Towards human-centered and efficient video synthesis: a survey of multimodal diffusion models

{"Abstract":[0],"Multimodal":[1],"video":[2,12,222],"diffusion":[3],"models":[4],"have":[5,43],"emerged":[6],"as":[7,114,179],"transformative":[8],"tools":[9],"for":[10,135,176,219],"controlled":[11],"synthesis,":[13],"integrating":[14],"text,":[15],"images,":[16],"audio,":[17],"and":[18,37,56,77,94,106,161,196,214],"pose":[19],"sequences":[20],"to":[21,120,205],"generate":[22],"semantically":[23],"meaningful":[24],"content.":[25],"Despite":[26],"significant":[27],"advances,":[28],"critical":[29],"gaps":[30],"persist":[31],"in":[32,59,190],"temporal":[33,115],"consistency,":[34],"multimodal":[35,78,147,221],"alignment,":[36],"human-centric":[38,88,149],"motion":[39,61,89,166],"generation.":[40,223],"Existing":[41],"surveys":[42],"not":[44],"addressed":[45],"clearly":[46],"the":[47,83,186],"complex":[48],"interplay":[49],"between":[50,103],"these":[51],"components,":[52],"particularly":[53],"physiological":[54,92],"constraints":[55,157],"identity":[57,95,162],"preservation":[58,163],"human":[60],"synthesis.":[62],"This":[63],"survey":[64],"provides":[65],"a":[66,70,180],"comprehensive":[67],"analysis":[68,99,192],"through":[69],"unified":[71],"architectural":[72],"framework,":[73],"examining":[74],"spatial-temporal":[75],"representations":[76],"conditioning":[79],"mechanisms.":[80],"We":[81,168],"present":[82],"first":[84],"systematic":[85],"evaluation":[86,212],"of":[87,201],"modeling,":[90],"addressing":[91],"plausibility":[93],"consistency":[96],"challenges.":[97],"Our":[98],"reveals":[100],"fundamental":[101],"trade-offs":[102],"computational":[104,123],"efficiency":[105],"generation":[107],"quality,":[108],"with":[109,128,145,165,173],"reported":[110],"specialized":[111],"techniques":[112],"such":[113],"block":[116],"pruning":[117],"achieving":[118],"up":[119],"$$523\\\\times":[121],"$$":[122],"savings":[124],"under":[125],"specific":[126],"baselines":[127],"minimal":[129],"quality":[130],"degradation":[131],"(see":[132],"Sect.":[133],"6.3":[134],"comparability":[136],"caveats).":[137],"Key":[138],"findings":[139],"indicate":[140],"that":[141,184],"current":[142],"approaches":[143],"struggle":[144],"seamless":[146],"integration,":[148],"applications":[150],"face":[151],"\\"uncanny":[152],"valley\\"":[153],"effects":[154],"when":[155],"physics":[156],"are":[158],"too":[159],"rigid,":[160],"conflicts":[164],"dynamics.":[167],"introduce":[169],"MIME-Vid":[170,202],"(Multi-modal":[171],"Integration":[172],"Motion":[174],"Enhancement":[175],"Video":[177],"Generation)":[178],"conceptual":[181],"reference":[182],"framework":[183],"operationalises":[185],"unifying":[187],"principles":[188],"identified":[189],"our":[191],"(reference-flexibility,":[193],"physics-perception":[194],"asymmetry,":[195],"hierarchical":[197],"disentanglement);":[198],"empirical":[199],"validation":[200],"is":[203],"deferred":[204],"follow-up":[206],"work.":[207],"Furthermore,":[208],"we":[209],"propose":[210],"novel":[211],"paradigms":[213],"identify":[215],"future":[216],"research":[217],"directions":[218],"advancing":[220]}

Authors

Institutions

Publication Details

Journal
Artificial Intelligence Review
Published
2026-09-13
DOI
https://doi.org/10.1007/s10462-026-11699-z
Primary Topic
Generative Adversarial Networks and Image Synthesis
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
article

Towards human-centered and efficient video synthesis: a survey of multimodal diffusion models

Alaa Abdullah Albaghdadi, Ahmad R. Naghsh-Nilchi
Artificial Intelligence Review
Generative Adversarial Networks and Image Synthesis
article

Towards human-centered and efficient video synthesis: a survey of multimodal diffusion models

Alaa Abdullah Albaghdadi, Ahmad R. Naghsh-Nilchi
article en

Abstract

Abstract Multimodal video diffusion models have emerged as transformative tools for controlled video synthesis, integrating text, images, audio, and pose sequences to generate semantically meaningful content. Despite significant advances, critical gaps persist in temporal consistency, multimodal alignment, and human-centric motion generation. Existing surveys have not addressed clearly the complex interplay between these components, particularly physiological constraints and identity preservation in human motion synthesis. This survey provides a comprehensive analysis through a unified architectural framework, examining spatial-temporal representations and multimodal conditioning mechanisms. We present the first systematic evaluation of human-centric motion modeling, addressing physiological plausibility and identity consistency challenges. Our analysis reveals fundamental trade-offs between computational efficiency and generation quality, with reported specialized techniques such as temporal block pruning achieving up to $$523\times $$ computational savings under specific baselines with minimal quality degradation (see Sect. 6.3 for comparability caveats). Key findings indicate that current approaches struggle with seamless multimodal integration, human-centric applications face "uncanny valley" effects when physics constraints are too rigid, and identity preservation conflicts with motion dynamics. We introduce MIME-Vid (Multi-modal Integration with Motion Enhancement for Video Generation) as a conceptual reference framework that operationalises the unifying principles identified in our analysis (reference-flexibility, physics-perception asymmetry, and hierarchical disentanglement); empirical validation of MIME-Vid is deferred to follow-up work. Furthermore, we propose novel evaluation paradigms and identify future research directions for advancing multimodal video generation.

Artificial Intelligence Review
University of Isfahan (IR)
Openalex Percentile: Top 13%
Generative Adversarial Networks and Image Synthesis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.