.S., Ayan, B.K., Mahdavi, S.S., Lopes, R.G., Salimans, T., Ho, J., Fleet, D.J., Norouzi, M.: Photorealistic text-to-image diffusion models with deep language understanding (2022). https://doi.org...

  26. [34]

    Advances in neural information processing systems35, 36479–36494 (2022) 2

    Saharia, C., Chan, W., Saxena, S., Li, L., Whang, J., Denton, E.L., Ghasemipour, K., Gontijo Lopes, R., Karagol Ayan, B., Salimans, T., et al.: Photorealistic text- to-image diffusion models with deep language understanding. Advances in neural information processing systems35,...

  27. [35]

    Shonenkov, A., Konstantinov, M., Bakshandaeva, D., Schuhmann, C., Ivanova, K., Klokova, N.: If by deepfloyd lab at stabilityai.https://github.com/deep-floyd/IF (2023), accessed: 2024-5-22 2, 4

  28. [36]

    arXiv preprint arXiv:2010.02502 (2020) 2

    Song, J., Meng, C., Ermon, S.: Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502 (2020) 2

  29. [37]

    Tancik, M.: Illusion-diffusion.https://github.com/tancik/Illusion-Diffusion (2023), accessed: 2024-5-22 2

  30. [38]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Tumanyan, N., Geyer, M., Bagon, S., Dekel, T.: Plug-and-play diffusion features for text-driven image-to-image translation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 1921–1930 (2023) 2 24 S. Baluja et al

  31. [39]

    New Ideas in Psychology36, 1–9 (2015) 1

    de Wit, M.M., van der Kamp, J., Withagen, R.: Visual illusions and direct percep- tion: Elaborating on gibson’s insights. New Ideas in Psychology36, 1–9 (2015) 1

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.