Animate Anyone 2

One-line summary: 2025 successor to AnimateAnyone from Alibaba's HumanAIGC team that adds environment-aware character animation.

What it is

A pose-driven character image animation model that — unlike its V1 predecessor — additionally extracts environmental context from the driving video so the animated character interacts plausibly with surroundings. Project name: "High-Fidelity Character Image Animation with Environment Affordance."

Why it matters to ai-video-generation

The V1 (AnimateAnyone) model from late 2023 was an early pose-driven animation reference; the V2 paper (arXiv:2502.06145, 2025) extends the lineage by incorporating scene/environment information rather than treating the background as inert. From 2026-05-07-ai-avatar-motion-mimicking-models-survey.

Key facts

• Authors: HumanAIGC team at Alibaba.
• Project page: humanaigc.github.io/animate-anyone-2/.
• Inputs: reference character image + driving video (the latter supplies both motion signals AND environmental representation).
• Stated baselines: outperforms MIMO and Viggle V3 on robustness and detail preservation.

Technical contributions over V1

• Environment affordance: the environment is formulated as "regions devoid of characters"; generated character animations maintain coherence and plausible interaction with that environment.
• Object guider: extracts features of interacting objects with spatial blending for feature injection.
• Pose modulation strategy: handles more diverse motion patterns than V1.
• End-to-end character–environment fusion for seamless integration.

Strengths

• Better object-interaction fidelity than V1 (cited).
• More diverse motion handling.

Weaknesses

• Code-release status for V2 is unclear from the survey — paper exists, dedicated GitHub repo for V2 was not surfaced.

Open questions

• Code release status of Animate Anyone 2 (paper-only? open weights?).

Sources

• 2026-05-07-ai-avatar-motion-mimicking-models-survey

Related

• mimicmotion
• stableanimator
• wan-animate — also from Alibaba; the DiT-based successor of this lineage.
• unianimate
• pose-driven-vs-audio-driven-avatars