Local AI video generation on Mac: what fits?
Apple Silicon can generate video locally, but the complete model bundle and workload matter. Compare video builds by unified memory, app, license and published measurement instead of relying on a single weight file.
Quick answer
Choose your memory tier below. Every model row and fit verdict comes from the shared media dataset and engine. Missing runtime evidence stays unknown.
Get the local AI newsletter for your Mac
One practical Thursday email with new local AI models, memory requirements and Mac setup guides.
For: MacBook Air M5 16 GB
You'll also get the weekly ModelFit email. By signing up you agree to our Privacy Policy. Unsubscribe from either email anytime, in one click.
Which apps can generate video locally?
- Draw Things lists Wan 2.2 and LTX-2.3 among its supported models and provides local generation in a native offline app.
- ComfyUI supports image and video workflows in its node graph, with reusable workflow JSON and custom nodes.
- Use only the app routes named on each model row. Support for one model or file format does not establish support for another.
Table 1
Which video models fit your Mac?
The eight tiers use the same engine as the media configurator. Fit and tight verdicts count as compatible in the tier summaries; no and unknown do not.
Run counts fit and tight verdicts; unknown means no verdict is established. Time appears only for an exact published match.
8 GB unified memory1 unknown · 4 no
| Model | Fits | Quality | App |
|---|---|---|---|
| Mochi 1 previewApache License 2.0 | unknownComplete: calcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAE7.7 GB download | #48 of 48 in text to video5,893 votessnapshot 2026-09-22 | Diffusers, ComfyUI |
| LTX-2.3LTX-2 Community License Agreement | Does not fit est.Complete: Unsloth GGUF dev Q4_K_M with GGUF Gemma 3 (ComfyUI)26.7 GB download | No independent score yet | Lightricks/LTX-2 ltx-pipelines, appautomaton/ltx-video-mlx, ComfyUI |
| MiniMax H3MiniMax H3 Community License Agreement | Does not fit est.Complete: Unsloth GGUF Q4_K denoiser with the Q4_K_M Qwen3-VL encoder35.5 GB download | #8 of 48 in text to video11,417 votessnapshot 2026-09-22 | Argus-AiTeam/minimax-h3-mac (MLX streaming), antirez/h3.c (Metal), ddalcu/mlx-serve, stable-diffusion.cpp, ComfyUI |
| Wan2.2 A14BApache License 2.0 | Does not fit est.Complete: QuantStack GGUF Q4_K_M (high and low noise)26.3 GB download | #42 of 48 in text to video10,458 votessnapshot 2026-09-22 | PocketAI Wan2.2 MLX (pocketai-wan-video), ComfyUI |
| Wan2.2 TI2V 5BApache License 2.0 | Does not fit est.Complete: QuantStack GGUF Q8_013.5 GB download | No independent score yet | Blaizzy/mlx-video, FastVideo MLX (FastMetal), ComfyUI |
Quality: blind human votes from Arena (arena.ai), CC BY 4.0, snapshot 2026-09-22. Scores are for the Arena checkpoint, not your local quantized build.
16 GB unified memory2 run · 2 unknown · 2 no
| Model | Fits | Quality | App |
|---|---|---|---|
| LTX-2.3LTX-2 Community License Agreement | Runs · Stated16 GB unified memory · ltx-2-mlx int4 pack · ltx-2-mlx README | No independent score yet | Lightricks/LTX-2 ltx-pipelines, appautomaton/ltx-video-mlx, ComfyUI |
| Wan2.1Apache License 2.0 | Runs · Stated16 GB unified memory · mlx-video 4-bit conversion · mlx-video Wan README | No independent score yet | ml-explore/mlx-examples Wan2.1, ComfyUI, FastVideo MLX (FastMetal) |
| Mochi 1 previewApache License 2.0 | unknownComplete: calcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAE7.7 GB download | #48 of 48 in text to video5,893 votessnapshot 2026-09-22 | Diffusers, ComfyUI |
| Wan2.2 TI2V 5BApache License 2.0 | unknownComplete: QuantStack GGUF Q8_013.5 GB download | No independent score yet | Blaizzy/mlx-video, FastVideo MLX (FastMetal), ComfyUI |
| MiniMax H3MiniMax H3 Community License Agreement | Does not fit est.Complete: Unsloth GGUF Q4_K denoiser with the Q4_K_M Qwen3-VL encoder35.5 GB download | #8 of 48 in text to video11,417 votessnapshot 2026-09-22 | Argus-AiTeam/minimax-h3-mac (MLX streaming), antirez/h3.c (Metal), ddalcu/mlx-serve, stable-diffusion.cpp, ComfyUI |
| Wan2.2 A14BApache License 2.0 | Does not fit est.Complete: QuantStack GGUF Q4_K_M (high and low noise)26.3 GB download | #42 of 48 in text to video10,458 votessnapshot 2026-09-22 | PocketAI Wan2.2 MLX (pocketai-wan-video), ComfyUI |
Quality: blind human votes from Arena (arena.ai), CC BY 4.0, snapshot 2026-09-22. Scores are for the Arena checkpoint, not your local quantized build.
24 GB unified memory4 run · 1 unknown · 1 no
| Model | Fits | Quality | App |
|---|---|---|---|
| LTX-2.3LTX-2 Community License Agreement | Runs · PublishedApple M5 Pro · 24 GB unified memory · ltx-2-mlx-swift int8 streamed, 704x512, 121 frames, 150 s · ltx-2-mlx-swift README | No independent score yet | Lightricks/LTX-2 ltx-pipelines, appautomaton/ltx-video-mlx, ComfyUI |
| MiniMax H3MiniMax H3 Community License Agreement | Runs · PublishedComplete: Argus calibrated INT8 DiT with the official text encoder and VAEs115.5 GB download | #8 of 48 in text to video11,417 votessnapshot 2026-09-22 | Argus-AiTeam/minimax-h3-mac (MLX streaming), antirez/h3.c (Metal), ddalcu/mlx-serve, stable-diffusion.cpp, ComfyUI |
| Wan2.1Apache License 2.0 | Runs · Stated16 GB unified memory · mlx-video 4-bit conversion · mlx-video Wan README | No independent score yet | ml-explore/mlx-examples Wan2.1, ComfyUI, FastVideo MLX (FastMetal) |
| Wan2.2 TI2V 5BApache License 2.0 | Runs · PublishedComplete: FastMetal 5B QAD (DMD2-distilled INT8 derivative, not the base model)19.5 GB download | No independent score yet | Blaizzy/mlx-video, FastVideo MLX (FastMetal), ComfyUI |
| Mochi 1 previewApache License 2.0 | unknownComplete: calcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAE7.7 GB download | #48 of 48 in text to video5,893 votessnapshot 2026-09-22 | Diffusers, ComfyUI |
| Wan2.2 A14BApache License 2.0 | Does not fit est.Complete: QuantStack GGUF Q4_K_M (high and low noise)26.3 GB download | #42 of 48 in text to video10,458 votessnapshot 2026-09-22 | PocketAI Wan2.2 MLX (pocketai-wan-video), ComfyUI |
Quality: blind human votes from Arena (arena.ai), CC BY 4.0, snapshot 2026-09-22. Scores are for the Arena checkpoint, not your local quantized build.
32 GB unified memory5 run · 2 unknown
| Model | Fits | Quality | App |
|---|---|---|---|
| LTX-2.3LTX-2 Community License Agreement | Runs · PublishedApple M2 Pro · 32 GB unified memory · ltx-2-mlx bf16 distilled with block streaming, 480x704, 33 frames · ltx-2-mlx project notes | No independent score yet | Lightricks/LTX-2 ltx-pipelines, appautomaton/ltx-video-mlx, ComfyUI |
| LTX-2.5LTX-2.x Community License Agreement | Runs · PublishedApple M2 Pro · 32 GB unified memory · ltx-2-mlx q8 with --low-ram, 768x512, 49 frames, 277 s · ltx-2-mlx pipelines docs | No independent score yet | Lightricks/LTX-2 ltx-pipelines, ComfyUI |
| MiniMax H3MiniMax H3 Community License Agreement | Runs · Published on M4 Pro · 24 GBComplete: Argus calibrated INT8 DiT with the official text encoder and VAEs115.5 GB download | #8 of 48 in text to video11,417 votessnapshot 2026-09-22 | Argus-AiTeam/minimax-h3-mac (MLX streaming), antirez/h3.c (Metal), ddalcu/mlx-serve, stable-diffusion.cpp, ComfyUI |
| Wan2.1Apache License 2.0 | Runs · Stated32 GB unified memory · Draw Things, 720p up to 81 frames · Draw Things wiki: video generation | No independent score yet | ml-explore/mlx-examples Wan2.1, ComfyUI, FastVideo MLX (FastMetal) |
| Wan2.2 TI2V 5BApache License 2.0 | Runs · Published on M5 · 24 GBComplete: FastMetal 5B QAD (DMD2-distilled INT8 derivative, not the base model)19.5 GB download | No independent score yet | Blaizzy/mlx-video, FastVideo MLX (FastMetal), ComfyUI |
| Mochi 1 previewApache License 2.0 | unknownComplete: calcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAE7.7 GB download | #48 of 48 in text to video5,893 votessnapshot 2026-09-22 | Diffusers, ComfyUI |
| Wan2.2 A14BApache License 2.0 | unknownComplete: PocketAI MLX Q4 (group size 64)28.6 GB download | #42 of 48 in text to video10,458 votessnapshot 2026-09-22 | PocketAI Wan2.2 MLX (pocketai-wan-video), ComfyUI |
Quality: blind human votes from Arena (arena.ai), CC BY 4.0, snapshot 2026-09-22. Scores are for the Arena checkpoint, not your local quantized build.
48 GB unified memory5 run · 2 unknown
| Model | Fits | Quality | App |
|---|---|---|---|
| LTX-2.3LTX-2 Community License Agreement | Runs · Stated48 GB unified memory · LTX Desktop on macOS 26+, full local generation · LTX Desktop README (Lightricks) | No independent score yet | Lightricks/LTX-2 ltx-pipelines, appautomaton/ltx-video-mlx, ComfyUI |
| LTX-2.5LTX-2.x Community License Agreement | Runs · Stated48 GB unified memory · LTX Desktop on macOS 26+, full local generation · LTX Desktop README (Lightricks) | No independent score yet | Lightricks/LTX-2 ltx-pipelines, ComfyUI |
| MiniMax H3MiniMax H3 Community License Agreement | Runs · PublishedComplete: ComfyUI GGUF Q5_K_M pruned denoiser with Parasyte Turbo (Bambushu workflow)36.6 GB download | #8 of 48 in text to video11,417 votessnapshot 2026-09-22 | Argus-AiTeam/minimax-h3-mac (MLX streaming), antirez/h3.c (Metal), ddalcu/mlx-serve, stable-diffusion.cpp, ComfyUI |
| Wan2.1Apache License 2.0 | Runs · Stated32 GB unified memory · Draw Things, 720p up to 81 frames · Draw Things wiki: video generation | No independent score yet | ml-explore/mlx-examples Wan2.1, ComfyUI, FastVideo MLX (FastMetal) |
| Wan2.2 TI2V 5BApache License 2.0 | Runs · Published on M4 Max · 36 GBComplete: FastMetal 5B QAD (DMD2-distilled INT8 derivative, not the base model)19.5 GB download | No independent score yet | Blaizzy/mlx-video, FastVideo MLX (FastMetal), ComfyUI |
| Mochi 1 previewApache License 2.0 | unknownComplete: calcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAE7.7 GB download | #48 of 48 in text to video5,893 votessnapshot 2026-09-22 | Diffusers, ComfyUI |
| Wan2.2 A14BApache License 2.0 | unknownComplete: PocketAI MLX Q4 (group size 64)28.6 GB download | #42 of 48 in text to video10,458 votessnapshot 2026-09-22 | PocketAI Wan2.2 MLX (pocketai-wan-video), ComfyUI |
Quality: blind human votes from Arena (arena.ai), CC BY 4.0, snapshot 2026-09-22. Scores are for the Arena checkpoint, not your local quantized build.
64 GB unified memory8 run
| Model | Fits | Quality | App |
|---|---|---|---|
| HunyuanVideo-1.5Tencent Hunyuan Community License | Runs · PublishedApple M2 Ultra · 64 GB unified memory · ComfyUI official workflow, I2V 832x480, 73 frames · ComfyUI discussion #15109 | #38 of 48 in text to video4,292 votessnapshot 2026-09-22 | ComfyUI |
| LTX-2.3LTX-2 Community License Agreement | Runs · Stated48 GB unified memory · LTX Desktop on macOS 26+, full local generation · LTX Desktop README (Lightricks) | No independent score yet | Lightricks/LTX-2 ltx-pipelines, appautomaton/ltx-video-mlx, ComfyUI |
| LTX-2.5LTX-2.x Community License Agreement | Runs · PublishedApple M1 Max · 64 GB unified memory · KinoMLX FP16, 768x448, 5 s clip in 1 min 45 s · KinoMLX LTX-2.5 performance | No independent score yet | Lightricks/LTX-2 ltx-pipelines, ComfyUI |
| MiniMax H3MiniMax H3 Community License Agreement | Runs · Published on M5 Pro · 48 GBComplete: ComfyUI GGUF Q5_K_M pruned denoiser with Parasyte Turbo (Bambushu workflow)36.6 GB download | #8 of 48 in text to video11,417 votessnapshot 2026-09-22 | Argus-AiTeam/minimax-h3-mac (MLX streaming), antirez/h3.c (Metal), ddalcu/mlx-serve, stable-diffusion.cpp, ComfyUI |
| Mochi 1 previewApache License 2.0 | Runs · PublishedComplete: Official Diffusers BF16 variant40.0 GB download | #48 of 48 in text to video5,893 votessnapshot 2026-09-22 | Diffusers, ComfyUI |
| Wan2.1Apache License 2.0 | Runs · Stated32 GB unified memory · Draw Things, 720p up to 81 frames · Draw Things wiki: video generation | No independent score yet | ml-explore/mlx-examples Wan2.1, ComfyUI, FastVideo MLX (FastMetal) |
| Wan2.2 A14BApache License 2.0 | Runs · PublishedApple M1 Max · 64 GB unified memory · ComfyUI Q5_K_M I2V with the Lightning 4-step LoRA, 480x320, 25 frames, 182.7 s · umamon on Zenn | #42 of 48 in text to video10,458 votessnapshot 2026-09-22 | PocketAI Wan2.2 MLX (pocketai-wan-video), ComfyUI |
| Wan2.2 TI2V 5BApache License 2.0 | Runs · Published on M4 Max · 36 GBComplete: FastMetal 5B QAD (DMD2-distilled INT8 derivative, not the base model)19.5 GB download | No independent score yet | Blaizzy/mlx-video, FastVideo MLX (FastMetal), ComfyUI |
Quality: blind human votes from Arena (arena.ai), CC BY 4.0, snapshot 2026-09-22. Scores are for the Arena checkpoint, not your local quantized build.
96 GB unified memory8 run
| Model | Fits | Quality | App |
|---|---|---|---|
| HunyuanVideo-1.5Tencent Hunyuan Community License | Runs · PublishedApple M2 Ultra · 64 GB unified memory · ComfyUI official workflow, I2V 832x480, 73 frames · ComfyUI discussion #15109 | #38 of 48 in text to video4,292 votessnapshot 2026-09-22 | ComfyUI |
| LTX-2.3LTX-2 Community License Agreement | Runs · Stated48 GB unified memory · LTX Desktop on macOS 26+, full local generation · LTX Desktop README (Lightricks) | No independent score yet | Lightricks/LTX-2 ltx-pipelines, appautomaton/ltx-video-mlx, ComfyUI |
| LTX-2.5LTX-2.x Community License Agreement | Runs · PublishedApple M1 Max · 64 GB unified memory · KinoMLX FP16, 768x448, 5 s clip in 1 min 45 s · KinoMLX LTX-2.5 performance | No independent score yet | Lightricks/LTX-2 ltx-pipelines, ComfyUI |
| MiniMax H3MiniMax H3 Community License Agreement | Runs · Published on M5 Pro · 48 GBComplete: ComfyUI GGUF Q5_K_M pruned denoiser with Parasyte Turbo (Bambushu workflow)36.6 GB download | #8 of 48 in text to video11,417 votessnapshot 2026-09-22 | Argus-AiTeam/minimax-h3-mac (MLX streaming), antirez/h3.c (Metal), ddalcu/mlx-serve, stable-diffusion.cpp, ComfyUI |
| Mochi 1 previewApache License 2.0 | Runs · Published on M4 Max · 64 GBComplete: Official Diffusers BF16 variant40.0 GB download | #48 of 48 in text to video5,893 votessnapshot 2026-09-22 | Diffusers, ComfyUI |
| Wan2.1Apache License 2.0 | Runs · Stated32 GB unified memory · Draw Things, 720p up to 81 frames · Draw Things wiki: video generation | No independent score yet | ml-explore/mlx-examples Wan2.1, ComfyUI, FastVideo MLX (FastMetal) |
| Wan2.2 A14BApache License 2.0 | Runs · PublishedApple M1 Max · 64 GB unified memory · ComfyUI Q5_K_M I2V with the Lightning 4-step LoRA, 480x320, 25 frames, 182.7 s · umamon on Zenn | #42 of 48 in text to video10,458 votessnapshot 2026-09-22 | PocketAI Wan2.2 MLX (pocketai-wan-video), ComfyUI |
| Wan2.2 TI2V 5BApache License 2.0 | Runs · Published on M4 Max · 36 GBComplete: FastMetal 5B QAD (DMD2-distilled INT8 derivative, not the base model)19.5 GB download | No independent score yet | Blaizzy/mlx-video, FastVideo MLX (FastMetal), ComfyUI |
Quality: blind human votes from Arena (arena.ai), CC BY 4.0, snapshot 2026-09-22. Scores are for the Arena checkpoint, not your local quantized build.
128 GB unified memory8 run
| Model | Fits | Quality | App |
|---|---|---|---|
| HunyuanVideo-1.5Tencent Hunyuan Community License | Runs · PublishedApple M2 Ultra · 64 GB unified memory · ComfyUI official workflow, I2V 832x480, 73 frames · ComfyUI discussion #15109 | #38 of 48 in text to video4,292 votessnapshot 2026-09-22 | ComfyUI |
| LTX-2.3LTX-2 Community License Agreement | Runs · PublishedComplete: Official FP8 dev checkpoint, quantized to 4-bit at load61.1 GB download | No independent score yet | Lightricks/LTX-2 ltx-pipelines, appautomaton/ltx-video-mlx, ComfyUI |
| LTX-2.5LTX-2.x Community License Agreement | Runs · PublishedApple M1 Max · 64 GB unified memory · KinoMLX FP16, 768x448, 5 s clip in 1 min 45 s · KinoMLX LTX-2.5 performance | No independent score yet | Lightricks/LTX-2 ltx-pipelines, ComfyUI |
| MiniMax H3MiniMax H3 Community License Agreement | Runs · PublishedComplete: MLX Serve FL2VA 4-bit pack with Turbo LoRA41.1 GB download | #8 of 48 in text to video11,417 votessnapshot 2026-09-22 | Argus-AiTeam/minimax-h3-mac (MLX streaming), antirez/h3.c (Metal), ddalcu/mlx-serve, stable-diffusion.cpp, ComfyUI |
| Mochi 1 previewApache License 2.0 | Runs · Published on M4 Max · 64 GBComplete: Official Diffusers BF16 variant40.0 GB download | #48 of 48 in text to video5,893 votessnapshot 2026-09-22 | Diffusers, ComfyUI |
| Wan2.1Apache License 2.0 | Runs · Stated32 GB unified memory · Draw Things, 720p up to 81 frames · Draw Things wiki: video generation | No independent score yet | ml-explore/mlx-examples Wan2.1, ComfyUI, FastVideo MLX (FastMetal) |
| Wan2.2 A14BApache License 2.0 | Runs · PublishedComplete: PocketAI MLX Q4 (group size 64)28.6 GB download | #42 of 48 in text to video10,458 votessnapshot 2026-09-22 | PocketAI Wan2.2 MLX (pocketai-wan-video), ComfyUI |
| Wan2.2 TI2V 5BApache License 2.0 | Runs · Published on M4 Max · 36 GBComplete: FastMetal 5B QAD (DMD2-distilled INT8 derivative, not the base model)19.5 GB download | No independent score yet | Blaizzy/mlx-video, FastVideo MLX (FastMetal), ComfyUI |
Quality: blind human votes from Arena (arena.ai), CC BY 4.0, snapshot 2026-09-22. Scores are for the Arena checkpoint, not your local quantized build.
Why does the exact clip matter?
Resolution, duration, frame count, frame rate, steps and quantization can all change a workload. A published time applies only to its recorded model, build, Mac and settings.
A model that fits for one clip may fail for a larger one. When exact runtime costs or a matching measurement are unavailable, the table does not invent a fit or time.
How should you choose a Mac for local video?
Start with the models and clip settings you need, then compare real Apple configurations by their engine fit counts. More unified memory increases capacity, but this guide does not turn hardware specifications into an unmeasured speed ranking.
Frequently asked questions
Can a Mac generate AI video without a cloud service?
Yes, supported open-weight models can run locally through apps such as Draw Things and ComfyUI. Check the exact model row for its app route and memory verdict.
Is 16 GB enough for local AI video?
It depends on the model build and clip. Open the 16 GB tier for catalog verdicts. Unknown means the runtime evidence is insufficient, not that the model fits.
Does a small GGUF file mean the whole model is small?
No. The complete workflow can also require text encoders, a VAE, audio components and runtime memory. The table evaluates complete builds when the required data is available.
How fast is local AI video on a Mac?
There is no single useful number. Time appears only when a published measurement matches the model, build, Mac and workload; otherwise it remains unknown. ModelFit does not run home benchmarks for these guides.
Sources
Official documentation and source code unless the label identifies an individual issue report.