Pith Number

pith:L3NA6IVR

pith:2026:L3NA6IVRNT557KRRZYRICUT76V

not attested not anchored not stored refs resolved

Head Forcing: Long Autoregressive Video Generation via Head Heterogeneity

Chi Zhang, Gang Yu, Jiahao Tian, Yiwei Wang

Attention heads in autoregressive video diffusion transformers naturally divide into local, anchor, and memory roles, enabling a training-free Head Forcing method to generate minute-long videos by assigning each type specialized KV cache策略.

arxiv:2605.14487 v1 · 2026-05-14 · cs.CV · cs.AI

Open paper page JSON Open Graph Bundle Merged state Verified badge What is a Pith Number?

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{L3NA6IVRNT557KRRZYRICUT76V}

Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge

Record completeness

1 Bitcoin timestamp

2 Internet Archive

3 Author claim open · sign in to claim

4 Citations open

5 Replications open

✓ Portable graph bundle live · download bundle · merged state

The bundle contains the canonical record plus signed events. A mirror can host it anywhere and recompute the same current state with the deterministic merge algorithm.

Claims

C1strongest claim

Without additional training, Head Forcing extends generation from 5 seconds to minute-level duration, supports multi-prompt interactive synthesis, and consistently outperforms existing baselines.

C2weakest assumption

That attention heads in AR video diffusion transformers naturally and reliably fall into distinct functional categories (local for detail refinement, anchor for structural stabilization, memory for long-range context) that can be identified and assigned effective tailored KV cache strategies without any model-specific training or validation.

C3one line summary

Head Forcing assigns tailored KV cache strategies to local, anchor, and memory attention heads plus head-wise RoPE re-encoding to extend autoregressive video generation from seconds to minutes without training.

References

73 extracted · 73 resolved · 26 Pith anchors

[1] Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets 2023 · arXiv:2311.15127

[2] In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition 2023

[3] In: Forty-first International Conference on Machine Learning (2024) 2024

[4] arXiv:2508.03841 (2025).https://doi.org/10.48550/arXiv.2508 2025 · doi:10.48550/arxiv.2508

[5] Advances in Neural Information Processing Systems37, 24081–24125 (2024) 2024

Receipt and verification

First computed	2026-05-17T23:39:06.477970Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

5eda0f22b16cfbdfaa31ce2281527ff57f9b9bd32944180336922f00e088c04d

Aliases

arxiv: 2605.14487 · arxiv_version: 2605.14487v1 · doi: 10.48550/arxiv.2605.14487 · pith_short_12: L3NA6IVRNT55 · pith_short_16: L3NA6IVRNT557KRR · pith_short_8: L3NA6IVR

Agent API

Resolver JSON Graph JSON Events JSON Schema Signing key

Verify this Pith Number yourself

curl -sH 'Accept: application/ld+json' https://pith.science/pith/L3NA6IVRNT557KRRZYRICUT76V \
  | jq -c '.canonical_record' \
  | python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: 5eda0f22b16cfbdfaa31ce2281527ff57f9b9bd32944180336922f00e088c04d

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "1d4ec4f37b71cade6bb4662ffbf739d1161d570f08b28cb2cb42a080a6ea2a43",
    "cross_cats_sorted": [
      "cs.AI"
    ],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "primary_cat": "cs.CV",
    "submitted_at": "2026-05-14T07:27:39Z",
    "title_canon_sha256": "37768ef9866f318c71d0fcbfc4e35c6f2acaca235c13e38eeda1ee854f4d176b"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.14487",
    "kind": "arxiv",
    "version": 1
  }
}