Pith. sign in

REVIEW 3 major objections 4 minor 123 references

How to Take a Memorable Picture? Empowering Users with Actionable Feedback

T0 review · 3 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read A training-free steering method gets AI to give photo advice that raises memorability.

desk verdict Useful new task and benchmark; the central claim about human memorability is credible but rests on a self-referential evaluation loop, so it needs careful refereeing and better evidence. read the letter →

arxiv 2602.21877 v2 pith:GTLDGM4X submitted 2026-02-25 cs.CV

Francesco Laiti , Davide Talon , Jacopo Staiano , Elisa Ricci This is my paper
classification cs.CV
keywords imagememorabilityactionablefeedbackmultimodallargelanguagemodelsactivationsteeringteacher-studentdistillationMemBenchCoachphotographyguidance
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

The paper claims that memorability—how likely a photo is to be remembered—can be not just predicted but instructed. It introduces Memorability Feedback (MemFeed), in which a model tells a photographer in plain language what to change in the current shot, and MemCoach, a training-free teacher-student activation steering method that shifts a multimodal vision-language model's internal activations toward patterns associated with more memorable images. On the new MemBench benchmark, the steered model's suggestions, when applied by an editing model, raise the edited photo's predicted memorability more often and by a larger margin than zero-shot prompting, across four open MLLMs. The authors read this as evidence that memorability can be taught and instructed, moving the field from passive prediction to actionable guidance at capture time.

What carries the argument

The central mechanism is a memorability steering vector: the per-layer average of activation differences between student responses to memorability-aware feedback and student responses to neutral feedback, computed on paired examples from the MemBench training split. At inference, the vector is added to the student's activations at a chosen layer with a strength coefficient α. The paper also builds MemBench, pairing least-to-most memorable photos within 1,570 scenes and captioning the transformation between them, which supplies the contrasting data and the evaluation protocol.

What would settle it

A repeat-detection human memory study on a held-out set: if people remember MemCoach-edited photos no better than zero-shot-edited photos while M scores them higher, the paper's central claim is falsified. A less expensive check is to swap in an independently trained regressor and see whether the gains disappear; the paper's cross-predictor experiment is a first step, but human recall is decisive.

Watch

Extended reading notes

Core claim

MemCoach distills the difference between neutral feedback and feedback that a teacher model gives when shown how to go from a less memorable to a more memorable shot of the same scene. The difference is captured as a steering vector in the student model's activation space and injected at inference time. The paper reports that this alone—no fine-tuning—raises the improvement ratio from 0.73 to 0.80 and the relative memorability gain from 5.47% to 7.21% on InternVL3.5, with consistent gains across Qwen2.5-VL, Idefics3, and LLaVA-OV. The authors also run a preliminary human memory experiment and a human-in-the-loop phone-camera study, which they report as consistent with the automated metrics.

Load-bearing premise

The load-bearing premise is that the memorability regressor M used to rank images, generate pairs, and score edits reflects genuine human memory; if M is biased, the improvements measure alignment with M, not with human recall.

Editorial extensions

If this is right

  • Because MemCoach is training-free, it can be applied to any MLLM that exposes intermediate activations, and the paper shows consistent improvement-ratio gains across four architectures.
  • The method closes most of the gap to teacher-oracle models that see the high-memorability image, while beating aesthetics-specialist models trained on photographic critique.
  • Steering requires very little data: with 1% of the training scenes, MemCoach matches full-data fine-tuning on the improvement ratio, pointing to a cheap recipe for perceptual feedback tasks.
  • The authors suggest the same activation-steering route can endow MLLMs with perceptual skills beyond memorability, such as interactive visual guidance.

Reading between the lines

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

  • Deployment inference: a live memorability coach in a phone viewfinder would let photographers get suggestions before pressing the shutter, which is the capture-time use case the paper motivates but only tests in a prototype app.
  • Evaluation inference: because the automated metrics define improvement through the same regressor that ranks the benchmark pairs, a natural next test is to ask independent human raters to pick the more memorable image from source-versus-edited pairs, or to re-train the regressor from human annotations on edited images.
  • Scientific inference: the recurring advice to show faces, direct gaze, and symmetric poses suggests a concrete hypothesis—these cues make photos memorable by engaging social attention—that could be tested by measuring gaze patterns on MemCoach-edited versus neutral images.
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 / 4 minor

Summary. The paper introduces Memorability Feedback (MemFeed), a task in which an MLLM produces natural-language, actionable advice for increasing a photograph's memorability. It contributes MemBench, a benchmark built from PPR10K by ranking same-scene images with a memorability regressor M and using a captioning MLLM to verbalize the transformation between low- and high-ranked images, and MemCoach, a training-free activation-steering method that derives a memorability steering vector from paired teacher/student feedback and applies it at inference. Experiments across four open MLLMs report improved improvement ratio (IR) and relative memorability (RM), plus lower perplexity, with ablations and preliminary human studies in the supplement. The central claim is that MemCoach makes MLLMs provide actionable feedback that, when applied, increases image memorability more often and by a larger margin than zero-shot prompting.

Significance. If the empirical claims hold, the paper opens a useful new direction by turning memorability prediction into explainable, actionable guidance; the benchmark and the model-agnostic steering recipe are concrete assets for follow-up work. The manuscript is generally well structured, includes reproducible experimental details (fixed seeds, structured decoding, project page), and reports cross-editor and cross-predictor generalizations. These are real strengths. However, the magnitude of the claimed human-memorability improvement is not yet established: the headline evaluation is tied to the same regressor M used to construct the data, to generate teacher feedback, and to tune hyperparameters, while the independent human evidence is preliminary and reported without inferential statistics. The contribution is therefore promising but needs substantially stronger validation.

major comments (3)
  1. [§3.2, §4.2, Eq. (3), Table 2] Headline evaluation is largely self-referential. MemBench pairs are ranked with M (Supp. A.4); teacher feedback f+ in §4.2 describes transforming an M-low into an M-high image; the steering vector (Eq. 3) is the activation difference between f+ and neutral feedback; and IR/RM in §3.2 are computed with the same M. The +9.59% IR and +31.81% RM for INTERNVL3.5 therefore measure alignment with M's preferences, not directly human memorability. Supp. A.4 acknowledges reliance on M, but M is also the evaluation and hyperparameter-selection signal, so the acknowledgment covers only part of the loop. Table 9 (VM→MB) helps, but ViTMem and M are trained on the same public memorability datasets; Supp. C's human studies (47 subjects in 3 arms, 27 scenes, 28 raters) are preliminary and lack inferential statistics. An independent, adequately powered human memory experiment is needed to support the abst
  2. [§5.1, Tables 2–3, Fig. 8] No uncertainty quantification is reported. Tables 2 and 3 give point estimates only, and Fig. 8 has no error bars. With the held-out split of ~1,570 scenes, an IR gain of 0.73→0.80 and a RM change such as IDEFICS3's 20.19→19.81 (Table 3) may be within noise; perplexity for IDEFICS3 also moves in the wrong direction (6.64→6.69). To support 'consistent' gains, the paper should report standard deviations or confidence intervals over evaluation subsets or repeated runs, plus significance tests on the primary metrics.
  3. [§3.2, §5.1, Table 2] The perplexity axis is self-referential. The ground-truth feedback is generated by INTERNVL3.5, the same model family used as student and default teacher, and the steering vector is distilled from that teacher's output. Lower perplexity therefore indicates better imitation of the teacher's text distribution, not human-likeness; the teacher-oracle rows in Table 2 have much lower perplexity (2.34–5.73) than MemCoach (4.99), consistent with this mechanism. An independent human rating of feedback quality (e.g., expanded from Supp. C.3) or human-written ground-truth feedback should be the primary evidence.
minor comments (4)
  1. [Tables 2–3] The column headers appear corrupted: Tables 2 and 3 list 'Editing Perplexity (↓) IR (↑) RM% (↑)', but the columns are actually IR, RM%, and Perplexity. Please fix the header order.
  2. [Table 2] 'n.d.' is used for perplexity of Edit model and GPT-5 MINI but never defined; define it (presumably 'not determined').
  3. [Fig. 2] The figure contains repeated 'Predictor dss' labels; this appears to be a typo and should be corrected.
  4. [Supp. C.3] The feedback quality study uses Oracle vs MemCoach but does not include a zero-shot baseline; without that comparison, the interpretability/actionability advantage over ordinary prompting cannot be evaluated from this table.

Circularity Check

3 steps flagged · score 7.0 of 10

Primary evaluation is a closed loop: M ranks MemBench pairs, teacher feedback is derived from those pairs, steering is tuned on M-based IR, and IR/RM are computed with the same M; perplexity uses ground truth from the same captioning model. Cross-predictor and small human studies partially mitigate.

  1. self definitional [§3.2 (Data pipeline, Evaluation protocol)]
    "Once images are associated to the corresponding memorability score, photographs within the same scene are ranked and pairs (xS, xD) are constructed from less to more memorable images. ... IMPROVEMENT RATIO (IR) evaluates the fraction of time the edited image has larger memorability than the source one."

    The destination images that define 'more memorable' are chosen by ranking with M, and the evaluation metric IR/RM measures whether edited images score higher under the same M. Therefore, the reported improvement is a measure of how well MemCoach's feedback moves images along M's preference gradient, not an independent test of human memorability. The loop is closed by construction: the training pairs and the evaluation criterion share the same scoring function.

  2. fitted input called prediction [§4.2 (Contrasting data generation, Eq. 3) and §5 (Implementation details)]
    "we leverage the teacher model to extract memorability-aware feedback f i + = ϕ teach(xi S, xi D, pa), with pa the feedback elicitation prompt in Sec. 3, yielding the actionable instructions on how to move from xi S to xi D and consequently, improve the current image memorability. ... We fix the steering layer to l=12 and the coefficient to α=55, selected via tuning on a held-out subset of the training data."

    The steering vector is computed from the difference between teacher feedback (derived from M-ranked pairs) and neutral student feedback, and the steering coefficient/layer are tuned on the M-based IR metric. Thus, MemCoach is explicitly fitted to produce feedback that maximizes the same M-score improvement that is later reported as the headline result. This is a fitted-input-called-prediction loop: the 'prediction' of improved memorability is the optimization target.

1 more flagged steps
  1. self definitional [§3.2 (Evaluation protocol) and §5.1 (Quantitative Results)]
    "we evaluate the likelihood that a model provides improving memorability feedback by computing the PERPLEXITY on ground truth memorability-aware feedback from the same captioning model. ... the lower perplexity achieved by MemCoach confirms its improved alignment with human-like memorability-aware feedback."

    The 'ground truth' feedback used for perplexity is generated by the same captioning model (INTERNVL3.5) that serves as the teacher for steering. MemCoach is steered to imitate that teacher's output, so low perplexity on that distribution is self-consistency with the training signal, not evidence of human-aligned memorability. The paper's interpretation of this as 'human-like' feedback is not supported by an independent ground truth.

full rationale

The central claim that MemCoach produces feedback which increases image memorability rests on a closed evaluation loop. The memorability predictor M is used to rank images within each scene and select the source/destination pairs; the teacher feedback describes how to transform the source into the M-selected destination; the steering vector is extracted from the contrast between this teacher feedback and neutral student feedback; the steering hyperparameters are tuned on M-based IR; and the headline IR/RM metrics are computed with the same M. Thus, the primary quantitative results measure alignment with M's preferences rather than human memorability. The perplexity metric is similarly circular because its 'ground truth' is generated by the same captioning model that provides the teacher signal. The paper includes mitigating evidence: the cross-predictor experiment (VM→MB) shows robustness across two predictors, and the human studies in Supp. C are genuinely independent and show improvement. However, both predictors are trained on the same paradigm of human memorability annotations, and the human studies are small and preliminary (47 participants, 27 scenes, 28 raters), without inferential statistics. Therefore, the evaluation is partially circular: the method is optimized for and evaluated on the same scoring function that defines its training signal, with only limited external validation. No self-citation chain is load-bearing here; the circularity is internal to the benchmark and evaluation design.

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

The central claim depends on the regressor M as a proxy for human memory and on the editing model's fidelity; both are loaded from prior work or assumed. MemCoach's own free hyperparameters (l, α) are tuned to the evaluation metric. No radically new entities are introduced.

free parameters (3)
  • Memorability predictor M (MLP on OpenCLIP ViT-SO400M-14-SigLIP-384 embeddings) = Spearman 0.82 on LaMem; weights not released
    Used to rank images within each scene (Sec 3.2), to extract teacher feedback, and to compute IR/RM metrics (Sec 5.1). Its validity as a proxy for human memorability is the central load-bearing assumption.
  • Steering layer l and coefficient α = l=12, α=55 for InternVL3.5; per-model values in Supp. Table 7
    Selected by tuning on a held-out subset of MemBench training scenes to maximize IR (Sec 5, Implementation details). This fits the method's free hyperparameters to the evaluation metric.
  • Feedback elicitation prompt p_a and memorability prompt p_m = Text given in Sec 3.2 and Supp. A.6
    The wording of these prompts determines the semantic content of the benchmark and the student's default feedback; the paper acknowledges prompt engineering to avoid image-referential phrasing.
assumptions (5)
  • domain assumption Memorability is a stable, image-computable property of visual content
    Adopted from prior work [26-28,32] and used to justify ranking images by a learned regressor (Sec 1, 3.2).
  • domain assumption The learned regressor M is a valid ground-truth proxy for human memorability
    Used to construct source/destination pairs and evaluation labels; the paper states it treats M as a well-established black box (Supp. A.4).
  • domain assumption The editing model e faithfully applies natural-language feedback to images
    IR/RM assume e(x_S, a) implements the feedback; paper tests multiple editing models but cannot guarantee faithful execution (Sec 3.2, Supp. D.4).
  • ad hoc to paper The teacher's description of differences between least and most memorable images constitutes actionable memorability feedback
    MemBench ground truth is generated by prompting INTERNVL3.5 to describe image-pair differences; no human verification that these actions are the cause of the memorability difference (Sec 3.2).
  • domain assumption Activation steering via mean-difference vectors (CAA) transfers to multimodal models and changes output behavior
    Adopted from [56,65,79]; the paper relies on the linear representation hypothesis but does not verify it for the chosen layers (Sec 4.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of How to Take a Memorable Picture? Empowering Users with Actionable Feedback." pith.science (2026). https://pith.science/paper/GTLDGM4X

@misc{pith2026260221877,
  author       = {Pith},
  title        = {Pith review of: How to Take a Memorable Picture? Empowering Users with Actionable Feedback},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GTLDGM4X}},
  note         = {Machine review of arXiv:2602.21877}
}
read the original abstract

Image memorability, i.e., how likely an image is to be remembered, has traditionally been studied in computer vision either as a passive prediction task, with models regressing a scalar score, or with generative methods altering the visual input to boost the image likelihood of being remembered. Yet, none of these paradigms supports users at capture time, when the crucial question is how to improve a photo memorability. We introduce the task of Memorability Feedback (MemFeed), where an automated model should provide actionable, human-interpretable guidance to users with the goal to enhance an image future recall. We also present MemCoach, the first approach designed to provide concrete suggestions in natural language for memorability improvement (e.g., "emphasize facial expression," "bring the subject forward"). Our method, based on Multimodal Large Language Models (MLLMs), is training-free and employs a teacher-student steering strategy, aligning the model internal activations toward more memorable patterns learned from a teacher model progressing along least-to-most memorable samples. To enable systematic evaluation on this novel task, we further introduce MemBench, a new benchmark featuring sequence-aligned photoshoots with annotated memorability scores. Our experiments, considering multiple MLLMs, demonstrate the effectiveness of MemCoach, showing consistently improved performance over several zero-shot models. The results indicate that memorability can not only be predicted but also taught and instructed, shifting the focus from mere prediction to actionable feedback for human creators.

Figures

Figures reproduced from arXiv: 2602.21877 by the authors.

Figure 1
Figure 1. Given an input photo, memorability feedback aims to generate natural-language suggestions to guide users toward a more memorable shot. MemCoach provides memorability-aware feedback, effectively assisting users to capture memorable images. els trained to regress a scalar memorability score from im￾ages [17, 32, 62] and explaining what makes an image mem￾orable [26–28]. These works identified key intrinsic factors suc… view at source ↗
Figure 2
Figure 2. Overview of MemBench generation and evaluation. Top: Data pipeline for constructing MemBench, including scene grouping, memorability regression, image ranking, and generation of actionable memorability-aware feedback. Bottom: Evaluation pipeline assessing feedback quality through editing-based memorability improvement and perplexity scoring. 3. Memorability Feedback In this section, we formally define the task of Me… view at source ↗
Figure 3
Figure 3. MemBench statistics. Data analysis in terms of (a) most frequent words; (b) distribution of memorability scores for the least and most memorable images within each scene; (c) feedback length as measured by content words; and (d) categorization of atomic sub-actions, where the width of each chord indicates the frequency of co-occurrence between categories. using GPT-5-MINI as an automatic annotator and report their c… view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: Overview of the proposed method. (a) Contrasting data generation : paired samples are built by coupling the memorability￾aware guidance of a teacher MLLM with the neutral responses of a student MLLM on the same scene; (b) Steering vector extraction : activation differe…
Figure 6
Figure 6. Figure 6: Common feedback patterns on source images. Mem￾Coach favors symmetric and socially connected compositions, re￾flecting principles of human photography. expressed in concise textual instructions (mostly involving verbs “Bring”, “Stand”, “Remove”) that can be directly im…
Figure 7
Figure 7. Figure 7: Qualitative feedback from MemCoach. For each source image (left), the model provides natural-language feedback (bottom) that is applied to produce the destination image (right). Each score represents the Relative Memorability (RM), indicating how suggested feedback aff…
Figure 8
Figure 8. Figure 8 [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Human memory performance across three different image settings. more memorable images. Experiment setup details. We implemented a mobile app allowing a user to use MemCoach in real-life scenarios. The app presents a live camera viewfinder and allows users to point thei…
Figure 10
Figure 10. Figure 10: A set of qualitative examples from MemBench. [PITH_FULL_IMAGE:figures/full_fig_p019_10.png]
Figure 11
Figure 11. Figure 11: A set of qualitative examples from MemBench. [PITH_FULL_IMAGE:figures/full_fig_p020_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

123 extracted references · 15 linked inside Pith

  1. [1]

    Deep learning using rectified linear units (relu).arXiv preprint arXiv:1803.08375, 2018

    Abien Fred Agarap. Deep learning using rectified linear units (relu).arXiv preprint arXiv:1803.08375, 2018. 13

    arXiv 2018

  2. [2]

    Flamingo: a vi- sual language model for few-shot learning

    Jean-Baptiste Alayrac, Jeff Donahue, Pauline Luc, Antoine Miech, Iain Barr, Yana Hasson, Karel Lenc, Arthur Men- sch, Katie Millicah, Malcolm Reynolds, Roman Ring, Eliza Rutherford, Serkan Cabi, Tengda Han, Zhitao Gong, Sina Samangooei, Marianne Monteiro, Jacob Menick, Sebastian Borgeaud, Andrew Brock, Aida Nematzadeh, Sahand Shar- ifzadeh, Mikolaj Binkow...

    2022

  3. [3]

    Llava-onevision-1.5: Fully open framework for democratized multimodal training.arXiv preprint arXiv:2509.23661, 2025

    Xiang An, Yin Xie, Kaicheng Yang, Wenkang Zhang, Xiuwei Zhao, Zheng Cheng, Yirui Wang, Songcen Xu, Changrui Chen, Chunsheng Wu, et al. Llava-onevision-1.5: Fully open framework for democratized multimodal training.arXiv preprint arXiv:2509.23661, 2025. 5, 6, 7

    arXiv 2025

  4. [4]

    Qwen technical report.arXiv preprint arXiv:2309.16609,

    Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xi- aodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, et al. Qwen technical report.arXiv preprint arXiv:2309.16609,

    arXiv

  5. [5]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. Qwen2. 5-vl technical report.arXiv preprint arXiv:2502.13923, 2025. 5, 6, 7

    arXiv 2025

  6. [6]

    Kv cache steering for controlling frozen llms.arXiv preprint arXiv:2507.08799, 2025

    Max Belitsky, Dawid J Kopiczko, Michael Dorkenwald, M Je- hanzeb Mirza, James R Glass, Cees GM Snoek, and Yuki M Asano. Kv cache steering for controlling frozen llms.arXiv preprint arXiv:2507.08799, 2025. 2

    arXiv 2025

  7. [7]

    FLUX.2 [klein], 2026

    Black Forest Labs. FLUX.2 [klein], 2026. 18

    2026

  8. [8]

    Intrinsic and extrinsic effects on image memorability.Vision research, 2015

    Zoya Bylinskii, Phillip Isola, Constance Bainbridge, Antonio Torralba, and Aude Oliva. Intrinsic and extrinsic effects on image memorability.Vision research, 2015. 1, 2, 7, 13

    2015

Show all 123 references
  1. [9]

    Memorability: An image-computable measure of information utility

    Zoya Bylinskii, Lore Goetschalckx, Anelise Newman, and Aude Oliva. Memorability: An image-computable measure of information utility. InHuman perception of visual infor- mation: Psychological and computational perspectives, 2021. 1, 2

    2021

  2. [10]

    Web-scale visual entity recognition: An llm-driven data approach

    Mathilde Caron, Alireza Fathi, Cordelia Schmid, and Ahmet Iscen. Web-scale visual entity recognition: An llm-driven data approach. InAdvances in Neural Information Processing Systems, 2024. 13

    2024

  3. [11]

    Persona vectors: Monitoring and con- trolling character traits in language models.arXiv preprint arXiv:2507.21509, 2025

    Runjin Chen, Andy Arditi, Henry Sleight, Owain Evans, and Jack Lindsey. Persona vectors: Monitoring and con- trolling character traits in language models.arXiv preprint arXiv:2507.21509, 2025. 2

    2025 arXiv

  4. [12]

    Videomem: Constructing, analyzing, predicting short-term and long-term video memorability

    Romain Cohendet, Claire-H´el`ene Demarty, Ngoc QK Duong, and Martin Engilberge. Videomem: Constructing, analyzing, predicting short-term and long-term video memorability. In Proceedings of the IEEE/CVF international conference on computer vision, 2019. 2

    2019

  5. [13]

    What makes an object memo- rable? InProceedings of the ieee international conference on computer vision, 2015

    Rachit Dubey, Joshua Peterson, Aditya Khosla, Ming-Hsuan Yang, and Bernard Ghanem. What makes an object memo- rable? InProceedings of the ieee international conference on computer vision, 2015. 1, 2

    2015

  6. [14]

    Modular memorability: Tiered representations for video mem- orability prediction

    Th´eo Dumont, Juan Segundo Hevia, and Camilo L Fosco. Modular memorability: Tiered representations for video mem- orability prediction. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, 2023. 2

    2023

  7. [15]

    E, Ohad Fried, Jingwan Lu, Jianming Zhang, Radom´ır Mech, Jose Echevarria, Pat Hanrahan, and James A

    Jane L. E, Ohad Fried, Jingwan Lu, Jianming Zhang, Radom´ır Mech, Jose Echevarria, Pat Hanrahan, and James A. Landay. Adaptive photographic composition guidance. InProceedings of the 2020 CHI Conference on Human Factors in Computing Systems, 2020. 2

    2020

  8. [16]

    Video unlearning via low-rank refusal vector.arXiv preprint arXiv:2506.07891, 2025

    Simone Facchiano, Stefano Saravalle, Matteo Migliarini, Edoardo De Matteis, Alessio Sampieri, Andrea Pilzer, Emanuele Rodol`a, Indro Spinelli, Luca Franco, and Fabio Galasso. Video unlearning via low-rank refusal vector.arXiv preprint arXiv:2506.07891, 2025. 2

    2025

  9. [17]

    Amnet: Memorability estimation with attention

    Jiri Fajtl, Vasileios Argyriou, Dorothy Monekosso, and Paolo Remagnino. Amnet: Memorability estimation with attention. InProceedings of the IEEE conference on computer vision and pattern recognition, 2018. 1, 13, 14

    2018

  10. [18]

    Memcat: a new category-based image set quantified on memorability.PeerJ,

    Lore Goetschalckx and Johan Wagemans. Memcat: a new category-based image set quantified on memorability.PeerJ,

  11. [19]

    Ganalyze: Toward visual definitions of cognitive im- age properties

    Lore Goetschalckx, Alex Andonian, Aude Oliva, and Phillip Isola. Ganalyze: Toward visual definitions of cognitive im- age properties. InProceedings of the ieee/cvf international conference on computer vision, 2019. 1, 2, 7, 16

    2019

  12. [20]

    Level up your photography skills with camera coach,

    Google. Level up your photography skills with camera coach,

  13. [21]

    Image memora- bility prediction with vision transformers.arXiv preprint arXiv:2301.08647, 2023

    Thomas Hagen and Thomas Espeseth. Image memora- bility prediction with vision transformers.arXiv preprint arXiv:2301.08647, 2023. 13, 14, 18

    2023 arXiv

  14. [22]

    Lora: Low-rank adaptation of large language models.ICLR, 2022

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models.ICLR, 2022. 8

    2022

  15. [23]

    Language is not all you need: Aligning perception with language models.Advances in Neural Information Processing Systems, 2023

    Shaohan Huang, Li Dong, Wenhui Wang, Yaru Hao, Saksham Singhal, Shuming Ma, Tengchao Lv, Lei Cui, Owais Khan Mohammed, Barun Patra, et al. Language is not all you need: Aligning perception with language models.Advances in Neural Information Processing Systems, 2023. 2

    2023

  16. [24]

    Aesexpert: Towards multi-modality foundation 9 model for image aesthetics perception

    Yipo Huang, Xiangfei Sheng, Zhichao Yang, Quan Yuan, Zhichao Duan, Pengfei Chen, Leida Li, Weisi Lin, and Guang- ming Shi. Aesexpert: Towards multi-modality foundation 9 model for image aesthetics perception. InProceedings of the 32nd ACM International Conference on Multimedia...

    2024

  17. [25]

    Openclip, 2021

    Gabriel Ilharco, Mitchell Wortsman, Ross Wightman, Cade Gordon, Nicholas Carlini, Rohan Taori, Achal Dave, Vaishaal Shankar, Hongseok Namkoong, John Miller, Hannaneh Ha- jishirzi, Ali Farhadi, and Ludwig Schmidt. Openclip, 2021. 13

    2021

  18. [26]

    Understanding the intrinsic memorability of images.Ad- vances in neural information processing systems, 2011

    Phillip Isola, Devi Parikh, Antonio Torralba, and Aude Oliva. Understanding the intrinsic memorability of images.Ad- vances in neural information processing systems, 2011. 1, 2, 13

    2011

  19. [27]

    What makes an image memorable? InCVPR 2011,

    Phillip Isola, Jianxiong Xiao, Antonio Torralba, and Aude Oliva. What makes an image memorable? InCVPR 2011,

    2011

  20. [28]

    What makes a photograph memorable?IEEE transactions on pattern analysis and machine intelligence,

    Phillip Isola, Jianxiong Xiao, Devi Parikh, Antonio Torralba, and Aude Oliva. What makes a photograph memorable?IEEE transactions on pattern analysis and machine intelligence,

  21. [29]

    Portraid: An ai- driven portrait assistant for professional-quality image compo- sition

    Jaspal Singh Kahlon and Gongbo Liang. Portraid: An ai- driven portrait assistant for professional-quality image compo- sition. InProceedings of the 2025 ACM Southeast Conference,

    2025

  22. [30]

    Image memorability and visual in- ception

    Aditya Khosla, Jianxiong Xiao, Phillip Isola, Antonio Tor- ralba, and Aude Oliva. Image memorability and visual in- ception. InSIGGRAPH Asia 2012 Technical Briefs, 2012. 1

    2012

  23. [31]

    Modifying the memorability of face photographs

    Aditya Khosla, Wilma A Bainbridge, Antonio Torralba, and Aude Oliva. Modifying the memorability of face photographs. InProceedings of the IEEE international conference on com- puter vision, 2013. 2

    2013

  24. [32]

    Understanding and predicting image memorability at a large scale

    Aditya Khosla, Akhil S Raju, Antonio Torralba, and Aude Oliva. Understanding and predicting image memorability at a large scale. InProceedings of the IEEE international conference on computer vision, 2015. 1, 2, 3, 4, 5, 13

    2015

  25. [33]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. InICLR, 2017. 13

    2017

  26. [34]

    Rukiye Savran Kiziltepe, Lorin Sweeney, Mihai Gabriel Con- stantin, Faiyaz Doctor, Alba Garc´ıa Seco de Herrera, Claire- H´el´ene Demarty, Graham Healy, Bogdan Ionescu, and Alan F. Smeaton. An annotated video dataset for computing video memorability.Data in Brief, 2021. 2

    2021

  27. [35]

    Prajneya Kumar, Eshika Khandelwal, Makarand Tapaswi, and Vishnu Sreekumar. Eye vs. ai: Human gaze and model atten- tion in video memorability.arXiv preprint arXiv:2311.16484,

    arXiv

  28. [36]

    Seeing eye to ai: Comparing human gaze and model attention in video memorability

    Prajneya Kumar, Eshika Khandelwal, Makarand Tapaswi, and Vishnu Sreekumar. Seeing eye to ai: Comparing human gaze and model attention in video memorability. In2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2025. 2

    2025

  29. [37]

    Black Forest Labs, Stephen Batifol, Andreas Blattmann, Fred- eric Boesel, Saksham Consul, Cyril Diagne, Tim Dockhorn, Jack English, Zion English, Patrick Esser, et al. Flux. 1 kon- text: Flow matching for in-context image generation and editing in latent space.arXiv preprint a...

    arXiv

  30. [38]

    A survey on image memorability prediction: From traditional to deep learning models

    Souad Lahrache and Rajae El Ouazzani. A survey on image memorability prediction: From traditional to deep learning models. In2022 2nd International Conference on Innovative Research in Applied Science, Engineering and Technology (IRASET), 2022. 2, 13

    2022

  31. [39]

    Building and better understanding vision-language models: insights and future directions

    Hugo Laurenc ¸on, Andr ´es Marafioti, Victor Sanh, and Leo Tronchon. Building and better understanding vision-language models: insights and future directions. InWorkshop on Re- sponsibly Building the Next Generation of Multimodal Foun- dational Models, 2024. 5, 6, 7

    2024

  32. [40]

    Blip- 2: Bootstrapping language-image pre-training with frozen image encoders and large language models

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip- 2: Bootstrapping language-image pre-training with frozen image encoders and large language models. InInternational conference on machine learning. PMLR, 2023. 2

    2023

  33. [41]

    Personality-assisted multi-task learning for generic and personalized image aesthetics assessment.IEEE Transactions on Image Processing, 2020

    Leida Li, Hancheng Zhu, Sicheng Zhao, Guiguang Ding, and Weisi Lin. Personality-assisted multi-task learning for generic and personalized image aesthetics assessment.IEEE Transactions on Image Processing, 2020. 1

    2020

  34. [42]

    Photo composition with real-time rating.Sensors, 20(3), 2020

    Yi-Feng Li, Chuan-Kai Yang, and Yi-Zhen Chang. Photo composition with real-time rating.Sensors, 20(3), 2020. 2

    2020

  35. [43]

    Ppr10k: A large-scale portrait photo retouching dataset with human-region mask and group-level consistency

    Jie Liang, Hui Zeng, Miaomiao Cui, Xuansong Xie, and Lei Zhang. Ppr10k: A large-scale portrait photo retouching dataset with human-region mask and group-level consistency. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021. 2, 3, 13

    2021

  36. [44]

    Photobot: Reference-guided interactive photography via natural language

    Oliver Limoyo, Jimmy Li, Dmitriy Rivkin, Jonathan Kelly, and Gregory Dudek. Photobot: Reference-guided interactive photography via natural language. In2024 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems (IROS),

  37. [45]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. InAdvances in neural information processing systems, 2023. 2

    2023

  38. [46]

    Smarteye: assisting instant photo taking via integrating user preference with deep view proposal network

    Shuai Ma, Zijun Wei, Feng Tian, Xiangmin Fan, Jianming Zhang, Xiaohui Shen, Zhe Lin, Jin Huang, Radom ´ır Mˇech, Dimitris Samaras, et al. Smarteye: assisting instant photo taking via integrating user preference with deep view proposal network. InProceedings of the 2019 CHI con...

    2019

  39. [47]

    Parameter- efficient adaptation of large vision—language models for video memorability prediction.Sensors (Basel, Switzerland),

    Iv´an Mart´ın-Fern´andez, Sergio Esteban-Romero, Fernando Fern´andez-Mart´ınez, and Manuel Gil-Mart ´ın. Parameter- efficient adaptation of large vision—language models for video memorability prediction.Sensors (Basel, Switzerland),

  40. [48]

    Multimodal memorability: Modeling effects of semantics and decay on video memora- bility

    Anelise Newman, Camilo Fosco, Vincent Casser, Allen Lee, Barry McNamara, and Aude Oliva. Multimodal memorability: Modeling effects of semantics and decay on video memora- bility. InEuropean Conference on Computer Vision, 2020. 2, 13

    2020

  41. [49]

    Yo’llava: Your personalized language and vision assistant

    Thao Nguyen, Haotian Liu, Yuheng Li, Mu Cai, Utkarsh Ojha, and Yong Jae Lee. Yo’llava: Your personalized language and vision assistant. InNeurIPS, 2024. 18

    2024

  42. [50]

    Gpt-5 system card

    OpenAI. Gpt-5 system card. Technical report, OpenAI, 2025. 6, 7, 15, 17

    2025

  43. [51]

    The linear representation hypothesis and the geometry of large language models.ICML, 2024

    Kiho Park, Yo Joong Choe, and Victor Veitch. The linear representation hypothesis and the geometry of large language models.ICML, 2024. 2 10

    2024

  44. [52]

    Is im- age memorability prediction solved? InProceedings of the IEEE/CVF conference on computer vision and pattern recog- nition workshops, 2019

    Shay Perera, Ayellet Tal, and Lihi Zelnik-Manor. Is im- age memorability prediction solved? InProceedings of the IEEE/CVF conference on computer vision and pattern recog- nition workshops, 2019. 2, 13

    2019

  45. [53]

    The photographer’s eye: Teaching multimodal large language models to see, and critique like photographers

    Daiqing Qi, Handong Zhao, Jing Shi, Simon Jenni, Yifei Fan, Franck Dernoncourt, Scott Cohen, and Sheng Li. The photographer’s eye: Teaching multimodal large language models to see, and critique like photographers. InProceedings of the IEEE/CVF Conference on Computer Vision and...

    2025

  46. [54]

    The photographer’s eye: Teaching multimodal large language models to see, and critique like photographers

    Daiqing Qi, Handong Zhao, Jing Shi, Simon Jenni, Yifei Fan, Franck Dernoncourt, Scott Cohen, and Sheng Li. The photographer’s eye: Teaching multimodal large language models to see, and critique like photographers. InProceedings of the Computer Vision and Pattern Recognition Co...

  47. [55]

    Learn- ing 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. Learn- ing transferable visual models from natural language supervi- sion. InInternational conference on machine learning. PmLR,

  48. [56]

    Steering llama 2 via con- trastive activation addition

    Nina Rimsky, Nick Gabrieli, Julian Schulz, Meg Tong, Evan Hubinger, and Alexander Turner. Steering llama 2 via con- trastive activation addition. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2024. 2

    2024

  49. [57]

    Lunar: Llm unlearning via neural acti- vation redirection.arXiv preprint arXiv:2502.07218, 2025

    William F Shen, Xinchi Qiu, Meghdad Kurmanji, Alex Ia- cob, Lorenzo Sani, Yihong Chen, Nicola Cancedda, and Nicholas D Lane. Lunar: Llm unlearning via neural acti- vation redirection.arXiv preprint arXiv:2502.07218, 2025. 2

    2025

  50. [58]

    Long-term ad memorability: Un- derstanding and generating memorable ads

    Harini SI, Somesh Singh, Yaman K Singla, Aanisha Bhat- tacharyya, Veeky Baths, Changyou Chen, Rajiv Ratn Shah, and Balaji Krishnamurthy. Long-term ad memorability: Un- derstanding and generating memorable ads. InProceedings of the IEEE international conference on computer visi...

    2024

  51. [59]

    How to make an image more memorable? a deep style transfer approach

    Aliaksandr Siarohin, Gloria Zen, Cveta Majtanovic, Xavier Alameda-Pineda, Elisa Ricci, and Nicu Sebe. How to make an image more memorable? a deep style transfer approach. InProceedings of the 2017 ACM on international conference on multimedia retrieval, 2017. 1, 2

    2017

  52. [60]

    Changing the image memorability: From basic photo editing to gans

    Oleksii Sidorov. Changing the image memorability: From basic photo editing to gans. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2019. 2

    2019

  53. [61]

    The proof and measurement of association between two things.International Journal of Epidemiology,

    C Spearman. The proof and measurement of association between two things.International Journal of Epidemiology,

  54. [62]

    Deep learning for predicting image memorability

    Hammad Squalli-Houssaini, Ngoc QK Duong, Marquant Gwena¨elle, and Claire-H ´el`ene Demarty. Deep learning for predicting image memorability. In2018 IEEE interna- tional conference on acoustics, speech and signal processing (ICASSP), 2018. 1

    2018

  55. [63]

    Mem- net: A persistent memory network for image restoration

    Ying Tai, Jian Yang, Xiaoming Liu, and Chunyan Xu. Mem- net: A persistent memory network for image restoration. In Proceedings of the IEEE international conference on com- puter vision, 2017. 13, 14

    2017

  56. [64]

    Seeing the abstract: Translating the ab- stract language for vision language models

    Davide Talon, Federico Girella, Ziyue Liu, Marco Cristani, and Yiming Wang. Seeing the abstract: Translating the ab- stract language for vision language models. InProceedings of the Computer Vision and Pattern Recognition Conference,

  57. [65]

    Steering language models with activation engineering.arXiv preprint arXiv:2308.10248, 2023

    Alexander Matt Turner, Lisa Thiergart, Gavin Leech, David Udell, Juan J Vazquez, Ulisse Mini, and Monte MacDiarmid. Steering language models with activation engineering.arXiv preprint arXiv:2308.10248, 2023. 2, 5

    2023 arXiv

  58. [66]

    Diffusers: State-of-the-art diffusion 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 diffusion models,

  59. [67]

    Internvl3

    Weiyun Wang, Zhangwei Gao, Lixin Gu, Hengjun Pu, Long Cui, Xingguang Wei, Zhaoyang Liu, Linglin Jing, Shenglong Ye, Jie Shao, et al. Internvl3. 5: Advancing open-source multimodal models in versatility, reasoning, and efficiency. arXiv preprint arXiv:2508.18265, 2025. 3, 5, 6, 7

    2025 arXiv

  60. [68]

    Efficient guided generation for large language models.arXiv preprint arXiv:2307.09702, 2023

    Brandon T Willard and R ´emi Louf. Efficient guided generation for large language models.arXiv preprint arXiv:2307.09702, 2023. 6, 14

    2023 arXiv

  61. [69]

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chau- mond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, R ´emi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, ...

    2020

  62. [70]

    Qwen-image technical report.preprint arXiv:2508.02324, 2025

    Chenfei Wu, Jiahao Li, Jingren Zhou, Junyang Lin, Kaiyuan Gao, Kun Yan, Sheng-ming Yin, Shuai Bai, Xiao Xu, Yilei Chen, et al. Qwen-image technical report.preprint arXiv:2508.02324, 2025. 18

    2025 arXiv

  63. [71]

    Q-instruct: Improving low-level visual abilities for multi-modality foundation models

    Haoning Wu, Zicheng Zhang, Erli Zhang, Chaofeng Chen, Liang Liao, Annan Wang, Kaixin Xu, Chunyi Li, Jingwen Hou, Guangtao Zhai, et al. Q-instruct: Improving low-level visual abilities for multi-modality foundation models. In Proceedings of the IEEE/CVF conference on computer v...

    2024

  64. [72]

    Interpretable aesthetic analysis model for intel- ligent photography guidance systems

    Xiaoran Wu. Interpretable aesthetic analysis model for intel- ligent photography guidance systems. InProceedings of the 27th International Conference on Intelligent User Interfaces,

  65. [73]

    Tumera: Tutor of photography beginners.arXiv preprint arXiv:2109.11365, 2021

    Xiaoran Wu and Jia Jia. Tumera: Tutor of photography beginners.arXiv preprint arXiv:2109.11365, 2021. 2

    2021 arXiv

  66. [74]

    Don’t judge before you clip: A unified approach for perceptual tasks.arXiv preprint arXiv:2503.13260, 2025

    Amit Zalcher, Navve Wasserman, Roman Beliy, Oliver Hein- imann, and Michal Irani. Don’t judge before you clip: A unified approach for perceptual tasks.arXiv preprint arXiv:2503.13260, 2025. 2, 13, 14

    2025 arXiv

  67. [75]

    Zar.Spearman Rank Correlation

    Jerrold H. Zar.Spearman Rank Correlation. John Wiley & Sons, Ltd, 2005. 13

    2005

  68. [76]

    Sigmoid loss for language image pre-training

    Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid loss for language image pre-training. 11 InProceedings of the IEEE/CVF international conference on computer vision, 2023. 13

    2023

  69. [77]

    Sigmoid loss for language image pre-training

    Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid loss for language image pre-training. InProceedings of the IEEE/CVF international conference on computer vision, 2023. 2

    2023

  70. [78]

    Minigpt-4: Enhancing vision-language understanding with advanced large language models

    Deyao Zhu, Jun Chen, Xiaoqian Shen, Xiang Li, and Mo- hamed Elhoseiny. Minigpt-4: Enhancing vision-language understanding with advanced large language models. InThe Twelfth International Conference on Learning Representa- tions, 2024. 2

    2024

  71. [79]

    Ad- just the brightness to match the one in Image B

    Andy Zou, Long Phan, Sarah Chen, James Campbell, Phillip Guo, Richard Ren, Alexander Pan, Xuwang Yin, Mantas Mazeika, Ann-Kathrin Dombrowski, et al. Representation engineering: A top-down approach to ai transparency.arXiv preprint arXiv:2310.01405, 2023. 2 12 How to Take a Mem...

    2023 arXiv

  72. [80]

    Rotate the perspective to face forward

  73. [81]

    Bring the hands up to cover the mouth

  74. [82]

    Hold a small yellow flower between the fingers

  75. [83]

    Mem score: 0.753 Mem score: 0.803 Feedbacka:

    Adjust the hair to frame the face evenly. Mem score: 0.753 Mem score: 0.803 Feedbacka:

  76. [84]

    Adjust the position of the person on the left to face forward with a slight smile

  77. [85]

    Raise the head of the person on the right and have them look slightly to the side with a gentle smile

  78. [86]

    Ensure both individuals are standing upright and close together, with the person on the right holding the handlebars of the scooter

  79. [87]

    Mem score: 0.885 Mem score: 0.965 Feedbacka:

    Maintain the floral arrangement and attire of both individuals as they are. Mem score: 0.885 Mem score: 0.965 Feedbacka:

  80. [88]

    Adjust the position of the left arm to rest on the hip, ensuring the hand is placed firmly on the hip bone

  81. [89]

    Shift the weight slightly to the left leg to create a more dynamic stance

  82. [90]

    Tilt the head slightly to the right to add a subtle angle to the gaze

  83. [91]

    Ensure the right hand is relaxed by the side, with fingers gently curved

  84. [92]

    Mem score: 0.578 Mem score: 0.981 Feedbacka:

    Adjust the lighting to maintain consistency in shadows and highlights across the outfit. Mem score: 0.578 Mem score: 0.981 Feedbacka:

  85. [93]

    Crop the image to focus on the upper body of the person, removing the background elements such as the airplane and the chair

  86. [94]

    Adjust the angle to a closer, more intimate shot, cen- tering the person’s face and upper torso

  87. [95]

    Ensure the person’s hand is positioned near the face, with the index finger touching the lips, maintaining the same pose but in a tighter frame

  88. [96]

    Remove any elements that are not part of the person’s attire or immediate surroundings, such as the grass and distant trees

  89. [97]

    Figure 10

    Enhance the lighting to create a softer, more even illumination on the person’s face and clothing. Figure 10. A set of qualitative examples from MemBench. 19 Mem score: 0.716 Mem score: 0.980 Feedbacka:

  90. [98]

    Stand up from the seated position on the railway track

  91. [99]

    Hold a wicker basket with both hands in front of the body

  92. [100]

    Adjust the hair to fall naturally over the shoulders

  93. [101]

    Shift the gaze to the side while maintaining a smile

  94. [102]

    Mem score: 0.710 Mem score: 0.752 Feedbacka:

    Ensure the background changes to a green train car with visible rust and metal details. Mem score: 0.710 Mem score: 0.752 Feedbacka:

  95. [103]

    Adjust the person’s posture to a standing position

  96. [104]

    Raise both arms to hold flowers above the head

  97. [105]

    Ensure the person is surrounded by lush greenery

  98. [106]

    Remove the basket and gardening tools from the scene

  99. [107]

    Mem score: 0.807 Mem score: 0.892 Feedbacka:

    Adjust the lighting to create a brighter and more vi- brant atmosphere. Mem score: 0.807 Mem score: 0.892 Feedbacka:

  100. [108]

    Reposition the couple from a seated to a lying down pose on a grassy surface

  101. [109]

    Adjust the angle of the shot to an overhead view

  102. [110]

    Remove the stone wall and window background, re- placing it with a grassy area

  103. [111]

    Ensure the bride’s dress and veil spread out naturally on the grass

  104. [112]

    Place the groom’s jacket and pants neatly on the grass beside them

  105. [113]

    Mem score: 0.749 Mem score: 0.829 Feedbacka:

    Position the bride’s bouquet on the grass near her hand. Mem score: 0.749 Mem score: 0.829 Feedbacka:

  106. [114]

    Adjust the position of the person so they are standing more centrally within the frame

  107. [115]

    Shift the perspective slightly to the right to include more of the water and the sculpture on the left

  108. [116]

    Reduce the brightness and contrast to create a softer, more subdued lighting effect

  109. [117]

    Reposition the person’s arm so it is relaxed by their side, not holding onto the structure

  110. [118]

    Mem score: 0.798 Mem score: 0.990 Feedbacka:

    Ensure the reflection on the water is more prominent by adjusting the angle of the light source. Mem score: 0.798 Mem score: 0.990 Feedbacka:

  111. [119]

    Reposition the individuals so that they are facing each other, with one person lifting the other into their arms

  112. [120]

    Adjust the arms so that the lifted person’s arms are wrapped around the other person’s neck

  113. [121]

    Ensure the lifted person’s legs are bent at the knees and held by the other person

  114. [122]

    Shift the gaze of both individuals to look at each other affectionately

  115. [123]

    Figure 11

    Maintain the scenic background with mountains and clouds, but adjust the angle slightly to accommodate the new pose. Figure 11. A set of qualitative examples from MemBench. 20

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.