Model Card for TempoFunk

<!-- Provide a quick summary of what the model is/does. [Optional] --> A community produced Text-To-Video model using Temporal Attention

Table of Contents

Model Details

Model Description

<!-- Provide a longer summary of what this model is/does. --> A community produced Text-To-Video model using Temporal Attention

Uses

<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->

The TempoFunk model is meant to be used as a Video Production Program.

Direct Use

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. --> <!-- If the user enters content, print that. If not, but they enter a task in the list, use that. If neither, say "more info needed." -->

Produce Generative Video

Downstream Use [Optional]

<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app --> <!-- If the user enters content, print that. If not, but they enter a task in the list, use that. If neither, say "more info needed." -->

Meme production Visualization Personalized Text-To-Video

Out-of-Scope Use

<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. --> <!-- If the user enters content, print that. If not, but they enter a task in the list, use that. If neither, say "more info needed." -->

Produce Disinformation Produce Gore

Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->

During usage of TempoFunk, it may generate obscene or otherwise unpleasant to look imagery. This is because of both the VAE and the low amount of samples seen by the TempoFunk model. Video generated by TempoFunk may be uncanny.

Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

Use superres or other methods to clean up visuals before publishing or using.

Training Details

Training Data

<!-- This should link to a Data Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

TempoFunk was trained on movement data from dancing videos. These dancing videos were scrapped and encoded into Stable Diffusion Vae Latents. More information forthcoming.

<!-- ## Training Procedure -->

<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->

Results

[https://huggingface.co/spaces/TempoFunk/makeavid-sd-jax]

Environmental Impact

<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

Technical Specifications [optional]

Model Architecture and Objective

The temporal layers are a port of Make-A-Video PyTorch to FLAX. The convolution is pseudo 3D and seperately convolves accross the spatial dimension in 2D and over the temporal dimension in 1D. Temporal attention is purely self attention and also separately attends to time.

Only the new temporal layers have been fine tuned on a dataset of videos themed around dance. The model has been trained for 80 epochs on a dataset of 18,000 Videos with 120 frames each, randomly selecting a 24 frame range from each sample.

Compute Infrastructure

TPU_V4

Hardware

TPU_V4

Software

Google JAX Google FLAX

Model Card Authors [optional]

<!-- This section provides another layer of transparency and accountability. Whose views is this model card representing? How many voices were included in its construction? Etc. --> Lopho, Chavez, Davut Emre, Julian Herrera

How to Get Started with the Model

Use the space below to get started! [https://huggingface.co/spaces/TempoFunk/makeavid-sd-jax]