Files read on Hugging Face: 2026-10-24

Mochi 1 preview GGUF files and Mac RAM

Mochi 1 preview is a 10B-parameter text-to-video model from Genmo built on an Asymmetric Diffusion Transformer. It generates 480p video and is a research preview, and one complete Mac measurement exists, on a 64 GB M4 Max.

Genmo · Updated 2026-10-24
Answer

No published Mac run with a GGUF build yet. The smallest complete GGUF download is 7.67 GB; a download that fits in memory does not prove the model runs.

Smallest Mac with a published run
none yetunknown below
Smallest complete download
7.67 GBcalcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAE
Files read
2026-10-24Hugging Face

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

Table 1

Which exact GGUF files make a complete build?

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

calcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAE

gguf · Q3_K_M · 7.67 GB complete

Mac routes: ComfyUI

Exact component files for calcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAE
FileComponentRequiredDownloadSource
mochi-q3_k_m.ggufcalcuis/mochiAsymmDiTditYes4.31 GBHugging FaceRead 2026-10-24
t5xxl_fp16-q4_0.ggufcalcuis/mochiT5-XXL text encodertext_encoderYes2.90 GBHugging FaceRead 2026-10-24
mochi_vae_fp8_e4m3fn.safetensorscalcuis/mochiAsymmVAEvaeYes0.46 GBHugging FaceRead 2026-10-24

Build sources: calcuis mochi GGUF 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 GBcalcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAEQ3_K_M · ComfyUI7.67 GBunknownunknown
16 GBcalcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAEQ3_K_M · ComfyUI7.67 GBunknownunknown
24 GBcalcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAEQ3_K_M · ComfyUI7.67 GBunknownunknown
32 GBcalcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAEQ3_K_M · ComfyUI7.67 GBunknownunknown
48 GBcalcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAEQ3_K_M · ComfyUI7.67 GBunknownunknown
64 GBcalcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAEQ3_K_M · ComfyUI7.67 GBunknownunknown
96 GBcalcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAEQ3_K_M · ComfyUI7.67 GBunknownunknown
128 GBcalcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAEQ3_K_M · ComfyUI7.67 GBunknownunknown

How does it rank in blind votes?

#48 of 48 in text to video blind votes, from 5,893 votes.

Quality: blind human votes from mochi-v1 (arena.ai), CC BY 4.0, snapshot 2026-09-22. Scores are for the Arena checkpoint, not your local quantized build.

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

What does the official configuration say?

"Official Diffusers BF16 example: 22GB VRAM (CUDA). Single-GPU operation of the genmo harness needs about 60GB VRAM. No Mac statement."

Source: Mochi 1 model card · Read 2026-10-24.

What known issues should you check?

  • Known issue 1

    The only complete Mac measurement ran with 19.8 GB of swap on a 64 GB machine.

    Source: Zenn: Mochi 1 on M4 Max 64GB · Read 2026-10-24

  • Known issue 2

    The official card calls Mochi 1 a living and evolving checkpoint under the research preview, with 480p output.

    Source: Mochi 1 model card · Read 2026-10-24

Sources