ai-video-generation

Seeding ingest from autoresearch survey of AI avatar generation models with motion mimicking. Established the thread's initial taxonomy: 11 entity pages (named models), 3 concept pages (cross-cutting architecture topics), 4 question pages (gaps the source itself flagged). Wiki was empty before this ingest.

Created entities:

• animate-anyone-2 (entity): Alibaba HumanAIGC pose-driven model with environment affordance.
• dispose-pose-conditioning (entity): plug-and-play disentangled-pose ControlNet.
• echomimic-v2 (entity): Ant Group audio-driven semi-body model.
• heygen-avatar-v (entity): HeyGen closed talking-head with full-token reference attention.
• mimicmotion (entity): Tencent SVD-based pose-driven animation.
• omnihuman-1-5 (entity): ByteDance closed audio-driven full-body model.
• persona-avatar-3d (entity): 3D Gaussian Splatting + diffusion hybrid (slug prefixed to avoid generic "persona" collision).
• runway-act-two (entity): Runway performance-capture model; covers Act-One and Act-Two.
• stableanimator (entity): CVPR 2025 ID-preserving framework.
• unianimate (entity): UniAnimate + UniAnimate-DiT.
• wan-2-2 (entity): the parent open video model family.
• wan-animate (entity): the dedicated avatar variant; current open-source SOTA.

Created concepts:

• identity-preservation-video-diffusion (concept): four families of identity-preservation strategies.
• pose-driven-vs-audio-driven-avatars (concept): the driving-signal taxonomy.
• video-diffusion-unet-to-dit (concept): the 2025 backbone migration.

Created questions:

• dreamactor-m1 (question): characterize this Wan-Animate competitor.
• hand-fidelity-comparison-across-avatar-models (question): apples-to-apples hand quality comparison.
• independent-avatar-benchmarks (question): vendor claims vs independent benchmarks.
• real-time-avatar-puppeting (question): consumer-hardware real-time feasibility.

Updated index.md: populated all four sections (Entities, Concepts, Open Questions, Sources) and added a one-line thread summary; previously TODO.

Notes:

• No contradictions surfaced — wiki was empty, so all claims came in clean.
• Vault-wide basename uniqueness verified for all 18 new pages before writing.
• All entity benchmark claims marked as author/vendor self-reported where they came from primary papers without independent verification — that uncertainty is captured in independent-avatar-benchmarks.
• DreamActor-M1, Aurora, Kling 3.0 Motion Control, and LPM 1.0 are mentioned in the source but not yet given dedicated entity pages — they had thinner technical detail in the survey. Worth filling in if a future source touches them.

Thread ai-video-generation scaffolded via /thread-create.

• Scope: undefined — set in SCOPE.md when ready