hideodaikoku / mlops

Machine Learning Toolset for Houdini

Houdini MLOPs 2.0

Free and Open Source Machine Learning Plugin for Houdini developed by Ambrosiussen Holding and Entagma, Licensed and Distributed by Bismuth Consultancy B.V. By downloading or using the plugin (or any of its contents), you are agreeing to the LICENSE found in this repository and Terms of Service of Bismuth Consultancy B.V.

Paul Ambrosiussen	Entagma	Discord

Promo Video

Installing for Houdini

To install the plugin for the first time, follow these steps:

1. Clone this repository and make note of the directory you have cloned it to.
2. Copy the MLOPs.json file found in the repository root, and paste it in the $HOUDINI_USER_PREF_DIR/packages/ folder.
3. Edit the MLOPs.json file you just pasted, and modify the $MLOPS path found inside. Set the path to where you cloned the repository to in step one.
4. Install git. Follow the instructions for your relevant OS here.
5. Launch Houdini and open the MLOPs shelf. Click the Install Dependencies shelf button. Restart Houdini once complete.
6. After restarting Houdini, open the MLOPs shelf. Click the Download Model button. Optionally change the Model Name parameter to a custom model, or just leave as is and hit Download to work with the default Stable Diffusion Model.
7. In the MLOPs nodes, use the dropdown on the [type] Model parameters to select a downloaded model to use. You can also provide a repo name from the Huggingface Library, and the nodes will download it for you. For example runwayml/stable-diffusion-v1-5.

Downloading Models

• By default, $MLOPS_SD_MODEL is the path to a SINGLE model used by all Stable Diffusion nodes by default. You can set this to be your preferred default model.
• By default, the plugin will cache all downloaded models to the folder specified by $MLOPS_MODELS. (Notice the S at the end) This will make them show up in the dropdowns for the model paths on the nodes. Both of the above varibles can be changed in the MLOPS.json to suit your preference.

Troubleshooting

• If you get an error saying "Torch not compiled with CUDA enabled". Uninstall pytorch in your system python, restart your PC and hit the Install Dependencies shelf button again.
• Metal users should set the compute device on MLOPs nodes to "MPS", and set the following environment variable for it to work: PYTORCH_ENABLE_MPS_FALLBACK=1
• If you get an error with "Could not load library cudnn_cnn_infer64_8.dll" and you have Octane installed as a plugin for Houdini, try disabling it and restart Houdini.
• If you get an error similar to: "Unexpected self.size(-1) must be divisible by 4 to view Byte as Float (different element sizes), but got 2683502, <class 'RuntimeError'>". Try deleting the model cache in $MLOPS_MODELS/cache/[your model] and try again. This is likely caused by a corrupt cache.
• Other plugins we know cause issues installing MLOPS dependencies: Renderman, Octane. Disable these while installing MLOPs and its dependencies. After installing MLOPs and its dependencies you can re-enable them.
• If you get strange Python errors and you have tried several things already, make sure you dont have a conflicting PYTHONPATH environment variable set. If that is the case remove it and restart Houdini (And the Launcher if you use it)

Notes

• We have provided a basic example file in this repo. You can find it in the hip/ folder.
• This plugin installs quite a few dependencies. You can find them in requirements.txt.
• Digital Assets (HDAs) are stored and distributed in the expanded format. You can use hotl to collapse them if need be.