Reference-Driven Human Image Animation
Animating a person or character in a still image to follow the motion of a driving video or pose sequence.
Tier Grades
Tier A: Frontier AI labs
feasible
Big tech releases DiT-based animation models.
Tier B: Indonesian AI labs
stretch
Training is a stretch for an Indonesian lab.
Tier C: Institution units such as ITB
feasible
A 6-author team produced DreamActor-M1; a university unit can work on methods.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
stretch
Client value is mostly in marketing content, a narrower pilot.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeTraining compute is not in the sources read.
- DataNot in the sources read; DreamActor-M1 uses implicit face features, 3D head spheres and 3D body skeletons as guidance.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 6 authors on the DreamActor-M1 paper (arXiv 2504.01724).
- Time horizonMonths.
- Shelf-life signalsGeneral video models absorb animation control, Identity misuse rules
- Pipeline roleTier A trains animation models; Tier C studies them; Tier D uses them in content pilots.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
Links
R&D TopicReference-Driven Human Image Animation
Models
Wan2.2-Animate-2-14BAlibaba (Qwen), avatarUniAnimateAlibaba (Qwen), avatar
DreamActor-M1ByteDance, avatar
First Order Motion ModelUniversity of Trento, avatar+4 more
Wan2.2-Animate-2-14BAlibaba (Qwen), avatarUniAnimateAlibaba (Qwen), avatarDreamActor-M1ByteDance, avatarFirst Order Motion ModelUniversity of Trento, avatar+4 moreUse Casesnone recorded
Who Runs Itnone recorded
Models 8
DreamActorByteDance, avatarproduced byDreamActor-M1ByteDance, avatarexemplarDreamActor-M2.0ByteDance, avatarproduced by
FaceShotTongji University, Shanghai AI Lab, avatarproduced byFirst Order Motion ModelUniversity of Trento, Snap Inc., avatarexemplarUniAnimateAlibaba (Qwen), avatarexemplarWan-Dancer-14BAlibaba (Qwen), avatarproduced byWan2.2-Animate-2-14BAlibaba (Qwen), avatarexemplarPapers
- DisCo: Disentangled Control for Realistic Human Dance GenerationTan Wang et al., CVPR24 (per arXiv comment), 2023 founding
- Champ: Controllable and Consistent Human Image Animation with 3D Parametric GuidanceShenhao Zhu et al., arXiv preprint, 2024 state of the art
- UniAnimate: Taming Unified Video Diffusion Models for Consistent Human Image AnimationXiang Wang et al., arXiv preprint, 2024 state of the art
- Animate Anyone: Consistent and Controllable Image-to-Video Synthesis for Character AnimationLi Hu et al., arXiv preprint, 2023 state of the art
- MagicAnimate: Temporally Consistent Human Image Animation using Diffusion ModelZhongcong Xu et al., arXiv preprint, 2023 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2504.01724 https://arxiv.org/abs/2504.01724
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Reference-Driven Human Image Animation", https://aurion.semai.id/topics/reference-driven-human-image-animation/, records of October 2026.