What does MuSS report about keeping the same subject across separate shots when a reference image is given, and why do reference-conditioned models look good on reference fidelity but fail between shots?
MuSS (arXiv 2604.23789, v3 11 Aug 2026, ACM MM 2026) pairs a movie-derived multi-shot dataset with a benchmark that decouples identity to the reference (Ref-Sub.Con) from identity between generated shots (Inter-Sub.Con, DINOv2 and InsightFace) and adds ACP-Var and CP-Rate for copy-paste. On 50 Track-2 test items the authors report Phantom 75.16 / 55.20 and VACE-14B 72.10 / 52.40 (Ref / Inter) versus their own EchoShot-based fine-tune 78.50 / 62.27, with CP-Rate 18.50% (Phantom), 21.68% (VACE-14B) and 7.35% (theirs). Multi-shot models without references were scored with a pseudo-reference taken from their first shot, so tables mix protocols. Only data-construction code is public; dataset files, benchmark code and checkpoints are not.
Structured fields
{
"citations": [
{
"url": "https://arxiv.org/html/2604.23789v3",
"note": "producer work copy work/muss_v3.html (sha256 equals pin; local read); producer work copy work/muss.html (sha256 equals pin; local read); producer work copy work/MuSS_README.md (sha256 equals pin; local read); producer work copy work/MuSS_download_README.md (sha256 equals pin; local read)"
},
{
"url": "https://github.com/zhang-haojie/muss",
"note": "cited at https://github.com/zhang-haojie/MuSS; producer work copy work/muss_v3.html (sha256 equals pin; local read); producer work copy work/muss.html (sha256 equals pin; local read); producer work copy work/MuSS_README.md (sha256 equals pin; local read); producer work copy work/MuSS_download_README.md (sha256 equals pin; local read)"
},
{
"url": "https://github.com/zhang-haojie/MuSS",
"note": "observed URL (dated/alias form of the cited catalogue source; canonical identity retained)"
}
]
}Evidence claims
- [sourced_fact] MuSS (arXiv 2604.23789, v3 11 Aug 2026, ACM MM 2026) pairs a movie-derived multi-shot dataset with a benchmark that decouples identity to the reference (Ref-Sub.Con) from identity between generated shots (Inter-Sub.Con, DINOv2 and InsightFace) and adds ACP-Var and CP-Rate for copy-paste. On 50 Track-2 test items the authors report Phantom 75.16 / 55.20 and VACE-14B 72.10 / 52.40 (Ref / Inter) versus their own EchoShot-based fine-tune 78.50 / 62.27, with CP-Rate 18.50% (Phantom), 21.68% (VACE-14B) and 7.35% (theirs). Multi-shot models without references were scored with a pseudo-reference taken from their first shot, so tables mix protocols. Only data-construction code is public; dataset files, benchmark code and checkpoints are not. · locator: producer work copy work/muss_v3.html (sha256 equals pin; local read); producer work copy work/muss.html (sha256 equals pin; local read); producer work copy work/MuSS_README.md (sha256 equals pin; local read); producer work copy work/MuSS_download_README.md (sha256 equals pin; local read) source
Rights
| component | license | hosting | redistribution | modification | training | Evidence and notes |
|---|---|---|---|---|---|---|
| documentation | unknown | permitted | permitted | unknown | unknown | Applies to this original summary only, not to the cited sources. |