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

Mac 18 GB for local AI

Which image and video models run on an Apple Silicon Mac with 18 GB of unified memory, which file to download for each, and how much of the memory they need.

Apple18 GB unified · ~14.4 GB for the GPU · 150–150 GB/s · Apple SiliconData 2026-09-25
Unified memory18 GB
For the GPU~14.4 GB
Older macOS~12.9 GB
Bandwidth150–150
ChipsM3 Pro
FP8 filesNo
BackendPyTorch MPS
Runs well18 of 49

Specs: support.apple.com

01What runs on it

ModelVerdictBest fileSizeNeeded
image models
FLUX.1 [dev]RunsQ6_K9.9 GB12.2 GB
FLUX.1 [schnell]RunsQ6_K9.8 GB12.1 GB
FLUX.1 Kontext [dev]RunsQ6_K9.8 GB12.4 GB
FLUX.1 Krea [dev]RunsQ6_K9.8 GB12.1 GB
FLUX.1 Fill [dev]RunsQ6_K9.9 GB12.5 GB
FLUX.2 [dev]Offload onlyQ2_K12.9 GB16.2 GB
FLUX.2 [klein] 9BRuns wellQ8_010.0 GB12.3 GB
FLUX.2 [klein] 4BRuns well16-bit7.8 GB9.6 GB
Krea 2 (Turbo)RunsQ6_K10.6 GB13.2 GB
Qwen-ImageTightQ3_K_M9.7 GB12.5 GB
Qwen-Image-Edit (2511)TightQ3_K_M9.9 GB12.9 GB
Qwen-Image 2.1Runs wellQ8_07.6 GB9.9 GB
Z-Image TurboRuns well16-bit12.3 GB14.3 GB
Z-Image (base)Runs well16-bit12.3 GB14.3 GB
Ideogram 4Offload onlyQ4_16.2 GB14.7 GB
Boogu-Image (Turbo)Runs wellQ8_011.6 GB14.2 GB
ERNIE-Image (Turbo)Runs wellQ8_08.7 GB11.0 GB
HiDream-O1-ImageOffload only16-bit16.4 GB19.2 GB
Mage-Flow (Microsoft)Runs well16-bit8.2 GB10.0 GB
Lumina Image 2.0Runs well16-bit5.2 GB7.0 GB
HiDream-I1 (Full)RunsQ4_K_M11.5 GB14.3 GB
HiDream-I1 (Dev)RunsQ4_K_M11.5 GB14.3 GB
Stable Diffusion 3.5 LargeRuns wellQ8_08.8 GB11.1 GB
Stable Diffusion 3.5 MediumRuns well16-bit5.1 GB6.9 GB
Chroma1-HDRuns wellQ8_09.7 GB12.0 GB
SDXL 1.0Runs well16-bit6.9 GB7.1 GB
Illustrious XL / Pony (SDXL anime)Runs well16-bit6.9 GB7.1 GB
Stable Diffusion 1.5Runs well16-bit2.1 GB3.3 GB
HunyuanImage 2.1TightQ3_K_M9.8 GB13.1 GB
video models
Wan 2.1 T2V 14BTightQ3_K_M7.6 GB11.9 GB
Wan 2.1 T2V 1.3BRuns well16-bit2.8 GB5.6 GB
Wan 2.1 I2V 14B 480PTightQ3_K_M8.6 GB12.9 GB
Wan 2.1 I2V 14B 720POffload onlyQ3_K_M8.6 GB15.4 GB
Wan 2.1 VACE 14BTightQ3_K_S7.8 GB12.6 GB
Wan 2.2 T2V A14BRunsQ4_K_M9.7 GB14.0 GB
Wan 2.2 I2V A14BRunsQ4_K_M9.7 GB14.0 GB
Wan 2.2 TI2V 5BRuns well16-bit10.0 GB13.8 GB
Wan 2.2 Animate 14BTightQ3_K_M8.6 GB13.9 GB
Wan Animate 2 (14B)TightQ3_K_M8.6 GB13.9 GB
Wan 2.2 S2V 14BTightQ2_K9.5 GB14.3 GB
SCAIL-2 (character animation)TightQ2_K7.3 GB12.6 GB
HunyuanVideo (13B, original)RunsQ5_K_M9.4 GB13.7 GB
LTX-Video 13B (0.9.8)RunsQ5_K_M9.8 GB14.1 GB
HunyuanVideo 1.5Runs wellQ8_09.0 GB13.3 GB
LTX-2 (19B)TightQ2_K8.1 GB12.9 GB
LTX-2.3 (22B)TightQ2_K8.3 GB13.1 GB
LTX-2.5 (22B)TightQ2_K8.8 GB13.6 GB
MiniMax H3 (33B)Offload onlyQ4_K_M19.9 GB25.7 GB
MiniMax H3 PrunedOffload onlyQ3_K_M8.9 GB14.7 GB

Calculated from real file sizes plus working memory. How the numbers work.

02Good to know

A Mac with 18 GB shares that memory between the CPU and the GPU. On current macOS the GPU may use about 14.4 GB of it by default (older macOS versions: about 12.9 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 18 GB: M3 Pro (memory bandwidth 150–150 GB/s — the higher, the faster). Everything about Macs and local AI →

03Measured and reported results

No measured results yet. Send yours.