Files read on Hugging Face: 2026-10-24

Wan2.2 A14B on Mac: files, RAM and published times

Wan2.2 A14B is a two-expert video model with a high-noise and a low-noise expert of about 14B parameters each, 27B in total and 14B active per step. It makes 5-second videos at 480P and 720P, and on a Mac it runs through community MLX packs.

Wan-AI · Updated 2026-10-24
Answer

Wan2.2 A14B runs on a Mac with 128 GB or more. A published run on M5 Max · 128 GB used the PocketAI MLX Q4 (group size 64) build (source, 2026-08-15).

Smallest Mac with a published run
128 GBM5 Max · 128 GB · PocketAI MLX Q4 (group size 64)
Smallest complete download
26.29 GBQuantStack GGUF Q4_K_M (high and low noise)
Files read
2026-10-24Hugging Face

Apache License 2.0. Commercial use: allowed. Excluded territories: none listed.

See the exact GGUF files and sizes

Table 1

What do you download?

Only builds with every required component and an exact probed file size are shown.

PocketAI MLX Q4 (group size 64)

mlx · Q4 · 28.63 GB complete · Recommended

Mac routes: PocketAI Wan2.2 MLX (pocketai-wan-video)

Exact component files for PocketAI MLX Q4 (group size 64)
FileComponentRequiredDownloadSource
q4/high_noise_model.safetensorsPocketAiHub/Wan2.2-T2V-A14B-MLXHigh-noise expertditYes8.38 GBHugging FaceRead 2026-10-24
q4/low_noise_model.safetensorsPocketAiHub/Wan2.2-T2V-A14B-MLXLow-noise expertditYes8.38 GBHugging FaceRead 2026-10-24
common/t5_encoder.safetensorsPocketAiHub/Wan2.2-T2V-A14B-MLXUMT5-XXL text encodertext_encoderYes11.36 GBHugging FaceRead 2026-10-24
common/vae.safetensorsPocketAiHub/Wan2.2-T2V-A14B-MLXWan2.1 VAEvaeYes0.51 GBHugging FaceRead 2026-10-24

Build sources: PocketAI Wan2.2 T2V A14B MLX card.

QuantStack GGUF Q4_K_M (high and low noise)

gguf · Q4_K_M · 26.29 GB complete

Mac routes: ComfyUI

Exact component files for QuantStack GGUF Q4_K_M (high and low noise)
FileComponentRequiredDownloadSource
HighNoise/Wan2.2-T2V-A14B-HighNoise-Q4_K_M.ggufQuantStack/Wan2.2-T2V-A14B-GGUFHigh-noise expertditYes9.65 GBHugging FaceRead 2026-10-24
LowNoise/Wan2.2-T2V-A14B-LowNoise-Q4_K_M.ggufQuantStack/Wan2.2-T2V-A14B-GGUFLow-noise expertditYes9.65 GBHugging FaceRead 2026-10-24
split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensorsComfy-Org/Wan-2.2UMT5-XXL text encodertext_encoderYes6.74 GBHugging FaceRead 2026-10-24
split_files/vae/wan_2.1_vae.safetensorsComfy-Org/Wan-2.2Wan2.1 VAEvaeYes0.25 GBHugging FaceRead 2026-10-24

Build sources: QuantStack Wan2.2 T2V A14B GGUF card · Comfy-Org Wan 2.2 repackaged card.

Comfy-Org repackaged FP8 scaled

safetensors · FP8 scaled · 35.58 GB complete

Mac routes: ComfyUI

Exact component files for Comfy-Org repackaged FP8 scaled
FileComponentRequiredDownloadSource
split_files/diffusion_models/wan2.2_t2v_high_noise_14B_fp8_scaled.safetensorsComfy-Org/Wan-2.2High-noise expertditYes14.29 GBHugging FaceRead 2026-10-24
split_files/diffusion_models/wan2.2_t2v_low_noise_14B_fp8_scaled.safetensorsComfy-Org/Wan-2.2Low-noise expertditYes14.29 GBHugging FaceRead 2026-10-24
split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensorsComfy-Org/Wan-2.2UMT5-XXL text encodertext_encoderYes6.74 GBHugging FaceRead 2026-10-24
split_files/vae/wan_2.1_vae.safetensorsComfy-Org/Wan-2.2Wan2.1 VAEvaeYes0.25 GBHugging FaceRead 2026-10-24

Build sources: Comfy-Org Wan 2.2 repackaged card.

Table 2

Does it fit your Mac's RAM?

Published names the exact measured machine. est. is a memory calculation. unknown means the run verdict is not established, not that the model fails.

Run verdicts by Mac memory tier
Mac RAMComplete buildDownloadRun verdictBasis
8 GBPocketAI MLX Q4 (group size 64)Q4 · PocketAI Wan2.2 MLX (pocketai-wan-video)28.63 GBDoes not fitest.
16 GBPocketAI MLX Q4 (group size 64)Q4 · PocketAI Wan2.2 MLX (pocketai-wan-video)28.63 GBDoes not fitest.
24 GBPocketAI MLX Q4 (group size 64)Q4 · PocketAI Wan2.2 MLX (pocketai-wan-video)28.63 GBDoes not fitest.
32 GBPocketAI MLX Q4 (group size 64)Q4 · PocketAI Wan2.2 MLX (pocketai-wan-video)28.63 GBunknownunknown
48 GBPocketAI MLX Q4 (group size 64)Q4 · PocketAI Wan2.2 MLX (pocketai-wan-video)28.63 GBunknownunknown
64 GBPocketAI MLX Q4 (group size 64)Q4 · PocketAI Wan2.2 MLX (pocketai-wan-video)28.63 GBunknownunknown
96 GBPocketAI MLX Q4 (group size 64)Q4 · PocketAI Wan2.2 MLX (pocketai-wan-video)28.63 GBunknownunknown
128 GBPocketAI MLX Q4 (group size 64)Q4 · PocketAI Wan2.2 MLX (pocketai-wan-video)28.63 GBRuns · PublishedPublishedApple M5 Max · 128 GB RAM

How does it rank in blind votes?

#42 of 48 in text to video blind votes, from 10,458 votes.

Quality: blind human votes from wan-v2.2-a14b (arena.ai), CC BY 4.0, snapshot 2026-09-22. Scores are for the Arena checkpoint, not your local quantized build.

Table 3

What published Mac times are available?

Third-party results keep their original machine and settings. ModelFit does not run these benchmarks.

Published Mac generation measurements
Machine and RAMSettingsAppResultSource
Apple M5 Max · 128 GB RAM512x320 · 5.1 s · 10 steps · 81 frames · 16 fps · Q4 · UniPC scheduler, guidance 3/4, identical prompt and seed. Time is end to end; peak is maximum process RSS. The repository carries no run date, so the date shown is its last-modified date.PocketAI Wan2.2 MLX (pocketai-wan-video)Success10:2516.37 GiBPocketAI Wan2.2 T2V A14B MLX cardPublished 2026-08-15 · Grade A

Which Mac software routes are listed?

  • ComfyUI

    Official

    Formats: safetensors, gguf. Memory policy: component stages stay resident.

    On Apple Silicon ComfyUI runs on PyTorch MPS, not MLX. GGUF files need the separate city96/ComfyUI-GGUF custom node. Resident size is the sum of the weights, an upper bound: ComfyUI can unload models between stages and publishes no per-model Mac peak.

    Sources: ComfyUI README · ComfyUI system requirements · ComfyUI CLI arguments · ComfyUI-GGUF README

  • PocketAI Wan2.2 MLX (pocketai-wan-video)

    Community

    Formats: mlx. Memory policy: component stages stay resident.

    Sources: PocketAI Wan2.2 T2V A14B MLX card

What does the official configuration say?

"The official T2V runner needs a GPU with at least 80GB VRAM. No official Mac memory minimum was found."

Source: Wan2.2 T2V A14B model card · Read 2026-10-24.

What known issues should you check?

  • Known issue 1

    PocketAI packs are community derivatives, not an official Wan-AI release.

    Source: PocketAI Wan2.2 T2V A14B MLX card · Read 2026-10-24

  • Known issue 2

    The PocketAI benchmark carries no run date. The date shown is the repository last-modified date.

    Source: PocketAI Wan2.2 T2V A14B MLX card · Read 2026-10-24

  • Known issue 3

    Wan2.2 A14B needs both experts for a complete run, so every build lists a high-noise and a low-noise file.

    Source: Wan2.2 README · Read 2026-10-24

Frequently asked questions

Does Wan2.2 A14B run on a Mac?

Yes, through community MLX packs. PocketAI's Q4 pack took 10:25 end to end on an M5 Max with 128 GB for a 512x320, 81-frame clip at 10 steps, with a 16.37 GiB peak process RSS. The official card asks for a GPU with at least 80GB VRAM.

Why does Wan2.2 A14B need two expert files?

The A14B design uses a high-noise expert for the early denoising stages and a low-noise expert for later detail refinement. Each has about 14B parameters, giving 27B in total with 14B active per step, so every build lists both files.

Sources