Datasheet for local AI49 models98 GPUsData read 2026-09-25
No adsNo tracking
Check my GPU →

Can the Mac 16 GB run Wan 2.1 I2V 14B 720P?

Video model · 16.4B16 GB unified · ~12.7 GB for the GPU · 68.25–200 GB/s · Apple SiliconData 2026-09-25
Offload
Not entirely in VRAM, but it can still run.

The smallest sensible file, Q4_K_M (11.3 GB), needs about 18.1 GB — 5.4 GB more than Mac 16 GB has. ComfyUI can still run it by streaming part of the model from system RAM. How much slower that is depends on your ComfyUI version, the file format and the PCIe link — see the note below.

Best fileQ4_K_M
File size11.3 GB
GPU memory needed~18.1 GB
Shared memory25 / 16 GB
Memory map · Mac 16 GB18.1 GB needed · 5.4 GB over 12.7 GB
09 GB18 GB
Weights Q4_K_M · 11.3 GBWorking memory · 6.0 GBReserve · 0.8 GBSpills to system RAM · 5.4 GB
calculated from real file sizes plus working memory. Not yet measured on this setup. How this works.

01What to download

PartFileFolderSize
Modelwan2.1-i2v-14b-720p-Q4_K_M.gguf
Q4_K_M · city96/Wan2.1-I2V-14B-720P-gguf
models/unet11.3 GBDownload →
Text encoderumt5-xxl-encoder-Q8_0.gguf
UMT5-XXL GGUF Q8_0 · city96/umt5-xxl-encoder-gguf
models/text_encoders6.0 GBDownload →
VAEwan_2.1_vae.safetensors
Wan 2.1 VAE · Comfy-Org/Wan_2.1_ComfyUI_repackaged
models/vae0.3 GBDownload →
Also neededclip_vision_h.safetensors
CLIP Vision H · Comfy-Org/Wan_2.1_ComfyUI_repackaged
models/clip_vision1.3 GBDownload →
Total download · keep about the same free on disk18.9 GB

Alternatives: UMT5-XXL FP16 (11.4 GB). The 16-bit text encoder is a little more faithful; the smaller one is chosen here because it loads faster and needs less RAM. On a Mac, FP8 text encoders do not load, so a 16-bit one is listed — or a GGUF one, loaded with “CLIPLoader (GGUF)” from the ComfyUI-GGUF nodes. Sizes read from Hugging Face (2026-09-25). “Download” links start the file directly; the file name links are the same files the official ComfyUI workflows use.

Shared memory: about 24.9 GB of your 16 GB. On this machine the model, the text encoder, the VAE and the operating system all use the same memory. That is about 8.9 GB more than there is. It still runs: while the model works, the system compresses or swaps the idle text encoder, so loading and changing the prompt get slower. A smaller file or text encoder avoids that. calculated

02Every Wan 2.1 I2V 720P file on 12.7 GB

FileSizeNeededOn this cardQualityDownload
16-bit
SAFETENSORS · Comfy-Org
32.8 GB39.6 GB26.9 GB too bigthe original weightsHugging Face →
Q8_0
GGUF · city96
18.1 GB24.9 GB12.2 GB too bigpractically identical to the originalHugging Face →
Q6_K
GGUF · city96
14.2 GB21.0 GB8.3 GB too bigvery close to the originalHugging Face →
Q5_K_M
GGUF · city96
12.7 GB19.5 GB6.8 GB too bigclose; small differences in fine detailHugging Face →
Q4_K_M ←
GGUF · city96
11.3 GB18.1 GB5.4 GB too biggood; some loss in fine detail and textHugging Face →
Q3_K_M
GGUF · city96
8.6 GB15.4 GB2.7 GB too bignoticeable loss of detailHugging Face →

Sizes from the Hugging Face file listing, read 2026-09-25. “Needed” = file + 6 GB working memory for this model + 0.8 GB kept free for the system. FP8 and INT8 files are left out on a Mac: Apple GPUs cannot compute FP8, so ComfyUI either fails to load them or converts them back to 16-bit, which saves no memory. Use a 16-bit or GGUF file.

03The text encoder

UMT5-XXL + CLIP Vision H: 11.4 GB as 16-bit, 6.7 GB as FP8, 3.7 GB as GGUF Q4_K_M (plus CLIP Vision H (1.3 GB) for the input image). ComfyUI encodes the prompt first and can push the encoder out of VRAM before sampling, so it does not have to fit together with the model. On 12.7 GB the FP8 encoder fits on its own, so prompt encoding stays fast.

04About this GPU

A Mac with 16 GB shares that memory between the CPU and the GPU. On current macOS the GPU may use about 12.7 GB of it by default (older macOS versions: about 11.5 GB); that is the figure used here. ComfyUI can go past it, but macOS then starts compressing and swapping memory and everything slows down. ComfyUI runs on Apple GPUs through PyTorch's MPS backend. 16-bit and GGUF files work; FP8 and INT8 files do not save memory on a Mac, so they are skipped. Speed is the catch: even the fastest Macs are several times slower per image than a desktop RTX card. Chips sold with 16 GB: M1, M1 Pro, M2, M2 Pro, M3, M4, M5, M6 (memory bandwidth 68.25–200 GB/s — the higher, the faster). Everything about Macs and local AI →

Mac 16 GB: 16 GB unified · ~12.7 GB for the GPU · 68.25–200 GB/s · Apple Silicon. Everything that runs on the Mac 16 GB →

05What more VRAM would change

With 32 GB you could run Wan 2.1 I2V 720P at 8-bit or better (Q8_0, 18.1 GB) with no offloading: for example on the RTX 5090 32 GB.

06Measured and reported results

Nobody has sent measured numbers for this pair yet. If you run Wan 2.1 I2V 720P on a Mac 16 GB, send your time per image and peak VRAM and it will appear here, credited.

07Questions

How much VRAM does Wan 2.1 I2V 720P need?

Around 15.4 GB with the smallest file (Q3_K_M) and 23.2 GB with an 8-bit file (FP8), counting working memory and a small system reserve. The full 16-bit file needs about 39.6 GB.

Which Wan 2.1 I2V 720P file should I download for the Mac 16 GB?

Q4_K_M (11.3 GB) from city96/Wan2.1-I2V-14B-720P-gguf. Load it with Unet Loader (GGUF) from the ComfyUI-GGUF node pack.

Is FP8 faster than GGUF on the Mac 16 GB?

No. This GPU has no FP8 compute, so ComfyUI converts FP8 weights back before the maths. FP8 only saves memory here; Q8_0 GGUF is the closer-to-original 8-bit pick.

How much do I need to download for Wan 2.1 I2V 720P on the Mac 16 GB?

About 18.9 GB for the model file, text encoder and VAE listed on this page, and the same again free on disk. With a 12.7 GB card, 32 GB of system RAM or more is recommended.