Pith. sign in

REVIEW 3 major objections 5 minor 52 references

Personalized Representation from Personalized Generation

T0 review · 3 major / 5 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 personalized representations trained with synthetic data, using only three real examples of an instance, outperform their pretrained counterparts across datasets, backbones, and downstream tasks.

desk verdict Worth refereeing: a practical recipe and new benchmark for few-shot personalized representation, with an honest limitations section; the mechanism claim needs one control experiment, and the intro overstates 'universal' gains. read the letter →

arxiv 2412.16156 v1 pith:ZTFICTCR submitted 2024-12-20 cs.CV cs.LG

Shobhita Sundaram , Julia Chae , Yonglong Tian , Sara Beery , Phillip Isola This is my paper · ORCID
classification cs.CVcs.LG
keywords personalizedrepresentationsyntheticdataDreamBoothcontrastivelearningInfoNCEfew-shotinstance-levelrecognitionPODSbenchmark
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 asks whether a few real photos of one object can be turned into a reusable personalized representation—a feature space specialized to that instance—by training on synthetic images of it. The authors' central claim is that contrastively fine-tuning a general-purpose vision backbone with InfoNCE on images generated by a DreamBooth-adapted text-to-image model, using only three real examples, outperforms the same pretrained backbone on classification, retrieval, detection, and segmentation of that instance. They introduce an evaluation suite with the new PODS dataset and reformulated DeepFashion2 and DogFaceNet splits, and they analyze which generator choices, such as classifier-free guidance, LLM-written prompts, masking, and cut-and-paste, matter most. The paper reports improvements in 32 of 36 backbone-task-dataset combinations, with large relative gains on detection for DINOv2 and on classification for CLIP.

What carries the argument

The machinery is the pairing of a personalized generator with a contrastive representation objective. A DreamBooth-adapted text-to-image diffusion model, conditioned on an identifier token and LLM-written prompts, produces synthetic positives of the target instance; the same model prompted with the generic category produces synthetic negatives. The InfoNCE loss then operates on the concatenated CLS and average-pooled patch features of a pretrained vision transformer adapted with LoRA, pulling real anchors toward synthetic positives and away from category negatives. This pairing carries the claim: the quality of the learned representation is set by the identity fidelity and diversity of the synthetic positives, which the paper measures through a DreamSim-based fidelity-diversity analysis.

What would settle it

Swap the synthetic positives for images of a different instance of the same category while keeping diversity and prompt distribution matched; if one-vs-all PR-AUC on real test images still rises, the paper's identity-preservation mechanism is not what carries the gain.

Watch

Extended reading notes

Core claim

Personalized representations trained with synthetic data, using only three real examples of an instance, significantly outperform their pretrained counterparts across datasets, backbones, and downstream tasks. The recipe is: fine-tune a text-to-image diffusion model with DreamBooth on three real photos of the instance; generate a pool of synthetic positives with diverse prompts and a category-level pool of negatives; LoRA-fine-tune a pretrained vision encoder with the InfoNCE loss so that real anchors and synthetic positives are pulled together and category negatives are pushed apart. The resulting features improve one-vs-all classification, retrieval, detection, and segmentation for that instance without training task-specific heads. The paper also finds that combining masked DreamBooth with cut-and-paste gives the best training data, that LLM-generated captions help global tasks more than dense tasks, and that different generators leave complementary biases—DreamBooth handles pose but confuses similar instances, while cut-and-paste keeps identity but overfits pose.

Load-bearing premise

Everything rests on DreamBooth, trained from three real photos, producing synthetic positives that keep the target instance's identity well enough for contrastive training to transfer to real test images; the paper's own hard-negative failures in Section D.2 show that when identity drifts, the representation can encode spurious generator features instead.

Editorial extensions

If this is right

  • With only three real images and no task-specific heads, a user can obtain an instance-tuned encoder that improves classification, retrieval, detection, and segmentation over the same pretrained encoder.
  • The paper's scaling experiment shows gains persist as real data grows: synthetic augmentation adds 27% classification accuracy at three real images and 8% at twenty, so generation complements rather than only substitutes for collected data.
  • Plugging the personalized encoder into a perSAM-style segmentation pipeline raises F1 scores on all three datasets, indicating the representation transfers to existing downstream methods.
  • Because generator choice changes which failures remain—pose robustness versus fine-grained discrimination—users can select a generator or combine them to match their deployment distribution.

Reading between the lines

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

  • If identity-preserving generation continues to improve, the practical minimum for personalizing a vision model could drop below three real images, and the same contrastive recipe might transfer to other modalities, such as audio or 3D object instances, wherever generative models exist.
  • The hard-negative failure cases the paper reports imply that adding a strict fidelity filter on synthetic positives—rejecting generations whose perceptual similarity to the real anchors falls below a threshold—could further narrow the gap between synthetic and fully real training.
  • A privacy-friendly consequence follows: because only three local images and a pretrained generator are needed, personalized models could be built without uploading user data or real negatives, keeping personalization siloed on the user's device.
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 / 5 minor

Summary. The paper proposes a method for learning instance-specific ("personalized") visual representations from very few real images. For a target instance, a DreamBooth model is fine-tuned on three real images, a pool of synthetic images is generated with varying prompts and CFG scales, and a pretrained vision encoder (DINOv2, CLIP, or MAE) is contrastively fine-tuned via LoRA using the synthetic images as positives and generic-category synthetic images as negatives. The authors introduce a new dataset, PODS, and reformulate DeepFashion2 and DogFaceNet for instance-level evaluation across classification, retrieval, detection, and segmentation. They report improvements over pretrained baselines in 32 of 36 settings, compare generation methods (Masked DreamBooth, Cut-and-Paste, Combined), and show that the learned representations can be used within the PerSAM pipeline. The paper is thorough in its ablations and releases code, data, and detailed metadata.

Significance. If the central mechanism holds, the paper demonstrates a novel and practically appealing use of personalized generation for representation learning in data-scarce, fine-grained settings, and it provides a reusable evaluation suite (PODS plus reformulated datasets) that should be valuable to the community. The empirical study is unusually broad: three backbones, three datasets, four tasks, multiple generation methods, and an application to PerSAM. The strengths include careful dataset construction, transparent reporting of full results in the appendix, and public code/data release. The main caveat is that the paper's explanatory claim—that identity-preserving personalized generation is what drives the gains—is not directly tested, leaving an alternative explanation (category-level cues or generator artifacts) open.

major comments (3)
  1. [Section 3.4, Eq. (1)] The InfoNCE loss as written is internally inconsistent: the paragraph defines the tuple as (x, x+, x0, ..., xN) with x the real anchor, x+ the synthetic positive, and xi for i=0,...,N the synthetic negatives, but the displayed equation uses x0 as the anchor in both the numerator and the denominator. The numerator should use the real anchor x (or the notation should be changed so that the anchor is named x0). As printed, the equation does not define a valid contrastive loss for the stated setup. Please correct the equation and the surrounding notation.
  2. [Section 3.2 / Section 3.4] The central claim is that personalized generation from the target instance is what enables the learned representation to transfer to real test images. However, no experiment isolates the role of identity preservation. The paper compares against real-image-only training, but that is not a control for the generator's personalization: the positives in "Real Imgs" are the actual real images, not non-personalized synthetic images. A proper control would replace the DreamBooth positives with synthetic images generated without instance personalization, e.g., images of the same category from the pretrained generator, or DreamBooth images personalized to a different instance. Section D.2 itself documents failure cases where similarity to hard negatives increases and attributes them to "noisy positives," but the frequency of these failures is never quantified. Without such a control, the aggregate gains in Tables 1 and 7-8 could be explained by the synthetic data providing category-level or artifact-level cues rather than identity-specific information, undermining the advertised mechanism.
  3. [Introduction / Section 5.1] The Introduction states that performance "universally improves," while Section 5.1 says improvements occur in "32/36 cases." These statements are inconsistent, and the full results show clear exceptions: MAE retrieval regresses on PODS (MAE-P 30.7 vs. pretrained 34.6) and DF2 (23.7 vs. 25.8), and several CLIP/MAE dense-task cells show near-zero or negative changes. Please correct the overclaim and explicitly discuss the conditions under which the proposed method fails to improve over the pretrained backbone.
minor comments (5)
  1. [Section 5.1] The relative improvements quoted in the text (e.g., "DINOv2 detection improves by 48%") are not defined as relative or absolute; please state the computation to avoid ambiguity.
  2. [Appendix A.3.2] The DreamSim filtering threshold is described as "empirically chosen" (0.6 for DF2/PODS, 0.55 for Dogs); please specify how this was selected (e.g., on a validation set) so the reader can assess overfitting risk.
  3. [Appendix C.4] The fidelity metric in Figure 12 measures DreamSim similarity between synthetic images and the mean embedding of real images; as the caption notes, this can be influenced by background features. The main text should acknowledge this limitation or report a more identity-focused fidelity measure.
  4. [Tables 7-8] The min/max ranges are reported only for the highlighted best rows; the other rows are presumably single-seed runs. Please clarify this in the caption or provide ranges for all rows, since several differences are small relative to the reported ranges.
  5. [Section 4.2] The same validation set is used both for hyperparameter sweeps and for selecting the best synthetic dataset per backbone. A brief note on the selection protocol (e.g., no test-set information used) would strengthen the reproducibility statement.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: the central claim is evaluated on held-out real test images; the few self-citations (StableRep, DreamSim) are non-load-bearing tools and do not make the result circular.

full rationale

The paper's central claim is empirical and externally grounded: a vision encoder is fine-tuned with InfoNCE on synthetic images generated by a DreamBooth model adapted from three real images, and then evaluated on held-out real test images across classification, retrieval, detection, and segmentation. The InfoNCE objective in Section 3.4 pulls real anchors toward synthetic positives, but nothing in that equation guarantees improved similarity to unseen real test images, so the reported gains are not forced by construction. Hyperparameters are swept on a validation set and the final numbers are reported on test splits, which is disclosed and standard. The DreamSim-based filtering in Section A.3.2 and the fidelity analysis in Section C.4 use a similarity model co-authored by present authors, but the main Table 1 results use unfiltered DreamBooth data, so the self-citation is not load-bearing. The StableRep citation in Sections 2 and 3.4 is also used only as inspiration for contrastive training on synthetic data; the paper's contribution and evaluation do not assume StableRep's correctness. The manuscript itself flags the most important validity threat in Section D.2: noisy synthetic positives can increase similarity to hard negatives, meaning some gains may be driven by spurious features rather than preserved identity. That is a correctness or mechanism concern, not a circularity concern, because the evaluation remains on real, held-out test images. Section 6.1 also acknowledges computational cost and inherited generator biases, again without introducing circularity. Overall, no prediction in the paper reduces to its inputs by definition, and no load-bearing premise is justified only by a self-citation chain.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The method is empirical; the ledger lists hyperparameters tuned on validation and the domain assumptions about generator fidelity, transfer, label quality, and pretrained backbones. No new physical or theoretical entities are introduced. The PODS dataset is a new empirical resource, not a postulated entity.

free parameters (5)
  • Number of synthetic positives |D_S| = 450
    Ablated on the validation set; performance plateaus at 450 generated positives, which is then used in all main experiments.
  • Anchor-positive pair count = 4500
    Validation sweep fixes a 1:10 ratio between the number of synthetic positives and sampled anchor-positive pairs; this ratio is used throughout.
  • CFG guidance scale = 4.0, 5.0, 7.5; best per backbone chosen on validation
    Classifier-free guidance is swept to control diversity of generated images; Table 7 shows the chosen best varies by backbone.
  • DreamBooth filtering threshold (DreamSim cosine) = 0.6 for DF2/PODS, 0.55 for Dogs
    Empirically chosen threshold in Appendix A.3.2 for filtering low-fidelity generated images before training.
  • LoRA hyperparameters = rank 16, alpha 0.5, dropout 0.3, learning rate 0.0003
    Fixed after validation experiments; standard fine-tuning choices listed in Appendix B.1.
assumptions (4)
  • domain assumption DreamBooth fine-tuning on three real images yields synthetic images that share instance identity with the target.
    The entire synthetic-positive pool in Section 3.2 is built on this premise. The paper acknowledges noisy positives in Section D.2, showing the assumption can fail.
  • domain assumption Contrastive fine-tuning on synthetic pairs transfers to real test images.
    The InfoNCE training in Section 3.4 assumes the synthetic-to-real domain gap is small enough for the representation to generalize to held-out real data.
  • domain assumption Instance-level labels and masks in PODS, DF2, and Dogs splits are accurate.
    PODS masks are partly manually annotated with TORAS and SAM, DogFaceNet required manual cleanup, and all evaluations depend on these labels.
  • domain assumption Pretrained backbones provide a suitable starting point for personalization.
    The method fine-tunes DINOv2, CLIP, and MAE with low-rank adaptation, assuming their generic features are a sufficient base for instance-level contrastive adaptation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Personalized Representation from Personalized Generation." pith.science (2026). https://pith.science/paper/ZTFICTCR

@misc{pith2026241216156,
  author       = {Pith},
  title        = {Pith review of: Personalized Representation from Personalized Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZTFICTCR}},
  note         = {Machine review of arXiv:2412.16156}
}
read the original abstract

Modern vision models excel at general purpose downstream tasks. It is unclear, however, how they may be used for personalized vision tasks, which are both fine-grained and data-scarce. Recent works have successfully applied synthetic data to general-purpose representation learning, while advances in T2I diffusion models have enabled the generation of personalized images from just a few real examples. Here, we explore a potential connection between these ideas, and formalize the challenge of using personalized synthetic data to learn personalized representations, which encode knowledge about an object of interest and may be flexibly applied to any downstream task relating to the target object. We introduce an evaluation suite for this challenge, including reformulations of two existing datasets and a novel dataset explicitly constructed for this purpose, and propose a contrastive learning approach that makes creative use of image generators. We show that our method improves personalized representation learning for diverse downstream tasks, from recognition to segmentation, and analyze characteristics of image generation approaches that are key to this gain.

Figures

Figures reproduced from arXiv: 2412.16156 by the authors.

Figure 1
Figure 1. Learning personalized representations from limited real data. In this paper we explore whether and how synthetic data can be used to train a personalized representation. Given a few real images of an instance, we generate novel images and contrastively fine-tune a general-purpose pretrained model to learn a personalized representation, useful for diverse downstream tasks. ABSTRACT Modern vision models excel at gener… view at source ↗
Figure 2
Figure 2. Personalized Representa￾tion Training Pipeline. Our three-stage training method: 1) Generative Model Training 2) Synthetic Data Generation 3) Contrastive LoRA Fine-Tuning. We generate personalized data from DR using Stable Dif￾fusion 1.5, a T2I model, as our generator gθ. We adapt gθ using DreamBooth (Ruiz et al., 2022) to generate novel images of c when conditioned on an identifier token. A T2I diffusion model gθ g… view at source ↗
Figure 3
Figure 3. (left) Examples of instances from our new PODS dataset. We showcase one example instance from each of the five object categories, displaying images from both the training and var￾ious test splits. We dim the surrounding scene, highlighting the instance of interest. This masking technique is not applied to our dataset images or during training. (right) We show example gen￾erated images from Dreambooth (LLM, cfg 5), w… view at source ↗
Figures from the paper (25 more)
Figure 4
Figure 4. Figure 4: Inference Pipelines. We visualize the global (classification, retrieval) and local (detection, segmentation) evaluation pipelines. Global inference uses cosine similarity between CLS embed￾dings, while local inference extracts patch features with spatial information. p…
Figure 5
Figure 5. Figure 5: Qualitative Results. Each triplet shows the test image (left), dense prediction maps for pretrained DINOv2 (center), and personalized (right). Prediction maps are computed via patchwise embedding similarity between the test and localized train images following [PITH_F…
Figure 7
Figure 7. Figure 7: DreamBooth vs Cut and Paste Model Fail￾ures. We show object pairs where DB-personalized and CP-personalized models differ most in predictions. Cut and Paste Masked DreamBooth [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 6
Figure 6. Figure 6 [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 9
Figure 9. Figure 9: DB, CP, and Combined per￾formance on PODS test splits. DB out￾performs CP on the Pose split, and un￾derperforms it on the Distractors split, whereas Combined performs best on both. These results support our qualitative obser￾vations. We also study the impact of fidelit…
Figure 10
Figure 10. Figure 10: DF2 train/test examples. Training images are of models wearing clothes, and test images are from consumers. Images and classes are randomly sampled. Train DOGS Dataset Test [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]
Figure 11
Figure 11. Figure 11: Dogs train/test images. Images and classes are randomly sampled. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_11.png]
Figure 12
Figure 12. Figure 12: Diversity and fidelity plot colored by accuracy across PODS, DF2 and Dogs synthetic datasets. Note that the fidelity metric may be influenced by the background features appearing in the cropped image, resulting in a reduced fidelity score for some samples. To better u…
Figure 13
Figure 13. Figure 13: DF2 Dense Predictions Images and classes are randomly sampled [PITH_FULL_IMAGE:figures/full_fig_p027_13.png]
Figure 14
Figure 14. Figure 14: Dogs Dense Predictions Images and classes are randomly sampled D.2 CHALLENGING EXAMPLES We visualize examples of hard negatives and hard positives from the Dogs dataset to better under￾stand the capabilities of personalized representations. For a given query image x o…
Figure 15
Figure 15. Figure 15: PODS Dense Predictions Images and classes are randomly sampled induced by noisy positives in the synthetic training dataset, leading the personalized representation to associate the target object with spurious features. Hard Positives in the Dogs Dataset [PITH_FULL_I…
Figure 16
Figure 16. Figure 16: Dogs Hard Positives. Given images of a target dog (leftmost of each row) we identify the positive test images with the lowest DINOv2 similarity to the query. DINOv2-P cosine similarity typically increases, even for cases with significant differences in setting, camera…
Figure 17
Figure 17. Figure 17: Dogs Hard Negatives. Given images of a target dog (leftmost of each row) we identify the negative test images with the highest DINOv2 similarity to the query. In some cases DINOv2-P cosine similarity decreases (top row) however we also identify failure cases where cos…
Figure 18
Figure 18. Figure 18: PODS Generated Images - DreamBooth without LLM prompting. Images and classes are randomly sampled [PITH_FULL_IMAGE:figures/full_fig_p030_18.png]
Figure 19
Figure 19. Figure 19: PODS Generated Images - DreamBooth with LLM prompting. Images and classes are randomly sampled 30 [PITH_FULL_IMAGE:figures/full_fig_p030_19.png]
Figure 20
Figure 20. Figure 20: PODS Generated Images - DreamBooth with LLM, Masking and Filtering. Images and classes are randomly sampled [PITH_FULL_IMAGE:figures/full_fig_p031_20.png]
Figure 21
Figure 21. Figure 21: PODS Generated Negatives. Images and classes are randomly sampled. Each row are sampled negatives for each object category. 31 [PITH_FULL_IMAGE:figures/full_fig_p031_21.png]
Figure 22
Figure 22. Figure 22: DF2 Generated Images - DreamBooth without LLM Images and classes are randomly sampled [PITH_FULL_IMAGE:figures/full_fig_p032_22.png]
Figure 23
Figure 23. Figure 23: DF2 Generated Images - DreamBooth with LLM Images and classes are randomly sampled 32 [PITH_FULL_IMAGE:figures/full_fig_p032_23.png]
Figure 24
Figure 24. Figure 24: DF2 Generated Images - DreamBooth with LLM, Masking and Filtering. Images and classes are randomly sampled [PITH_FULL_IMAGE:figures/full_fig_p033_24.png]
Figure 25
Figure 25. Figure 25: DF2 Generated Negatives. Images and classes are randomly sampled 33 [PITH_FULL_IMAGE:figures/full_fig_p033_25.png]
Figure 26
Figure 26. Figure 26: Dogs Generated Images - DreamBooth without LLM prompting. Images and classes are randomly sampled [PITH_FULL_IMAGE:figures/full_fig_p034_26.png]
Figure 27
Figure 27. Figure 27: Dogs Generated Images - DreamBooth with LLM prompting. Images and classes are randomly sampled 34 [PITH_FULL_IMAGE:figures/full_fig_p034_27.png]
Figure 28
Figure 28. Figure 28: Dogs Generated Images - DreamBooth with LLM, Masking and Filtering. Images and classes are randomly sampled [PITH_FULL_IMAGE:figures/full_fig_p035_28.png]
Figure 29
Figure 29. Figure 29: Dogs Generated Negatives. Images and classes are randomly sampled 35 [PITH_FULL_IMAGE:figures/full_fig_p035_29.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 52 canonical work pages

  1. [1]

    Filtered DogFaceNet dog classes to keep classes with above 10 images per instance

    work page

  2. [2]

    Performed a random train-test split for every instance, keeping 3 images for train and rest for test

    work page

  3. [3]

    This entailed looking through the training and test data and making sure that no test images were from the same sequential footage as the train data

    Manually inspected every instance in the dataset to remove data poisoning. This entailed looking through the training and test data and making sure that no test images were from the same sequential footage as the train data. When such images were discovered, they were removed from the test set

    work page

  4. [4]

    canonical pose

    After data-cleanup, we removed instances with less then 4 remaining test images. The above procedure resulted in 80 total dog classes. Below is the metadata of our subselected dataset. • Unique Instances Selected: 80 • Total # Training Images: 240 • # of Training Images per Instance: 3 • Total # Test Images: 1218 • Range of Test Images per Instance: [6, 3...

    work page 2023

  5. [5]

    A <new1> mug on a wooden desk

    work page

  6. [6]

    A <new1> mug in a cozy living room

    work page

  7. [7]

    A <new1> mug on a windowsill • Bottles

    work page

  8. [8]

    A <new1> bottle on a picnic table

    work page

Show all 52 references
  1. [9]

    A <new1> bottle in a backpack pocket

  2. [10]

    A <new1> bottle on a yoga mat • Screwdrivers

  3. [11]

    A <new1> screwdriver in a toolbox

  4. [12]

    A <new1> screwdriver on a wooden workbench

  5. [13]

    A <new1> screwdriver in a carpenter’s tool belt • Totes (Bags)

  6. [14]

    A <new1> bag in a car trunk

  7. [15]

    A <new1> bag on a park bench

  8. [16]

    A <new1> bag in a shopping cart

  9. [17]

    A <new1> bag on a library shelf

  10. [18]

    A <new1> bag in a gym locker

  11. [19]

    A <new1> bag on a wooden table • Shoes

  12. [20]

    A <new1> shoe in the rain

  13. [21]

    A <new1> shoe on a sandy beach 20

  14. [22]

    A <new1> shoe in a gym locker DF2 Dataset

  15. [23]

    A person wearing a <new1> shirt at a park

  16. [24]

    A <new1> shirt on a mannequin

  17. [25]

    A person wearing a <new1> shirt at a party

  18. [26]

    A <new1> shirt on a clothesline

  19. [27]

    A person wearing a <new1> shirt at a concert

  20. [28]

    A <new1> shirt on a chair

  21. [29]

    A person wearing a <new1> shirt at a caf´e

  22. [30]

    A <new1> shirt on a laundry basket

  23. [31]

    A person wearing a <new1> shirt at a stadium

  24. [32]

    A <new1> shirt on a hook

  25. [33]

    A person wearing a <new1> shirt at a bus stop

  26. [34]

    A <new1> shirt on a drying rack

  27. [35]

    A person wearing a <new1> shirt at a gym

  28. [36]

    A <new1> shirt on a shelf

  29. [37]

    A person wearing a <new1> shirt at a picnic Dogs Dataset

  30. [38]

    A <new1> dog in the park

  31. [39]

    A <new1> dog at the vet

  32. [40]

    A <new1> dog in a car

  33. [41]

    A <new1> dog at the groomer

  34. [42]

    A <new1> dog on a walk

  35. [43]

    A <new1> dog in the snow

  36. [44]

    A <new1> dog at the lake

  37. [45]

    A <new1> dog in the backyard

  38. [46]

    A <new1> dog at the <new1> dog park

  39. [47]

    A <new1> dog in a sweater

  40. [48]

    A <new1> dog in a bed

  41. [49]

    A <new1> dog at the farm

  42. [50]

    A <new1> dog in the woods

  43. [51]

    A <new1> dog in a kennel

  44. [52]

    <new1>" specifi- cation from every prompt: e.g

    A <new1> dog at a barbecue A.3.2 M ASKED DREAM BOOTH - FILTERING We apply automatic filtering to the Masked DreamBooth pipeline as an additional data-processing step that we can take when masks are available, to ensure high-quality generated data. We use the masks to extract a...

    2023

Pith tools

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