---
title: AnimeGen-T2V-A14B-Lightning-int4
canonical_url: "https://www.modelscope.cn/models/mlx-community/AnimeGen-T2V-A14B-Lightning-int4"
md_url: "https://www.modelscope.cn/models/mlx-community/AnimeGen-T2V-A14B-Lightning-int4.md"
repository: mlx-community/AnimeGen-T2V-A14B-Lightning-int4
last_updated: 2026-07-14
license: apache-2.0
model_type:
  - t2v
base_model:
  - aidealab/AnimeGen-T2V
  - Wan-AI/Wan2.2-T2V-A14B
base_model_relation: finetune
parameters: 10.7B
tensor_type:
  - BF16
  - U32
  - F32
library_name:
  - mlx
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 24
stars: 0
tags:
  - text-to-video
  - anime
  - mlx
  - wan2.2
---

# AnimeGen-T2V-A14B-Lightning-int4

> AnimeGen-T2V-A14B-Lightning-int4 - mlx-community 在 ModelScope 开源的模型。AnimeGen-T2V-A14B-Lightning-int4 (MLX)

mlx-community/AnimeGen-T2V-A14B-Lightning-int4 是 ModelScope 魔搭社区上的 10.7B 参数机器学习模型，采用 apache-2.0 许可，基于 aidealab/AnimeGen-T2V、Wan-AI/Wan2.2-T2V-A14B 构建。

- **Repository**: mlx-community/AnimeGen-T2V-A14B-Lightning-int4
- **License**: apache-2.0
- **Parameters**: 10.7B
- **Base model**: aidealab/AnimeGen-T2V, Wan-AI/Wan2.2-T2V-A14B
- **Tags**: text-to-video, anime, mlx, wan2.2
- **Downloads**: 24
- **Stars**: 0
- **Last updated**: 2026-07-14

Source: https://www.modelscope.cn/models/mlx-community/AnimeGen-T2V-A14B-Lightning-int4

---

# AnimeGen-T2V-A14B-Lightning-int4 (MLX)

An **MLX (Apple Silicon)** conversion of [aidealab/AnimeGen-T2V](https://huggingface.co/aidealab/AnimeGen-T2V) — an
anime-specialized **full fine-tune of Wan2.2-T2V-A14B** (dual-expert MoE) — quantized to **int4** and shipped as a
pre-merged **Lightning** checkpoint for few-step, CFG-free inference.

This variant is built for the Swift-MLX `wan-core` / Bernini-R runtime: it uses the exact flat, 1095-key A14B
checkpoint layout that runtime loads, so the DiT experts drop in unchanged.

## What's in here

Flat checkpoint (stock-Wan VAE + umT5 reused from Wan2.2-T2V-A14B, byte-identical):

| File | What |
|---|---|
| `high_noise_model.safetensors` | fine-tuned high-noise expert, int4-quantized (attention + FFN linears) |
| `low_noise_model.safetensors` | fine-tuned low-noise expert, int4-quantized |
| `vae.safetensors` | 16-channel Wan VAE (stock Wan2.2) |
| `t5_encoder.safetensors` | umT5-XXL text encoder (stock Wan2.2) |
| `config.json` | Wan2.2-A14B config + `quantization` metadata |

## How it was made

1. The two fine-tuned experts from `aidealab/AnimeGen-T2V` (original-Wan key naming, bf16) were converted 1:1 into
   the MLX A14B layout (`sanitize_wan_transformer_weights`; the key set is asserted equal to the base A14B contract,
   1095 keys, 0 added / 0 dropped).
2. AIdeaLab's acceleration recipe — the [lightx2v Wan2.2-Lightning](https://huggingface.co/lightx2v/Wan2.2-Lightning)
   `Wan2.2-T2V-A14B-4steps-lora-250928` LoRA — was **merged offline** per-expert with AIdeaLab's weighting
   (`set_adapters([high, low], [2.0, 1.0])`; kohya `W += strength·(alpha/rank)·(up @ down)`).
3. The attention + FFN linears of both experts were quantized to int4 (group size 64); VAE + umT5 kept full precision.

## Recommended inference recipe (from the AnimeGen model card)

- Sampler: **euler**, **CFG-free** (guidance 1.0)
- **8 steps**, **shift 3.0** (FlowMatchEuler)
- 832×480 or 1280×720, frames `16·seconds + 1` (must be 4n+1), 16 fps

## Footprint

int4 measured **~42 GB peak** at 832×480 / 17 frames / 8 steps on an M-series GPU (fits a 64 GB Mac).

## License & attribution

Apache-2.0. Derived from `aidealab/AnimeGen-T2V` (AIdeaLab, developed with GENIAC/METI support) and the lightx2v
Lightning LoRA. This is an unofficial community MLX conversion; not affiliated with AIdeaLab.