README.md

---
license: apache-2.0
base_model: Wan-AI/Wan2.2-TI2V-5B-Diffusers
pipeline_tag: text-to-video
library_name: mlx-gen
tags:
- mlx
- mlx-gen
- mflux
- apple-silicon
- bf16
- wan
- wan2.2
- video-generation
- text-to-video
- image-to-video
---
# wan2.2-ti2v-5b-diffusers-bf16

This repository contains BF16 MLX-Gen saved weights for
[`Wan-AI/Wan2.2-TI2V-5B-Diffusers`](https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B-Diffusers).
It is designed for local Apple Silicon inference with
[`mlx-gen`](https://github.com/lpalbou/mlx-gen).

It uses the mflux/MLX saved-weight layout. It is not a Diffusers or Transformers
`from_pretrained()` checkpoint.

## Source Model

Original model: [`Wan-AI/Wan2.2-TI2V-5B-Diffusers`](https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B-Diffusers).

This prepared derivative follows the Apache 2.0 license of the source model.

## Precision

The upstream TI2V-5B source snapshot is not uniformly 16-bit on disk: the transformer and VAE
safetensors are FP32, while the UMT5 text encoder is BF16. MLX-Gen loads Wan transformer/VAE
weights at BF16 runtime precision, so this prepared BF16 package reduces storage and download size
but is not a runtime-memory optimization versus source generation.

Use this package when you want a smaller reusable MLX-Gen folder that preserves source behavior.
Use the mixed q8/BF16 package when you want a smaller model footprint.

## Measurements

Measured on 2026-06-04 with `mlx-gen 0.18.10` on an Apple M5 Max with 128 GiB unified memory.

Validation profile: `1280x704`, 17 frames, 20 denoising steps, guidance `5`, 24 fps, seed `321`,
explicit empty negative prompt.

| Layout | Storage | Logical Model | Full-Process Physical Peak | Max RSS | MLX Peak | Total Time | Output |
| --- | ---: | ---: | ---: | ---: | ---: | ---: | --- |
| Upstream source snapshot | 31.9 GiB | 10.6 GiB | 102.7 GiB | 13.7 GiB | 58.5 GiB | 216.2 s | [base-source.mp4](validation/ti2v5b-clean/base-source.mp4) |
| This BF16 package | 21.2 GiB | 10.6 GiB | 102.6 GiB | 14.5 GiB | 58.5 GiB | 261.6 s | [prepared-bf16.mp4](validation/ti2v5b-clean/prepared-bf16.mp4) |
| Mixed q8/BF16 package | 16.9 GiB | 6.3 GiB | 103.7 GiB | 13.8 GiB | 54.2 GiB | 243.4 s | [mixed-q8-bf16.mp4](validation/ti2v5b-clean/mixed-q8-bf16.mp4) |

The source and this BF16 package produced byte-identical decoded MP4 frames. The mixed q8/BF16
package stayed visually in the same family with mean frame MAE `1.66` versus source/BF16.

`Storage` is the Hugging Face repository total. `Logical Model` is the loaded Wan transformer plus
VAE tensor footprint measured from MLX arrays. `Full-Process Physical Peak` is Darwin
`phys_footprint` sampled from model initialization through MP4 save and health validation.

Validation assets:

- [contact-sheet.png](validation/ti2v5b-clean/contact-sheet.png)
- [metrics.json](validation/ti2v5b-clean/metrics.json)

## Usage

```bash
python -m pip install -U mlx-gen

mlxgen download --model AbstractFramework/wan2.2-ti2v-5b-diffusers-bf16

mlxgen generate \
  --model AbstractFramework/wan2.2-ti2v-5b-diffusers-bf16 \
  --prompt "A short cinematic video of a glowing orange glass sphere floating above calm teal water, soft reflections, gentle camera movement" \
  --negative-prompt "" \
  --width 1280 \
  --height 704 \
  --frames 17 \
  --steps 20 \
  --guidance 5 \
  --fps 24 \
  --seed 321 \
  --output video.mp4
```

TI2V-5B also supports first-frame image-to-video in MLX-Gen when one input image is supplied.

## Attribution

MLX-Gen is based on [mflux](https://github.com/filipstrand/mflux) by Filip Strand and the original
mflux contributors.

Prepared and contributed by [@lpalbou](https://huggingface.co/lpalbou).