ai-video-generation

refreshed 1, created 0, skipped 0

• refreshed ai-video-by-job ← independent avatar-benchmarks academic pass: VBench is general T2V; talking-head benches exist but do not name HeyGen/Act-Two/Wan-Animate; pose-driven fairness gap (HumanVid) + author-tied HyperMotionX. Question left open.

Academic-research pass on independent avatar benchmarks (queue independent-avatar-benchmarks-academic). Consensus abstracts only; priors skipped; T2V Elo / pricing not redone.

• Created vbench (concept): VBench / VBench++ / VBench-2.0 as independent general T2V/I2V suites, not an avatar task board.
• Created talking-head-evaluation-benchmarks (concept): Chen 2020, THEval, Zhang 2026, Pei 2024.
• Created humanvid (entity): names the private-data fairness gap for pose-driven animation.
• Created hypermotionx-bench (entity): public complex-motion bench, author-tied.
• Updated independent-avatar-benchmarks: still open; talking-head academic benches exist; no retrieved live avatar/identity/hand/commercial board.
• Updated identity-preservation-video-diffusion: identity is a talking-head eval axis; VBench identity-inconsistency is a general T2V dim.
• Updated dispose-pose-conditioning: reported VBench gains are author-applied general-video scores.
• Updated hand-fidelity-comparison-across-avatar-models: no animation hand board in this corpus.
• Updated pose-driven-vs-audio-driven-avatars: eval coverage is asymmetric (talking-head benches vs pose-driven gap).
• Updated artificial-analysis-video-arena: VBench-class is a second general-video object; Elo not restated.
• Updated index.md: new entities, concepts, source; question line notes still-open.
• No mechanism page — source is an evaluation-object survey, not a ≥2-step causal chain.
• No contradictions requiring adjudication.

Extended 1, created 0, skipped 0

• extended ai-video-by-job ← extended ← video-diffusion-unet-to-dit

created 1

• created ai-video-by-job ← what-are-the-best-tools-the-best-performing-most-cost, ai-video-quality-versus-cost, artificial-analysis-video-arena, pose-driven-vs-audio-driven-avatars, identity-preservation-video-diffusion

Queue ai-video-generation tools pass: independent AA Elo + official list prices for general T2V/I2V; avatar lane updated one-line only.

• Created gemini-omni-flash (entity): AA T2V-with-audio leader, official ~$0.10/s.
• Created minimax-h3 (entity): open-weight (partial) near-tie on Elo.
• Created seedance-2-0 (entity): AA I2V-with-audio leader.
• Created wan-2-7 (entity): Apr 2026 suite still on the AA Wan rows.
• Created wan-3-0 (entity): Aug 2026 30 s hosted API; no AA Elo yet.
• Created veo-3-1 (entity): official Standard $0.40/s, mid-pack Elo.
• Created artificial-analysis-video-arena (concept): the ranking method.
• Created ai-video-quality-versus-cost (concept): frontier Elo cluster is also the cheaper API cluster.
• Updated what-are-the-best-tools-the-best-performing-most-cost: status open → narrowed with dated pick-list.
• Updated wan-2-2: pointer to 2.7 / 3.0 successors.
• Updated heygen-avatar-v: 2026 credit rates (rate card vs FAQ tension).
• Updated runway-act-two: 5 credits/sec on the 2026-08-18 Academy table.
• Updated independent-avatar-benchmarks: AA exists but is not avatar-specific.
• Updated wan-animate: related links only (no SOTA rewrite).
• Updated index.md: new entities, concepts, question, source.
• No mechanism page — source is a tools/price survey, not a ≥2-step causal chain.
• No contradictions requiring adjudication (vendor Elo vs live AA filed as an open tension on the AA concept page).

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