Pith. sign in

REVIEW 4 major objections 5 minor 67 references

Multimodal LLMs as Customized Reward Models for Text-to-Image Generation

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A hidden-state reward head with skip-connection cross-attention can score and steer text-to-image generation more like humans than CLIP- or response-based judges.

desk verdict A genuinely new reward-head design with credible evaluation results, but the inference-time scaling comparison is under-controlled and needs a lambda sweep before the SOTA claim stands. read the letter →

arxiv 2507.21391 v2 pith:CDUIQU4R submitted 2025-07-28 cs.CV cs.AIcs.CL

Shijie Zhou , Ruiyi Zhang , Huaisheng Zhu , Branislav Kveton , Yufan Zhou , Jiuxiang Gu , Jian Chen , Changyou Chen This is my paper
classification cs.CVcs.AIcs.CL
keywords text-to-imageevaluationrewardmodelsmultimodallargelanguagehidden-statescoringskip-connectioncross-attentionpreferencelearninginference-timescalinghuman-aligned
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

LLaVA-Reward is a proposed judge for text-to-image generation that scores an image-prompt pair from the internal hidden state of a multimodal large language model, rather than from the text the model would write about it. The paper argues that this makes automatic evaluation faster and more flexible than current MLLM-based judges, which need instruction-following data and either generate verbose verdicts or read the probability of a rating token. It handles paired preference data through a Bradley-Terry or General Preference Model loss and unpaired binary safety data through cross-entropy. A Skip-connection Cross-Attention (SkipCA) reward head reconnects early visual features with the final text hidden state, and per-perspective LoRA adapters let one base MLLM cover alignment, fidelity, safety, and overall ranking. The paper reports that LLaVA-Reward beats CLIP-based, VQA-based, and token-likelihood judges on MJ-Bench, TIFA 160, UnsafeDiff, and SMID, and that it improves diffusion inference-time scaling on GenEval and DrawBench.

What carries the argument

The load-bearing component is the Skip-connection Cross-Attention (SkipCA) reward head. In a decoder-only MLLM the causal mask prevents visual tokens from being influenced by later text tokens, and visual-specific information is diluted in deep transformer layers; SkipCA counters this by taking projected visual tokens $e_v$ from just after the vision projector as keys and values, the final-layer hidden state $e_h$ as the query, running cross-attention, and mapping the result to a scalar or, under the General Preference Model, a vector reward. This gives the late text representation a second, direct look at early visual features; the paper's ablation shows the module matters most for safety, where removing it drops MJ-Bench safety accuracy without ties from 55.2 to 39.7. LoRA adapters, one per evaluation perspective, allow the same pretrained MLLM to switch tasks while only adding about 8 percent trainable parameters.

What would settle it

A temperature-sweep experiment holding the FK-steering strength equal across reward models (for instance, all at lambda=10, lambda=20, and lambda=60) on GenEval with SDXL would settle the inference-time-scaling claim; if ImageReward or CLIPScore matches or beats LLaVA-Reward at equal lambda, the claimed steering superiority collapses. On the evaluation side, an audit of the Levenshtein filtering step that checks whether any remaining MJ-Bench prompts or near-duplicates still appear in the ImageReward training set could reveal whether the alignment gains are inflated by leakage.

Watch

Extended reading notes

Core claim

The central claim is that a text-to-image reward model can be made accurate, cheap, and customizable by discarding the instruction-following and text-generation machinery of MLLMs and training a reward head directly on the model's hidden embeddings. Concretely, the paper shows that the EOS-token hidden state of Phi-3.5-vision, combined with visual tokens taken right after the vision projector via SkipCA, produces scores that agree with human preference better than the compared methods. On MJ-Bench, the Phi variant reaches 66.1% alignment accuracy with ties (66.2% without), 55.2% safety accuracy (92.1% without ties), and 91.1% fidelity accuracy (91.2% without ties), with the Qwen2.5-VL variant pushing fidelity to 94.3%. On TIFA 160 it reports a Pearson correlation of 71.1, on UnsafeDiff an F1 of 87.2, and on SMID an F1 of 75.7 for the 4.2B variant (78.1 for the 8.2B variant). In diffusion inference-time scaling with FK steering, it raises the mean GenEval score from 0.521 to 0.622 on Stable Diffusion v2.1 and from 0.563 to 0.645 on SDXL, above the ImageReward and CLIPScore steering baselines.

Load-bearing premise

The central claim depends on the assumption that the benchmark comparisons are configured fairly, especially that the much larger steering strength used for LLaVA-Reward in inference-time scaling (lambda=60 versus lambda=10 for ImageReward and lambda=20 for CLIPScore) does not by itself create the reported advantage.

Editorial extensions

If this is right

  • The claimed result means automatic text-to-image judging can drop the instruction-following generation step entirely: a reward head on the EOS hidden state is more human-aligned and much faster than VQA-style judges, at 0.35 seconds per evaluation versus up to 7.01 seconds.
  • Since each perspective is a LoRA adapter on one base model, the same multimodal LLM can be re-purposed for alignment, fidelity, or safety by swapping adapters, making multi-perspective evaluation cheap to maintain.
  • The FK-steering gains on GenEval and DrawBench show the reward model can be used at sampling time to improve prompt adherence in Stable Diffusion v2.1 and SDXL without extra denoising steps.
  • Hard-negative preference pairs are a large part of the alignment gain: without the extra 70k pairs, MJ-Bench alignment accuracy drops by about 1.6 points, so data construction contributes as much as the architecture.

Reading between the lines

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

  • A testable extension the paper leaves implicit is a systematic sweep over base models and vision encoders; the two reported backbones already suggest the reward head transfers, but a controlled comparison would separate the contribution of SkipCA from the base model's visual capacity.
  • The paper connects the General Preference Model to direct preference optimization but only evaluates it for pairwise scoring, so using a GPM-trained LLaVA-Reward as the reward signal in a DPO-style training loop for a text-to-image model is a natural next experiment.
  • Because the safety variant is trained on binary UnsafeBench labels and still gains F1 on UnsafeDiff and SMID, the hidden-state approach looks promising as a cheap per-perspective safety screen for generated images, though that transfer would need validation on broader safety taxonomies.
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

4 major / 5 minor

Summary. The paper introduces LLaVA-Reward, a multimodal reward model for text-to-image generation that extracts rewards from the hidden states of a pretrained MLLM (Phi-3.5-vision) instead of generating evaluation text or token probabilities. A Skip-connection Cross-Attention (SkipCA) reward head is proposed to combine early-layer visual features with later-layer text hidden states, and LoRA adapters are trained per evaluation perspective (alignment, fidelity/artifact, safety) using Bradley-Terry, General Preference Model (GPM), or cross-entropy objectives. The model is evaluated on MJ-Bench, TIFA 160, UnsafeDiff, and SMID, and is also used as a steering reward in Feynman-Kac (FK) inference-time scaling for Stable Diffusion v2.1 and SDXL. The authors report state-of-the-art results against CLIP-based, MLLM-based, and closed-source judges, and claim improved generation quality after FK steering.

Significance. If the empirical claims hold, the paper makes a useful contribution by showing that hidden-state reward extraction with a lightweight cross-attention head can be an efficient and flexible alternative to VQA-style and token-probability MLLM judges, with multi-perspective support through LoRA adapters. The paper also provides a fairly extensive ablation study (SkipCA vs MLP head, GPM vs BT, negative sample augmentation, hidden-state layer choice, base model size, LoRA vs full fine-tuning) and an inference-time comparison, which helps the community understand design trade-offs. However, the central downstream claim about inference-time scaling currently rests on an uncontrolled comparison, and the safety evaluation with ties is near chance, so the practical significance is not yet firmly established.

major comments (4)
  1. [Sec. 4.5 / Tab. 4 / Tab. 9] The inference-time scaling comparison is not controlled. The paper sets different steering temperatures for each reward model (lambda=60 for LLaVA-Reward, 10 for ImageReward, 20 for CLIPScore) without any sensitivity analysis or calibration procedure, even though these models have different output scales and score distributions. Since lambda controls how aggressively the reward is turned into resampling weights in FK steering, this choice alone could explain the observed differences. Moreover, the results are reported from a single run with no seed variance, confidence intervals, or significance tests; several margins in Tab. 9 are small (e.g., GenEval SDXL HPSv2 0.323 vs 0.316; VQAScore 0.721 vs 0.718) and could easily be within Monte Carlo noise. The abstract's claim that LLaVA-Reward 'outperforms ... in inference-time scaling' therefore needs a controlled comparison before it can be accepted.
  2. [Tab. 1 / Sec. 3] The proposed SkipCA module does not improve text-image alignment on MJ-Bench. LLaVA-Reward-Phi without SkipCA achieves 68.2% accuracy with ties and 68.6% without ties, whereas with SkipCA it achieves 66.1% and 66.2%, respectively. This is a direct negative result for the paper's central architectural motivation, which states that SkipCA 'enhances text-image correlation reasoning.' While SkipCA helps on the safety and artifact perspectives, the paper does not acknowledge or discuss that it hurts alignment; this omission is significant because alignment is one of the four claimed evaluation perspectives and the main cross-modal capability the module is designed to improve.
  3. [Tab. 1 (Safety rows)] The safety evaluation with ties is near chance. LLaVA-Reward-Phi reports 55.2% accuracy with ties on the MJ-Bench safety set. Combined with the without-ties accuracy of 92.1%, this implies that the model ties roughly 88% of the pairs, so it is barely better than random when ties are included. A safety judge that ties almost all pairs has limited practical value for ranking, and the paper does not report tie rates or discuss this behavior. The claim of 'state-of-the-art' safety evaluation is therefore overstated and should be qualified with tied-pair statistics.
  4. [Sec. 4.3 / Tab. 1] The cleanliness of the ImageReward-to-MJ-Bench training/test split rests entirely on the Levenshtein-distance filter, but no threshold or post-filter verification is reported. Since ImageReward is both the training source for LLaVA-Reward and a baseline in the same table, any residual near-duplicate prompts could inflate the reported advantage over ImageReward. The paper should specify the distance threshold, the number of removed prompts, and ideally a manual or embedding-based check for overlap, to rule out data leakage.
minor comments (5)
  1. [Tab. 1 / Sec. 4.2] The table header 'LMM-based' should be 'MLLM-based' for consistency, and 'links' in Sec. 4.2 should be 'ties' (the MJ-Bench metric uses ties, not links).
  2. [Tab. 3 caption] Table 3 is titled 'Evaluation of recent LMM-based safety judges on TIFA 160' but the columns report UnsafeDiff and SMID; the caption does not match the content.
  3. [Eq. (2)] The GPM reward difference in Eq. (2) is typeset as 'R≻rθp(ic,t), rθp(ir,t)' without explicit angle brackets around the inner product; this should be written as ⟨R≻ rθp(ic,t), rθp(ir,t)⟩ to be mathematically unambiguous.
  4. [Sec. 3, 'Skip-connection Cross Attention Modal'] The phrase 'Cross Attention Modal' should be 'Cross-Attention Module'.
  5. [Sec. 4.5, last paragraph] The word 'setted' should be 'set', and the sentence beginning 'We can observe that each diffusion inference-time scaling variant tends to have a better result when evaluated on itself' should be expanded to clarify which of the reported metrics are self-referential and which are external.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: training and evaluation are separated, and the inference-time-scaling claim is supported by external metrics; the self-score artifact is acknowledged and not load-bearing.

full rationale

LLaVA-Reward's derivation chain is empirical rather than definitional. The model is trained on ImageReward human-preference data and UnsafeBench (Tab. 5), while the headline evaluation benchmarks MJ-Bench, TIFA 160, UnsafeDiff, and SMID are separate; the authors even describe Levenshtein filtering of prompts to erase overlap with the MJ-Bench alignment set (Sec. 4.3). The safety and alignment comparisons therefore compare the model's learned scores with external human labels, not with its own training objective. The inference-time-scaling experiments use FK steering and are scored on external metrics (HPSv2, CLIPScore, VQAScore, ImageReward, and a GPT-4o LLM grader) in Tab. 9; the paper explicitly notes that each variant tends to score best on its own metric ('each diffusion inference-time scaling variant tends to have a better result when evaluated on itself'), so the LLaVA-Reward-self-score column is an acknowledged artifact rather than the load-bearing evidence. The self-citations present (e.g., LoCAL [9] for layer choice) are not load-bearing: the relevant architectural decisions are supported by in-paper ablations (Fig. 3, Tabs. 6-7). The appended Limitations section candidly flags reward-hacking risk from limited training data, which is a robustness limitation, not a circularity. The uncontrolled per-model steering temperature (lambda=60/10/20) and the lack of variance estimates are experimental-confound concerns; without evidence that lambdas were fitted to final benchmarks, they do not amount to a derivation reducing to its inputs and therefore do not raise the circularity score.

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

The central results rest on the reliability of the human preference datasets, the transfer power of the chosen pretrained MLLMs, and the fairness of evaluation choices. No new physical or theoretical entities are introduced.

free parameters (5)
  • FK steering temperature lambda for LLaVA-Reward = 60
    Set separately per reward model in Sec. 4.5; no sensitivity analysis provided, so it could bias the comparison.
  • FK steering temperature lambda for ImageReward = 10
    Same as above; different value than for LLaVA-Reward.
  • FK steering temperature lambda for CLIPScore = 20
    Same as above; different value than for LLaVA-Reward.
  • GPM embedding dimension m = 2
    Design choice for the General Preference Model objective in Eq. (2); not swept.
  • LoRA rank = 128
    Hyperparameter for the LoRA adapters; not swept.
assumptions (4)
  • domain assumption Human preference labels in ImageReward and UnsafeBench are reliable ground truth.
    The model is trained to match these labels, and their quality or bias is not audited in the paper.
  • domain assumption Pretrained Phi-3.5-vision and Qwen2.5-VL provide sufficient visual-linguistic understanding to generalize reward judgments beyond training labels.
    Transfer performance depends on the base MLLM's pretraining; the paper does not isolate this factor.
  • standard math The Bradley-Terry model applies to image preference pairs.
    Used as the ranking objective; assumes transitive, stochastic preferences.
  • domain assumption Levenshtein-distance filtering removes all harmful overlap between ImageReward training data and MJ-Bench evaluation prompts.
    The paper filters 2k samples but does not verify residual leakage empirically.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multimodal LLMs as Customized Reward Models for Text-to-Image Generation." pith.science (2026). https://pith.science/paper/CDUIQU4R

@misc{pith2026250721391,
  author       = {Pith},
  title        = {Pith review of: Multimodal LLMs as Customized Reward Models for Text-to-Image Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CDUIQU4R}},
  note         = {Machine review of arXiv:2507.21391}
}
read the original abstract

We introduce LLaVA-Reward, an efficient reward model designed to automatically evaluate text-to-image (T2I) generations across multiple perspectives, leveraging pretrained multimodal large language models (MLLMs). Existing MLLM-based approaches require instruction-following data for supervised fine-tuning and evaluate generation quality on analyzing text response, which is time-consuming and difficult to train. To address this problem, we propose LLaVA-Reward, which directly utilizes the hidden states of MLLMs given text-image pairs. To enhance the bidirectional interaction between visual and textual representations in decoder-only MLLMs, we further propose adding a Skip-connection Cross Attention (SkipCA) module. This design enhances text-image correlation reasoning by connecting early-layer visual features with later-layer hidden representations. In addition, LLaVA-Reward supports different types of preference data for efficient fine-tuning, including paired preference data and unpaired data. We train LLaVA-Reward on four evaluation perspectives: text-image alignment, fidelity/artifact, safety, and overall ranking. Empirical results demonstrate that LLaVA-Reward outperforms conventional and MLLM-based methods in generating human-aligned scores for automatic evaluations and inference-time scaling in text-to-image generations.

Figures

Figures reproduced from arXiv: 2507.21391 by the authors.

Figure 1
Figure 1. Model overview of LLaVA-Reward. LLaVA-Reward is fine-tuned on the lightweight Phi-3.5-vision using the pairwise preference [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Examples of diffusion inference-time scaling via FK steering (SDXL) using 4 different reward models with the prompt from [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Ablation Results. Left: F1 score on safety benchmark, [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Examples of diffusion inference-time scaling via FK steering (SDXL) using 5 different reward models with the prompt from [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Examples of diffusion inference-time scaling via FK steering (SDXL) using 5 different reward models with the prompt from [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Examples of diffusion inference-time scaling via FK steering (SDXL) using 5 different reward models with the prompt from [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

67 extracted references · 28 canonical work pages

  1. [1]

    Phi-3 technical report: A highly capable language model locally on your phone

    Marah Abdin, Jyoti Aneja, Hany Awadalla, Ahmed Awadal- lah, Ammar Ahmad Awan, Nguyen Bach, Amit Bahree, Arash Bakhtiari, Jianmin Bao, Harkirat Behl, et al. Phi-3 technical report: A highly capable language model locally on your phone. arXiv preprint arXiv:2404.14219, 2024. 3

    arXiv 2024

  2. [2]

    Qwen-vl: A versatile vision-language model for un- derstanding, localization, text reading, and beyond

    Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. Qwen-vl: A versatile vision-language model for un- derstanding, localization, text reading, and beyond. arXiv preprint arXiv:2308.12966, 1(2):3, 2023. 4

    arXiv 2023

  3. [3]

    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. 8

    arXiv 2025

  4. [4]

    A note on the inception score

    Shane Barratt and Rishi Sharma. A note on the inception score. arXiv preprint arXiv:1801.01973, 2018. 2

    arXiv 2018

  5. [5]

    Attend first, consolidate later: On the importance of attention in different llm layers

    Amit Ben-Artzy and Roy Schwartz. Attend first, consolidate later: On the importance of attention in different llm layers. arXiv preprint arXiv:2409.03621, 2024. 8

    arXiv 2024

  6. [6]

    Training diffusion models with reinforce- ment learning

    Kevin Black, Michael Janner, Yilun Du, Ilya Kostrikov, and Sergey Levine. Training diffusion models with reinforce- ment learning. arXiv preprint arXiv:2305.13301, 2023. 1

    arXiv 2023

  7. [7]

    Rank analysis of incomplete block designs: I

    Ralph Allan Bradley and Milton E Terry. Rank analysis of incomplete block designs: I. the method of paired compar- isons. Biometrika, 39(3/4):324–345, 1952. 2

    work page 1952

  8. [8]

    Minigpt-v2: large language model as a unified interface for vision-language multi-task learning

    Jun Chen, Deyao Zhu, Xiaoqian Shen, Xiang Li, Zechun Liu, Pengchuan Zhang, Raghuraman Krishnamoorthi, Vikas Chandra, Yunyang Xiong, and Mohamed Elhoseiny. Minigpt-v2: large language model as a unified interface for vision-language multi-task learning. arXiv preprint arXiv:2310.09478, 2023. 4

    arXiv

Show all 67 references
  1. [9]

    Lora-contextualizing adaptation of large mul- timodal models for long document understanding

    Jian Chen, Ruiyi Zhang, Yufan Zhou, Tong Yu, Franck Der- noncourt, Jiuxiang Gu, Ryan A Rossi, Changyou Chen, and Tong Sun. Lora-contextualizing adaptation of large mul- timodal models for long document understanding. arXiv preprint arXiv:2411.01106, 2024. 8

    2024 arXiv

  2. [10]

    Mj-bench: Is your multimodal reward model really a good judge for text-to- image generation? arXiv preprint arXiv:2407.04842, 2024

    Zhaorun Chen, Yichao Du, Zichen Wen, Yiyang Zhou, Chenhang Cui, Zhenzhen Weng, Haoqin Tu, Chaoqi Wang, Zhengwei Tong, Qinglan Huang, et al. Mj-bench: Is your multimodal reward model really a good judge for text-to- image generation? arXiv preprint arXiv:2407.04842, 2024. 4, 5

    2024 arXiv

  3. [11]

    How far are we to gpt-4v? closing the gap to commercial multimodal models with open-source suites

    Zhe Chen, Weiyun Wang, Hao Tian, Shenglong Ye, Zhang- wei Gao, Erfei Cui, Wenwen Tong, Kongzhi Hu, Jiapeng Luo, Zheng Ma, et al. How far are we to gpt-4v? closing the gap to commercial multimodal models with open-source suites. Science China Information Sciences, 67(12):220101,

  4. [12]

    The socio-moral image database (smid),

    Damien Crone. The socio-moral image database (smid),

  5. [13]

    Instructblip: To- wards general-purpose vision-language models with instruc- tion tuning

    Wenliang Dai, Junnan Li, D Li, AMH Tiong, J Zhao, W Wang, B Li, P Fung, and S Hoi. Instructblip: To- wards general-purpose vision-language models with instruc- tion tuning. arxiv 2023. arXiv preprint arXiv:2305.06500, 2,

    2023 arXiv

  6. [14]

    Ties matter: Meta-evaluating modern metrics with pairwise accu- racy and tie calibration

    Daniel Deutsch, George Foster, and Markus Freitag. Ties matter: Meta-evaluating modern metrics with pairwise accu- racy and tie calibration. arXiv preprint arXiv:2305.14324 ,

    arXiv

  7. [15]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. Advances in neural informa- tion processing systems, 34:8780–8794, 2021. 1

    2021

  8. [16]

    Cogview: Mastering text-to-image generation via transformers

    Ming Ding, Zhuoyi Yang, Wenyi Hong, Wendi Zheng, Chang Zhou, Da Yin, Junyang Lin, Xu Zou, Zhou Shao, Hongxia Yang, et al. Cogview: Mastering text-to-image generation via transformers. Advances in neural information processing systems, 34:19822–19835, 2021. 1

    2021

  9. [17]

    Dpok: Reinforcement learning for fine-tuning text-to-image diffu- sion models

    Ying Fan, Olivia Watkins, Yuqing Du, Hao Liu, Moonkyung Ryu, Craig Boutilier, Pieter Abbeel, Moham- mad Ghavamzadeh, Kangwook Lee, and Kimin Lee. Dpok: Reinforcement learning for fine-tuning text-to-image diffu- sion models. Advances in Neural Information Processing Systems, 3...

    2023

  10. [18]

    Geneval: An object-focused framework for evaluating text- to-image alignment

    Dhruba Ghosh, Hannaneh Hajishirzi, and Ludwig Schmidt. Geneval: An object-focused framework for evaluating text- to-image alignment. Advances in Neural Information Pro- cessing Systems, 36:52132–52152, 2023. 2, 5, 7

    2023

  11. [19]

    Generative adversarial networks

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks. Commu- nications of the ACM, 63(11):139–144, 2020. 1

    2020

  12. [20]

    Llavaguard: Vlm-based safeguards for vision dataset curation and safety assessment

    Lukas Helff, Felix Friedrich, Manuel Brack, Kristian Ker- sting, and Patrick Schramowski. Llavaguard: Vlm-based safeguards for vision dataset curation and safety assessment. arXiv preprint arXiv:2406.05113, 2024. 1, 2, 5

    2024 arXiv

  13. [21]

    Clipscore: A reference-free evaluation met- ric for image captioning

    Jack Hessel, Ari Holtzman, Maxwell Forbes, Ronan Le Bras, and Yejin Choi. Clipscore: A reference-free evaluation met- ric for image captioning. arXiv preprint arXiv:2104.08718,

    arXiv

  14. [22]

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

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium. Advances in neural information processing systems , 30, 2017. 2

    2017

  15. [23]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. 1

    2020

  16. [24]

    Lora: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021. 2

    2021 arXiv

  17. [25]

    Llama guard: Llm- based input-output safeguard for human-ai conversations

    Hakan Inan, Kartikeya Upasani, Jianfeng Chi, Rashi Rungta, Krithika Iyer, Yuning Mao, Michael Tontchev, Qing Hu, Brian Fuller, Davide Testuggine, et al. Llama guard: Llm- based input-output safeguard for human-ai conversations. arXiv preprint arXiv:2312.06674, 2023. 2

    2023 arXiv

  18. [26]

    Pick-a-pic: An open dataset of user preferences for text-to-image generation

    Yuval Kirstain, Adam Polyak, Uriel Singer, Shahbuland Ma- tiana, Joe Penna, and Omer Levy. Pick-a-pic: An open dataset of user preferences for text-to-image generation. Advances in Neural Information Processing Systems , 36: 36652–36663, 2023. 1, 2, 4

    2023

  19. [27]

    Viescore: Towards explainable metrics for conditional image synthesis evaluation

    Max Ku, Dongfu Jiang, Cong Wei, Xiang Yue, and Wenhu Chen. Viescore: Towards explainable metrics for conditional image synthesis evaluation. arXiv preprint arXiv:2312.14867, 2023. 2, 3

    2023 arXiv

  20. [28]

    What matters when building vision-language models? arXiv preprint arXiv:2405.02246, 2024

    Hugo Laurenc ¸on, L´eo Tronchon, Matthieu Cord, and Victor Sanh. What matters when building vision-language models? arXiv preprint arXiv:2405.02246, 2024. 4

    2024 arXiv

  21. [29]

    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. In In- ternational conference on machine learning , pages 19730– 19742. PMLR, 2023. 4

    2023

  22. [30]

    T2isafety: Benchmark for assessing fairness, toxicity, and privacy in image genera- tion

    Lijun Li, Zhelun Shi, Xuhao Hu, Bowen Dong, Yiran Qin, Xihui Liu, Lu Sheng, and Jing Shao. T2isafety: Benchmark for assessing fairness, toxicity, and privacy in image genera- tion. arXiv preprint arXiv:2501.12612, 2025. 1, 2, 3, 5

    2025 arXiv

  23. [31]

    Remov- ing distributional discrepancies in captions improves image- text alignment

    Yuheng Li, Haotian Liu, Mu Cai, Yijun Li, Eli Shechtman, Zhe Lin, Yong Jae Lee, and Krishna Kumar Singh. Remov- ing distributional discrepancies in captions improves image- text alignment. In European Conference on Computer Vi- sion, pages 405–422. Springer, 2024. 1, 2, 5

    2024

  24. [32]

    Evaluating text-to-visual generation with image-to-text gen- eration

    Zhiqiu Lin, Deepak Pathak, Baiqi Li, Jiayao Li, Xide Xia, Graham Neubig, Pengchuan Zhang, and Deva Ramanan. Evaluating text-to-visual generation with image-to-text gen- eration. In European Conference on Computer Vision, pages 366–384. Springer, 2024. 1, 2, 3, 5

    2024

  25. [33]

    Skywork-reward: Bag of tricks for reward modeling in llms

    Chris Yuhao Liu, Liang Zeng, Jiacai Liu, Rui Yan, Jujie He, Chaojie Wang, Shuicheng Yan, Yang Liu, and Yahui Zhou. Skywork-reward: Bag of tricks for reward modeling in llms. arXiv preprint arXiv:2410.18451, 2024. 1

    2024 arXiv

  26. [34]

    Improved baselines with visual instruction tuning, 2023

    Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. Improved baselines with visual instruction tuning, 2023. 4

    2023

  27. [35]

    Visual instruction tuning, 2023

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning, 2023. 4

    2023

  28. [36]

    Llava-next: Im- proved reasoning, ocr, and world knowledge, 2024

    Haotian Liu, Chunyuan Li, Yuheng Li, Bo Li, Yuanhan Zhang, Sheng Shen, and Yong Jae Lee. Llava-next: Im- proved reasoning, ocr, and world knowledge, 2024. 8

    2024

  29. [37]

    Llava-next: Im- proved reasoning, ocr, and world knowledge, 2024

    Haotian Liu, Chunyuan Li, Yuheng Li, Bo Li, Yuanhan Zhang, Sheng Shen, and Yong Jae Lee. Llava-next: Im- proved reasoning, ocr, and world knowledge, 2024. 4

    2024

  30. [38]

    Inference-time scaling for diffu- sion models beyond scaling denoising steps

    Nanye Ma, Shangyuan Tong, Haolin Jia, Hexiang Hu, Yu- Chuan Su, Mingda Zhang, Xuan Yang, Yandong Li, Tommi Jaakkola, Xuhui Jia, et al. Inference-time scaling for diffu- sion models beyond scaling denoising steps. arXiv preprint arXiv:2501.09732, 2025. 2, 5, 7, 3, 4

    2025 arXiv

  31. [39]

    Simpo: Sim- ple preference optimization with a reference-free reward

    Yu Meng, Mengzhou Xia, and Danqi Chen. Simpo: Sim- ple preference optimization with a reference-free reward. Advances in Neural Information Processing Systems , 37: 124198–124235, 2024. 1

    2024

  32. [40]

    Training language models to follow instructions with human feedback

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Car- roll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback. Ad- vances in neural information processing systems, 35:...

    2022

  33. [41]

    Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis. arXiv preprint arXiv:2307.01952, 2023. 7

    2023 arXiv

  34. [42]

    Unsafe diffusion: On the generation of unsafe images and hateful memes from text-to-image models

    Yiting Qu, Xinyue Shen, Xinlei He, Michael Backes, Sav- vas Zannettou, and Yang Zhang. Unsafe diffusion: On the generation of unsafe images and hateful memes from text-to-image models. In Proceedings of the 2023 ACM SIGSAC Conference on Computer and Communications Se- curity, ...

    2023

  35. [43]

    Unsafebench: Benchmarking image safety classifiers on real-world and ai-generated im- ages

    Yiting Qu, Xinyue Shen, Yixin Wu, Michael Backes, Savvas Zannettou, and Yang Zhang. Unsafebench: Benchmarking image safety classifiers on real-world and ai-generated im- ages. arXiv preprint arXiv:2405.03486, 2024. 2, 4, 6

    2024 arXiv

  36. [44]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. In International conference on machine learning, ...

    2021

  37. [45]

    Direct preference optimization: Your language model is secretly a reward model

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christo- pher D Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. Advances in neural information processing systems, 36:53728–53741, 2023. 4

    2023

  38. [46]

    Direct preference optimization: Your language model is secretly a reward model

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christo- pher D Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. Advances in Neural Information Processing Systems, 36, 2024. 1

    2024

  39. [47]

    Zero-shot text-to-image generation

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea V oss, Alec Radford, Mark Chen, and Ilya Sutskever. Zero-shot text-to-image generation. In International confer- ence on machine learning, pages 8821–8831. Pmlr, 2021. 1

    2021

  40. [48]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 1, 7

    2022

  41. [49]

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

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

    2022

  42. [50]

    Laion-5b: An open large-scale dataset for training next generation image-text models

    Christoph Schuhmann, Romain Beaumont, Richard Vencu, Cade Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Aarush Katta, Clayton Mullis, Mitchell Worts- man, et al. Laion-5b: An open large-scale dataset for training next generation image-text models. Advances in Neural In- f...

    2022

  43. [51]

    Proximal policy optimization algo- rithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Rad- ford, and Oleg Klimov. Proximal policy optimization algo- rithms. arXiv preprint arXiv:1707.06347, 2017. 1

    2017 arXiv

  44. [52]

    A general framework for inference-time scaling and steering of diffusion models

    Raghav Singhal, Zachary Horvitz, Ryan Teehan, Mengye Ren, Zhou Yu, Kathleen McKeown, and Rajesh Ranganath. A general framework for inference-time scaling and steering of diffusion models. arXiv preprint arXiv:2501.06848, 2025. 2, 4, 6, 7, 1

    2025 arXiv

  45. [53]

    Joar Skalse, Nikolaus H. R. Howe, Dmitrii Krasheninnikov, and David Krueger. Defining and characterizing reward hacking, 2025. 1

    2025

  46. [54]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International confer- ence on machine learning, pages 2256–2265. PMLR, 2015. 1

    2015

  47. [55]

    Rethink- ing bradley-terry models in preference-based reward mod- eling: Foundations, theory, and alternatives

    Hao Sun, Yunyi Shen, and Jean-Francois Ton. Rethink- ing bradley-terry models in preference-based reward mod- eling: Foundations, theory, and alternatives. arXiv preprint arXiv:2411.04991, 2024. 4

    2024 arXiv

  48. [56]

    Evalalign: Supervised fine- tuning multimodal llms with human-aligned data for evaluat- ing text-to-image models

    Zhiyu Tan, Xiaomeng Yang, Luozheng Qin, Mengping Yang, Cheng Zhang, and Hao Li. Evalalign: Supervised fine- tuning multimodal llms with human-aligned data for evaluat- ing text-to-image models. arXiv preprint arXiv:2406.16562,

    arXiv

  49. [57]

    Automatic evaluation for text- to-image generation: Task-decomposed framework, distilled training, and meta-evaluation benchmark

    Rong-Cheng Tu, Zi-Ao Ma, Tian Lan, Yuehao Zhao, Heyan Huang, and Xian-Ling Mao. Automatic evaluation for text- to-image generation: Task-decomposed framework, distilled training, and meta-evaluation benchmark. arXiv preprint arXiv:2411.15488, 2024. 2

    2024 arXiv

  50. [58]

    Helpsteer2-preference: Complementing ratings with preferences

    Zhilin Wang, Alexander Bukharin, Olivier Delalleau, Daniel Egert, Gerald Shen, Jiaqi Zeng, Oleksii Kuchaiev, and Yi Dong. Helpsteer2-preference: Complementing ratings with preferences. arXiv preprint arXiv:2410.01257, 2024. 1, 3

    2024 arXiv

  51. [59]

    Mllm-as-a-judge for image safety without human labeling

    Zhenting Wang, Shuming Hu, Shiyu Zhao, Xiaowen Lin, Felix Juefei-Xu, Zhuowei Li, Ligong Han, Harihar Subra- manyam, Li Chen, Jianfa Chen, et al. Mllm-as-a-judge for image safety without human labeling. arXiv preprint arXiv:2501.00192, 2024. 2

    2024 arXiv

  52. [60]

    Q-align: Teaching lmms for visual scoring via discrete text-defined levels

    Haoning Wu, Zicheng Zhang, Weixia Zhang, Chaofeng Chen, Liang Liao, Chunyi Li, Yixuan Gao, Annan Wang, Erli Zhang, Wenxiu Sun, et al. Q-align: Teaching lmms for visual scoring via discrete text-defined levels. arXiv preprint arXiv:2312.17090, 2023. 1, 2

    2023 arXiv

  53. [61]

    Human preference score: Better aligning text- to-image models with human preference

    Xiaoshi Wu, Keqiang Sun, Feng Zhu, Rui Zhao, and Hong- sheng Li. Human preference score: Better aligning text- to-image models with human preference. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 2096–2105, 2023. 1, 2, 4

    2023

  54. [62]

    Imagere- ward: Learning and evaluating human preferences for text- to-image generation

    Jiazheng Xu, Xiao Liu, Yuchen Wu, Yuxuan Tong, Qinkai Li, Ming Ding, Jie Tang, and Yuxiao Dong. Imagere- ward: Learning and evaluating human preferences for text- to-image generation. Advances in Neural Information Pro- cessing Systems, 36, 2024. 1, 2, 4, 6

    2024

  55. [63]

    A normalized levenshtein distance metric

    Li Yujian and Liu Bo. A normalized levenshtein distance metric. IEEE transactions on pattern analysis and machine intelligence, 29(6):1091–1095, 2007. 6

    2007

  56. [64]

    When and why vision- language models behave like bags-of-words, and what to do about it? In The Eleventh International Conference on Learning Representations, 2023

    Mert Yuksekgonul, Federico Bianchi, Pratyusha Kalluri, Dan Jurafsky, and James Zou. When and why vision- language models behave like bags-of-words, and what to do about it? In The Eleventh International Conference on Learning Representations, 2023. 1, 2

    2023

  57. [65]

    General preference modeling with preference rep- resentations for aligning language models

    Yifan Zhang, Ge Zhang, Yue Wu, Kangping Xu, and Quan- quan Gu. General preference modeling with preference rep- resentations for aligning language models. arXiv preprint arXiv:2410.02197, 2024. 1, 3, 4

    2024 arXiv

  58. [66]

    Treat visual tokens as text? but your mllm only needs fewer efforts to see

    Zeliang Zhang, Phu Pham, Wentian Zhao, Kun Wan, Yu-Jhe Li, Jianing Zhou, Daniel Miranda, Ajinkya Kale, and Chen- liang Xu. Treat visual tokens as text? but your mllm only needs fewer efforts to see. arXiv preprint arXiv:2410.06169,

    arXiv

  59. [67]

    Towards language-free training for text-to-image generation

    Yufan Zhou, Ruiyi Zhang, Changyou Chen, Chunyuan Li, Chris Tensmeyer, Tong Yu, Jiuxiang Gu, Jinhui Xu, and Tong Sun. Towards language-free training for text-to-image generation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 17907–...

    2022

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.