{"id":"https://openalex.org/W7168264862","doi":"https://doi.org/10.48550/arxiv.2607.11886","title":"Read It Back: Pretrained MLLMs Are Zero-Shot Reward Models for Text-to-Image Generation","display_name":"Read It Back: Pretrained MLLMs Are Zero-Shot Reward Models for Text-to-Image Generation","publication_year":2026,"publication_date":"2026-07-13","ids":{"openalex":"https://openalex.org/W7168264862","doi":"https://doi.org/10.48550/arxiv.2607.11886"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2607.11886","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.11886","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2607.11886","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5140687029","display_name":"Runhui Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Runhui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140579193","display_name":"Qihui Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Qihui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140606694","display_name":"Zhe Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Zhe","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140579225","display_name":"Yu Gao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gao, Yu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140735962","display_name":"Jie Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Jie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5140648315","display_name":"Hengshuang Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Hengshuang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.35749998688697815,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.35749998688697815,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.2572999894618988,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.11550000309944153,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.5372999906539917},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.5320000052452087},{"id":"https://openalex.org/keywords/preference","display_name":"Preference","score":0.49380001425743103},{"id":"https://openalex.org/keywords/function","display_name":"Function (biology)","score":0.4778999984264374},{"id":"https://openalex.org/keywords/reinforcement","display_name":"Reinforcement","score":0.40790000557899475}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7002999782562256},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5866000056266785},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.5372999906539917},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.5320000052452087},{"id":"https://openalex.org/C2781249084","wikidata":"https://www.wikidata.org/wiki/Q908656","display_name":"Preference","level":2,"score":0.49380001425743103},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.4778999984264374},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.41119998693466187},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.40790000557899475},{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.35899999737739563},{"id":"https://openalex.org/C115961682","wikidata":"https://www.wikidata.org/wiki/Q860623","display_name":"Image (mathematics)","level":2,"score":0.3497999906539917},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.2549999952316284}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2607.11886","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.11886","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2607.11886","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.11886","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"In":[0],"this":[1],"paper,":[2],"we":[3],"propose":[4],"SpectraReward,":[5],"a":[6,29,52,85,108,124,201],"training-free":[7],"reward":[8,16,101,114,137,171,179,194],"function":[9],"that":[10,156,177,197],"turns":[11],"pretrained":[12,72],"MLLMs":[13,180],"into":[14],"off-the-shelf":[15],"models":[17,91,115],"for":[18,88,103,204],"image-generation":[19,126,206],"reinforcement":[20],"learning.":[21],"Instead":[22],"of":[23],"asking":[24],"the":[25,41,48,60,66,70,93,100],"MLLM":[26,138,142],"to":[27,146],"judge":[28],"generated":[30,49],"image":[31,50],"or":[32,116,189],"answer":[33],"decomposed":[34],"verification":[35],"questions,":[36],"SpectraReward":[37,122,158],"measures":[38],"how":[39],"well":[40],"original":[42],"prompt":[43,63],"can":[44,187],"be":[45],"recovered":[46],"from":[47,140],"through":[51,123],"single":[53],"image-conditioned,":[54],"teacher-forced":[55],"forward":[56],"pass.":[57],"We":[58,81],"use":[59],"average":[61],"image-conditioned":[62],"log-likelihood":[64],"as":[65,99],"reward,":[67],"directly":[68],"reusing":[69],"MLLM's":[71],"image-text":[73],"alignment":[74,199],"ability":[75],"without":[76,112],"preference":[77],"labels,":[78],"reward-model":[79],"fine-tuning.":[80],"further":[82],"introduce":[83],"Self-SpectraReward,":[84],"special":[86],"case":[87],"unified":[89],"multimodal":[90],"where":[92],"policy's":[94],"own":[95],"understanding":[96],"branch":[97],"serves":[98],"model":[102],"its":[104],"generation":[105,165],"branch,":[106],"forming":[107],"closed-loop":[109],"self-improving":[110],"framework":[111],"external":[113,117,193],"knowledge.":[118],"Extensive":[119],"experiments":[120],"validate":[121],"broad":[125],"RL":[127,134],"study":[128],"covering":[129],"two":[130],"diffusion":[131],"models,":[132,195],"three":[133],"algorithms,":[135],"nine":[136],"backbones":[139],"four":[141],"families":[143],"spanning":[144],"4B":[145],"235B":[147],"parameters,":[148],"and":[149,159,162,167],"five":[150],"out-of-distribution":[151],"text-to-image":[152],"benchmarks.":[153],"Results":[154],"show":[155],"both":[157],"Self-SpectraReward":[160,186],"significantly":[161],"consistently":[163],"improve":[164],"performance":[166],"outperform":[168],"prior":[169],"MLLM-derived":[170],"training":[172],"methods.":[173],"Further":[174],"analysis":[175],"reveals":[176],"larger":[178,192],"are":[181],"not":[182],"always":[183],"better,":[184],"while":[185],"match":[188],"surpass":[190],"much":[191],"suggesting":[196],"reward-policy":[198],"is":[200],"key":[202],"factor":[203],"effective":[205],"RL.":[207],"Project":[208],"Page:":[209],"https://huangrh99.github.io/SpectraReward/":[210]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-07-15T00:00:00"}