How to run MimicMotion on a cloud GPU
Rent a GPU on Deploy, then open a terminal in JupyterLab or over SSH. Install MimicMotion from the official Tencent/MimicMotion repository the way the README prints, download the weights that file names, and run python inference.py --inference_config configs/test.yaml. Copy the mp4 under outputs/ off the instance before you destroy it, because destroying the instance deletes its disk.
Portrait animation from a driving video is How to run LivePortrait on a cloud GPU. Audio-driven portrait video whose README lists tested GPUs as A100(80G), RTX4090D (24G), and V100(16G) is How to run EchoMimicV2 on a cloud GPU. Image-to-video checkpoints in ComfyUI are How to run Stable Video Diffusion in ComfyUI on a cloud GPU. Installing ComfyUI itself is How to install ComfyUI and ComfyUI-Manager on a cloud GPU. If the process runs out of memory: How to fix CUDA out of memory on a cloud GPU.
The repository LICENSE file says MimicMotion is licensed under the Apache License Version 2.0 except for the third-party components listed in that file. The copyright line is Copyright (C) 2023 Tencent. The GitHub license field on 2026-11-13 was Other. The model card for tencent/MimicMotion that day had cardData.license other and gated false, lastModified 2025-07-29T09:23:10.000Z. The weight LICENSE file on that model repository is a different text: it says to use MimicMotion only for academic, research and education purposes, and to refrain from using it for any commercial or production purposes. The model NOTICE says the weights are fine-tuned with Stable Video Diffusion 1.1 under the Stability AI Community License, and that agreement says the license terminates if annual revenue is more than USD $1,000,000. That dollar figure is the upstream license threshold. Read both files before you redistribute the program or the weights.
The commands below are copied from the upstream files named in each step. They were not run on a CloudGPU instance for this page. Anything this page adds around them is marked.
What the README says about memory
README.md (commit ba2044950de7cf371443b38250ffa439aca8434f, 2025-06-26) says it recommends Python 3+ with torch 2.x, validated with an Nvidia V100 GPU. It does not say which V100. It says that for the initial released checkpoint, videos have a maximum of 72 frames at a 576x1024 resolution, and that if you encounter insufficient memory you can reduce the number of frames. The same file’s news line for 2024-07-08 says the 1.1 checkpoint expanded the maximum from 16 frames to 72. This page quotes both sentences and does not merge them into one claim.
The VRAM section says the 72-frame model requires 16GB VRAM and, in parentheses, names a 4060ti. That name is not a row in gpu-prices.json. The same sentence also states a runtime on a 4090. This page does not copy that runtime. The next sentence says the minimum VRAM requirement for the 16-frame U-Net model is 8GB, and that the VAE decoder demands 16GB, with the option to run the VAE decoder on CPU. A later tip says to try PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:256 when GPU memory is limited. That assignment is not a GB figure.
GET /v1/public/compute-capabilities does not list GPUs. It returns feature flags. The memory column is vram_gb from gpu-prices.json, updated_at 2026-11-13T16:03:04.945Z. GET /v1/public/gpu-catalog the same day (generated_at 2026-11-13T16:03:03.877Z) listed the same names and the same vram_gb values, plus Ascend 910B at 64, which is not in gpu-prices.json and is not listed here. The catalog response also includes an availability field. This page does not copy it. Every prices row has gpu_count 1. Every catalog row has min_gpu_count 1. Catalog V100 16G is 16 and catalog V100 32G is 32. The README does not say which of those the validation used. This page does not mark a row as enough. Which listing you can start is on Deploy.
The catalog template list had no MimicMotion template.
| Catalog name | vram_gb |
|---|---|
| RTX 2080 | 8 |
| RTX 3080 | 10 |
| RTX 2080 Ti | 11 |
| RTX 3080 Ti | 12 |
| V100 16G | 16 |
| A30 | 24 |
| RTX 3090 | 24 |
| RTX 4090 | 24 |
| RTX 4090D | 24 |
| RTX 4080 SUPER 32G | 32 |
| RTX 5090 | 32 |
| RTX 5090D | 32 |
| V100 32G | 32 |
| A100 40G | 40 |
| A800 40G | 40 |
| A40 | 48 |
| RTX 4090 48G | 48 |
| RTX 4090D 48G | 48 |
| RTX A6000 | 48 |
| NVIDIA RTX PRO 5000 72G | 72 |
| RTX PRO 5000 72G | 72 |
| A100 80G | 80 |
| A800 80G | 80 |
| H800 | 80 |
| RTX 6000D | 83 |
| H20 96G | 96 |
| RTX PRO 6000 96G | 96 |
| RTX PRO 6000 Max-Q | 96 |
| H20-3e 141G | 141 |
Install
On Deploy, choose Raw GPU. The README install block does not print git clone. environment.yaml is a file in that repository, and the first command is conda env create -f environment.yaml. The clone below is this page’s, and it comes first so that file exists. It was not run. A conda environment is the first command in the README block: How to set up a conda environment on a cloud GPU.
# The README install block does not print a clone. These two lines are this page's. They were not run.
git clone https://github.com/Tencent/MimicMotion.git
cd MimicMotion
The install block, copied.
conda env create -f environment.yaml
conda activate mimicmotion
environment.yaml pins python=3.11, pytorch=2.0.1, torchvision=0.15.2, and pytorch-cuda=11.7. The pip section pins diffusers==0.27.0, transformers==4.32.1, and decord==0.6.0. environment.yaml was commit 1c91d48d3c0f15a5a20691beb71b81e80eed2e8b, committed 2024-07-01T02:42:05Z. This page did not run conda.
The RTX 5090 guide says that card is compute capability 12.0 (sm_120) and that PyTorch 2.7 and later, on a build for CUDA 12.8 or newer, is the line that announced support. environment.yaml pins pytorch=2.0.1 and pytorch-cuda=11.7. This page did not re-test MimicMotion on an RTX 5090. Catalog RTX 5090 and RTX 5090D are 32. That is the catalog number only.
Weights
The README says that if you have connection issues with Hugging Face, set HF_ENDPOINT as follows. That line is copied. This page did not run it.
export HF_ENDPOINT=https://hf-mirror.com
The README then prints cd MimicMotions/ and mkdir models. git clone https://github.com/Tencent/MimicMotion.git creates a directory named MimicMotion, not MimicMotions. This page does not put cd MimicMotions/ in the runnable block. The downloads below are the README’s, and they assume the shell is already in the clone directory from the install section.
mkdir models
mkdir -p models/DWPose
wget https://huggingface.co/yzd-v/DWPose/resolve/main/yolox_l.onnx?download=true -O models/DWPose/yolox_l.onnx
wget https://huggingface.co/yzd-v/DWPose/resolve/main/dw-ll_ucoco_384.onnx?download=true -O models/DWPose/dw-ll_ucoco_384.onnx
wget -P models/ https://huggingface.co/tencent/MimicMotion/resolve/main/MimicMotion_1-1.pth
The README says the tree should then be models/DWPose/yolox_l.onnx, models/DWPose/dw-ll_ucoco_384.onnx, and models/MimicMotion_1-1.pth. It says the SVD model stabilityai/stable-video-diffusion-img2vid-xt-1-1 will be automatically downloaded. configs/test.yaml sets base_model_path to that same id and ckpt_path to models/MimicMotion_1-1.pth. The model NOTICE names the original SVD model as stabilityai/stable-video-diffusion-img2vid-xt, without the -1-1 suffix. This page does not treat those two ids as the same repository. This page did not run the download.
Inference
The README’s inference command, copied. It says a sample configuration is test.yaml, and that you can modify the configurations.
python inference.py --inference_config configs/test.yaml
configs/test.yaml (commit 08f5e7c6315577ffe91b52eecb776b519927d4ed, 2024-07-09) sets ref_video_path to assets/example_data/videos/pose1.mp4, ref_image_path to assets/example_data/images/demo1.jpg, num_frames to 72, resolution to 576, frames_overlap to 6, num_inference_steps to 25, noise_aug_strength to 0, guidance_scale to 2.0, sample_stride to 2, fps to 15, and seed to 42. The documented command does not pass --output_dir.
inference.py defaults --output_dir to outputs/. It creates that directory, writes a log named with datetime.now() as %Y%m%d%H%M%S, and saves the mp4 as {basename of the reference video, without the extension}_{that same timestamp pattern}.mp4. For the documented config the video basename is pose1, so the prefix is pose1_. The clock value is not fixed. inference.py was commit 4bc49f24f0af3222d4fbd4a7f863be59f9bd1ba4, committed 2024-07-09T10:45:03Z. This page did not run it.
The README’s limited-memory tip, copied as an assignment. It was not run.
export PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:256
ComfyUI
The README does not document ComfyUI. The ComfyUI examples top-level directory list on 2026-11-13 had no directory named mimicmotion. Examples HEAD was f9431bb000ce792094ff345446e22cac1ea6cef3, committed 2025-11-26T20:33:58Z. This page does not copy a community ComfyUI install. Installing ComfyUI itself is How to install ComfyUI and ComfyUI-Manager on a cloud GPU.
The README does not print a host or a port. This page does not state one. The tunnel page, for programs that do listen, is How to use SSH port forwarding on a cloud GPU.
When it fails
CUDA out of memory. The README’s figures are 16GB for the 72-frame model, 8GB as the 16-frame U-Net minimum, and 16GB for the VAE decoder, with the option to run that decoder on CPU. It also says to reduce the number of frames, and to try PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:256. If the process raises that error, use how to fix CUDA out of memory on a cloud GPU.
“CUDA error: no kernel image is available for execution on the device” on an RTX 5090. See the install section. This page did not re-test the pinned torch build on that card.
The download directory name does not match. The README prints cd MimicMotions/. The clone URL in the install section creates MimicMotion. This page did not run either form.
Save the video, then destroy
Copy outputs/ from the documented inference.py command before you destroy the instance. Save your work before you destroy a cloud GPU is the checklist. The terms say data on local instance storage is permanently deleted when the instance is destroyed.
The pricing page FAQ states the billing rule in these words:
Per-minute options have a 1-minute minimum; unused prepaid minutes are refunded when you destroy the instance. Some configurations use whole-hour billing with a 1-hour minimum; unused time within a charged hour is not refunded on voluntary stop. The billing period, minimum charge and initial prepayment are shown before confirmation. No subscription.
Frequently asked questions
What GPU memory figures does the MimicMotion README print?
The README says the 72-frame model requires 16GB VRAM and names a 4060ti in parentheses. That card is not a row in gpu-prices.json. The same sentence states a runtime. This page does not copy that runtime. The README also says the minimum VRAM requirement for the 16-frame U-Net model is 8GB and that the VAE decoder demands 16GB, with the option to run the VAE decoder on CPU. It says the setup was validated with an Nvidia V100 and does not say which V100. The smallest vram_gb in gpu-prices.json (updated_at 2026-11-13T16:03:04.945Z) is 8, on RTX 2080. This page did not run MimicMotion and does not mark a row as enough. Which listing you can start is on https://cloudgpu.app/deploy.
Where does the video go?
inference.py defaults --output_dir to outputs/. It writes the reference-video basename, without the extension, then an underscore, then a timestamp from datetime.now(), then .mp4, in that directory. It also writes a log with that timestamp pattern. configs/test.yaml sets ref_video_path to assets/example_data/videos/pose1.mp4, so the basename prefix is pose1_. This page did not run the command and does not print a fixed filename.
Does the README document a web UI or ComfyUI?
The README inference command is python inference.py --inference_config configs/test.yaml. It does not print a host or a port. The ComfyUI examples top-level directory list on 2026-11-13 had no mimicmotion directory. This page does not copy a community ComfyUI install. This page did not start a server.
Which license covers the code and the weights?
The repository LICENSE file says MimicMotion is licensed under the Apache License Version 2.0 except for the third-party components listed in that file. The GitHub license field on 2026-11-13 was Other. The model card license field was other. The weight LICENSE file says to use MimicMotion only for academic, research and education purposes and to refrain from using it for any commercial or production purposes. The NOTICE says the weights are fine-tuned with Stable Video Diffusion 1.1 under the Stability AI Community License, and that agreement says licenses terminate above USD $1,000,000 in annual revenue. That figure is the upstream license threshold. Section 2 of the Apache file grants a no-charge, royalty-free copyright license for the Work that file covers. That wording is the upstream license, not a CloudGPU price and not a storage allowance.
Sources (read 2026-11-13, Asia/Shanghai)
- MimicMotion README (Python 3+ and torch 2.x, V100 sentence, conda block,
cd MimicMotions/, DWPose andMimicMotion_1-1.pthdownloads, SVD auto-download,python inference.py --inference_config configs/test.yaml, 16GB / 8GB / 16GB sentences,PYTORCH_CUDA_ALLOC_CONF): https://github.com/Tencent/MimicMotion/blob/main/README.md commitba2044950de7cf371443b38250ffa439aca8434f. Repository HEAD is6907bdcc259a6a048d41a365e840d22274f9256c, committed 2025-11-18T06:29:06Z environment.yaml(python=3.11,pytorch=2.0.1,pytorch-cuda=11.7,diffusers==0.27.0): commit1c91d48d3c0f15a5a20691beb71b81e80eed2e8bconfigs/test.yaml(stabilityai/stable-video-diffusion-img2vid-xt-1-1,models/MimicMotion_1-1.pth,pose1.mp4,num_frames72): commit08f5e7c6315577ffe91b52eecb776b519927d4edinference.py(--output_dirdefaultoutputs/, mp4 and log names): commit4bc49f24f0af3222d4fbd4a7f863be59f9bd1ba4- Repository
LICENSE(Apache 2.0 except listed third-party components, copyright 2023 Tencent): commitc053153a1d124abae8c08568925ae88debc63001. GitHub license field Other - Model card
tencent/MimicMotion(cardData.licenseother,gatedfalse,lastModified2025-07-29T09:23:10.000Z, shadb35a2c5d0a12db9155060f14688e72dedbfb370). WeightLICENSE(academic, research and education; no commercial or production) andNOTICE(Stability AI Community License, more than USD $1,000,000 annual revenue, original model idstabilityai/stable-video-diffusion-img2vid-xt), read 2026-11-13 - ComfyUI examples top-level directories (no
mimicmotiondirectory): https://github.com/comfyanonymous/ComfyUI_examples commitf9431bb000ce792094ff345446e22cac1ea6cef3 - Public GPU memory: https://cloudgpu.app/gpu-prices.json (
updated_at2026-11-13T16:03:04.945Z) and https://cloudgpu.app/v1/public/gpu-catalog (generated_at2026-11-13T16:03:03.877Z).GET /v1/public/compute-capabilitiesreturned feature flags and no GPU list. The catalog had no MimicMotion template
Cards you can start are on the Deploy page. Prices are on https://cloudgpu.app/pricing and are not repeated here.
Last updated: 2026-11-13 (Asia/Shanghai).