Pin official source and exact weights
Mutable model names are not immutable checkpoints.
Record exact 40-character SHA, file inventory, config and runtime.
Unpinned dependencies can change outputs.
MEITUAN LONGCAT · FIELD VERIFIED
Primary sources last checked on 17 Nov 2026. Verification freshness is tracked separately from factual change, and unknown values remain explicit rather than inferred.
MODEL INTELLIGENCE · SOURCE-FIRST
Open Model Weights records LongCat-Video-Avatar 1.6 as a Pending verification-parameter audio-driven video generation open-weight model in the Audio-driven video generation family, with Meituan LongCat listed as the developer. The checked source repository provides 21 recognized weight artifacts, format evidence for ONNX, PyTorch, Safetensors and a context window that is not declared in the checked standard sources. Values that are not evidenced in the checked sources remain explicitly undisclosed or unknown.
The repository declares mit. Commercial use allowed under mit conditions. Open Model Weights reports the checked license evidence and classification; this is not legal advice.
The checked sources declare meituan-longcat/LongCat-Video as lineage evidence. Training recipe status: not-disclosed-in-checked-repository; data disclosure status: not-disclosed-in-standard-metadata-or-obvious-model-card-section.
+ Save to my shortlistMonitor changes →Local browser monitoring · no background email alerts
DIRECT ANSWER · FEATURED MODEL BRIEFING
Source-grounded editorial layer · reviewed 2026-11-17
Meituan LongCat-Video-Avatar 1.6 is a publicly downloadable audio-driven human video-generation update with separate base-model weight shards, optional INT8 DiT shards, two LoRA adapters and audio-related support components. Its official repository declares MIT license terms in the model card and metadata; no separate standard license file was confirmed for this exact revision. Some inference workflows also require independently distributed LongCat-Video foundation weights and supporting tools, so the release must not be described as a single self-sufficient 77.6 GB model. Open Model Weights has not reproduced its video-quality or throughput claims.
WHY IT MATTERS
Meituan's 1.6 release focuses on audio-led avatar video, including lip synchronization, facial expressions, coordinated head and body motion and temporal consistency across long sequences. Its official model card describes both single-character and multi-character generation.
The official repository contains 21 recognized weight and index paths: six base diffusion Safetensors shards and their index, optional four-part INT8 model and index, two task adapters, and separately packaged Whisper, vocal-separation and other support weights.
Meituan's setup instructions separately download the LongCat-Video foundation repository and configure a specified Python/PyTorch and attention-backend environment. The optional INT8 model is not evidence that every model component has been quantized or that a particular consumer GPU will suffice.
MODEL-SELECTION VIEW
Meituan's 1.6 release focuses on audio-led avatar video, including lip synchronization, facial expressions, coordinated head and body motion and temporal consistency across long sequences. Its official model card describes both single-character and multi-character generation.
The update is a specialist continuation of the earlier LongCat-Video-Avatar line. It should not be confused with a newly pretrained general-purpose image or video model, and the publisher's own preference studies are not OMW measurements.
The official repository contains 21 recognized weight and index paths: six base diffusion Safetensors shards and their index, optional four-part INT8 model and index, two task adapters, and separately packaged Whisper, vocal-separation and other support weights.
The published roughly 77.6 GB repository tree describes downloaded package volume, not measured peak GPU memory. Developers must distinguish mutually optional weight variants from files required simultaneously for a chosen inference mode.
Meituan's setup instructions separately download the LongCat-Video foundation repository and configure a specified Python/PyTorch and attention-backend environment. The optional INT8 model is not evidence that every model component has been quantized or that a particular consumer GPU will suffice.
The LoRA directories include DMD and on-policy distillation adapters that serve different roles. Pinned dependency versions, audio preprocessing, memory headroom and compatible CUDA kernels should be recorded before making cost or speed comparisons.
The model card describes video continuation beyond ten minutes and a six-system human-preference study with 33,051 valid individual votes. Its reported second-place position depends on the chosen clips, participants and comparison procedure.
No independent OMW reproduction of temporal identity preservation, lip alignment, human preference or end-to-end throughput has been performed. A proper evaluation should include multiple voices, languages, audio lengths, face poses, scene changes and failure rates.
The publisher marks the 1.6 model weights as MIT in the official card and repository metadata. No distinct standard LICENSE file was detected at this immutable revision; the checked source therefore supports a metadata-declared classification, not a separately verified license-file claim.
Some packaging includes third-party components and requires other model downloads. License notices, patent and trademark scope, consent for likeness and voice, and applicable portrait/privacy rules must be checked individually.
This release is notable for reproducible research into audio-driven character performance and long-form avatar generation, not for verified universal superiority or guaranteed real-time output.
Before deployment, pin both avatar and LongCat-Video repositories, audit each component's rights, reproduce the exact LoRA/int8 choice, and measure generated lip-sync accuracy, frame quality, memory usage and latency under clearly documented input conditions.
IMPLEMENTATION & EVALUATION
Source-grounded deployment checklist; no Open Model Weights inference or benchmark was performed.
Mutable model names are not immutable checkpoints.
Record exact 40-character SHA, file inventory, config and runtime.
Unpinned dependencies can change outputs.
Repository labels do not cover every downstream dependency.
Check original model license declarations, extra licenses and required notices.
Unlicensed redistribution, privacy violations or incompatible integrations.
Source size and publisher numbers are not peak runtime requirements.
Record hardware, software versions, memory, latency, inputs and failure modes.
OOM, unsupported operators and unstable behavior.
Publisher measurements are not independent model rankings.
Compare pinned baselines on held-out representative tasks, report negative results.
False attribution of vendor performance to OMW.
Proposed tests, not OMW benchmark results. Read the official model card at the pinned source revision ↗ Inspect OMW field evidence ↗
PRIMARY & INDEPENDENT SOURCES
FAQ
Yes. The pinned official repository lists public Safetensors shards, LoRA adapters and supporting weights.
No. Repository volume includes alternative variants and support files, and memory depends on runtime configuration.
Yes. The publisher lists optional base_model_int8 shards as weight-only INT8 for the DiT.
No. The six-system, 33,051-vote study is attributed to Meituan, not reproduced by Open Model Weights.
No. The official instructions also reference separate LongCat-Video foundation weights and runtime components.
The official 1.6 model card and metadata declare MIT. A separate license file was not verified in the checked revision.
03 · PRACTICAL TECHNICAL PROFILE
Inspect weight-memory estimates, model lineage and runtime evidence separately from promotional or untested performance claims.
DEPLOYMENT PROFILE
These bars visualize weight-only decimal-GB estimates. They exclude KV cache, activations, runtime overhead, optimizer state and sharding.
LICENSE & USE
Commercial use allowed under mit conditions
This section summarizes the checked repository evidence. It is not legal advice.
LINEAGE & PROVENANCE
Open Model Weights does not infer a hidden base model or training corpus when the checked sources do not declare one.
RUNTIME SIGNALS
Declared and model-card-mentioned runtimes are discovery signals. They are not treated as Open Model Weights runtime tests unless explicitly measured.
EVIDENCE TRAIL
This is where the article remains auditable: source links on one side, the registry's observed change history on the other.
Checked Hugging Face API metadata, exact repository file list, model card and config when accessible.
04 · EVIDENCE AND UPDATES
Inspect dated source checks and any differences between published claims and the evidence; these checks are not independent model benchmarks.
SOURCE-BASED VISUAL BRIEF · NOTABLE
Three evidence-linked visual summaries: parameter scale, illustrative precision memory and verifiable provenance. No performance ranking is implied.
No comparable single-model parameter total is available.
No reliable single-model weight-only memory figure.
FRESHNESS & VERIFICATION
A verification check does not pretend the model changed. Open Model Weights updates the verification date after a real primary-source check, while factual-update dates move only when retained source evidence changes.
CONFLICT INTELLIGENCE
05 · SOURCE RECORD
Full repository-level attributes, licensing, format details and provenance are available below without hiding or rewriting the original data.
01 · WEIGHTS
base_model/diffusion_pytorch_model-00001-of-00006.safetensorsbase_model/diffusion_pytorch_model-00002-of-00006.safetensorsbase_model/diffusion_pytorch_model-00003-of-00006.safetensorsbase_model/diffusion_pytorch_model-00004-of-00006.safetensorsbase_model/diffusion_pytorch_model-00005-of-00006.safetensorsbase_model/diffusion_pytorch_model-00006-of-00006.safetensorsbase_model/diffusion_pytorch_model.safetensors.index.jsonbase_model_int8/quantized_model-00001-of-00004.safetensorsbase_model_int8/quantized_model-00002-of-00004.safetensorsbase_model_int8/quantized_model-00003-of-00004.safetensorsbase_model_int8/quantized_model-00004-of-00004.safetensorsbase_model_int8/quantized_model.safetensors.index.jsonloraA/lora.safetensorsloraB/lora.safetensorsvocal_separator/Kim_Vocal_2.onnxwhisper-large-v3/model.fp32-00001-of-00002.safetensorswhisper-large-v3/model.fp32-00002-of-00002.safetensorswhisper-large-v3/model.safetensorswhisper-large-v3/pytorch_model.binwhisper-large-v3/pytorch_model.fp32-00001-of-00002.binwhisper-large-v3/pytorch_model.fp32-00002-of-00002.bin02 · LICENSE
Classification describes the checked license terms; it is not legal advice.
03 · HARDWARE
Weight-only estimate; excludes runtime overhead.
04 · FORMATS
“Not observed” means no matching repository artifact/name/dtype signal was found; it does not prove no third-party conversion exists.
05 · LINEAGE
06 · TRAINING ASSETS
07 · RUNTIME SUPPORT
Open Model Weights separates model-repository evidence from external runtime-release evidence. An external support statement does not change the official model repository's weight formats and does not imply an Open Model Weights inference benchmark.
08 · SOURCES
09 · VERIFICATION HISTORY
Checked Hugging Face API metadata, exact repository file list, model card and config when accessible.
MODELS FOR AI AGENTS · ROLE EVIDENCE
Specialized model. A supporting component. This checkpoint does not, by category alone, serve as the planning and tool-execution controller.
The curated notable tier is not an agent-readiness rating. Model-specific tool-call success, long-horizon task accuracy and compatible runtime configurations are not independently established here.
Source-linked license: mit (allowed commercial-use classification; see original terms).
Context: Unknown or not declared.
BF16 weights: Not calculated.
Compare agent model candidates → · Agent evaluation method → · Inspect original model repository ↗
These records are separately sourced. They do not replace independent model-plus-runtime tests.
No model-specific agent capability evaluation is recorded for this checkpoint. Missing evidence is unknown, not a failed task.
Read evidence details and benchmarks → · Use the agent model selector →