Omni Animation - AnimateGPT (text to animation) has been released!

​

We're excited to announce our latest development – a cutting-edge AI generation technique capable of producing high fidelity 3D humanoid animations. Omni animation uses advanced algorithms and deep learning to generate animations with realistic movement, detail, and depth.

Interested in learning more about our new service? Sign up at OmniAnimation.ai to receive updates, get exclusive discounts, and to be the first to know when we launch and change the way animations are created in games, film, and architectural visualization.

 

We launched the OmniAnimation.ai last week without much description and by now I'm sure you're wondering what exactly is Omni Animation.

Omni Animation is a deep learning network capable of producing high fidelity animations. It is designed to speed up your animation workflow and as a result, save you a bunch of time in creating new animations. Omni Animation is perfect for personalized hero characters, NPCs, crowds, and motion matching.

Omni Animation does all of this by using an deep learning autoregressive model. This model produces an output based on the previous state. That output is then fed in as input to the next time step. When this process is repeated it produces a set of frames over a period of time, resulting in the generated animation. One of the things that has taken a very long time to research is how to do this without having any artifacts. The animations produced will be production quality animations without any necessary cleanup.

We understand that you may have some questions about how our service works. We have a series of articles lined up to guide you through the features of Omni Animation. Among other topics you'll learn about:

• Our process of converting raw mocap animations to generated animations.
• The bulk downloads feature designed for NPCs and motion matching.
• The IK features that allow for fine-grain adjustments.
• How styles can completely change the look of your animation.

At launch, we will have both AI generated and non-AI generated animations for you to choose from. We will be regularly releasing new animations as we continue to grow.

 

One of our original goals of Omni Animation has been to produce high fidelity animations that can be used in your project without any cleanup. We started developing Omni Animation in late 2020 and progress throughout this time has not been linear. There are some weeks when work is being done but the results are not tangible. There are other weeks when it seems like we are moving at light speed.

We thought it would be interesting to show an example of the type of results that we see throughout development. Taking a look at the walk cycle, some of our very first AI generated walk animations came out looking like this…

We’d inadvertently developed a spider animation! After some more research the animation started to resemble an actual walking animation:

Later the walking animation became much smoother, albeit had some foot sliding issues:

This sliding was fixed with a post processing technique that ensures the character moves the correct distance based on the feet movement. The end result is a walking animation that you can use without any cleanup:

It’s neat to think that the entire animation was generated using AI. It’s also fun to look back and see the first crazy results.

We should have some more news soon on the features and animations that Omni Animation will have when we first release the beta. The website is getting close to being done and the machine learning developers are continuously improving the deep learning model.

 

Fantastic results! Is the AI able to generate animations at runtime or only in the editor?

 

Thanks!

Right now it's completely outside of Unity. The workflow is similar to other animation websites where you download the animation file (we'll support fbx and bvh initially) and then import into the program that you want to use it in. This means that it can work with Unity, Unreal, Godot, etc.

In Unity I did a proof of concept which allows users to define their animations based on a set of parameters and then import those generated animations using the addressables system at runtime. We won't have the API fleshed out for this when we first release but I can definitely see adding it later this year.

Runtime animation generation would also be very cool, but that's a completely different scope and we will want to get everything working well on the server first

 

wow that's fantastic! You guys should advertise it more! I can't wait to use it! When do you guys plan to release?

 

Cheers! We will be releasing the beta this summer - we have a booth at devcom so definitely need to be out by then If you haven't signed up for the newsletter make sure you do so as those who are signed up will be the first to have access to it.

I am working on the marketing material for the next couple of months but as we get closer to release there will be more and more about it. We're getting close!

 

The goal for Omni Animation from the very start has been to generate high fidelity animations. We want to generate animations that do not require any cleanup and can be generated quickly without many iterations in order to get it just right.

After a lot of thought we decided that a series of sliders and drop downs is the best approach in order to achieve these goals. These parameters allow for complete control over the generated result so you are not guessing what value should be used in order to generate a good result.

Lets take a look at an example. Below is a screenshot of the website with the run animation.



In this image you can see the generated animation in the scene view and the available parameters on the right pane. By default the character is a male character that runs at a decent speed with a normal step size. For our example lets say that we now have a character that is female who has a smaller step size and slightly less velocity with her run so it's more of a jog. Using the parameter pane we can make these adjustments and generate a new animation:

As you can see with these parameters you are guaranteed to get good results no matter what combination of values you use. Lets take this one step further and change the style of the animation. The default animation uses the Generic style but we want the animation to more reflect a zombie style. With the change of the drop down we can easily generate that.

The rest of the parameters are still used so in this situation we have a female jogging zombie with a small step size.

One advantage of the parameter approach is that it makes for a really intuitive API. We have a prototype of a Unity SDK where we can load in new animations at runtime. This opens up a lot of use cases and we'll show more in the next newsletter.

 

With the uncertainty around using generative AI in Steam games you won't have to worry about that with Omni Animation. We own all of the training data. The EULA makes it clear that you can use the animations in any way besides redistributing the raw file.

​

 

Here's a new video showing more of the zombie style:

 

When Omni Animation releases in beta this summer you will be able to download generated animations in fbx or bvh format. One of the fun things that we have been experimenting with is the idea of providing an API so you can load the generated animations dynamically.

A great use case for this is if you are developing an RPG with a character customization screen. In this scenario the player can customize their character, and based on the player’s customization the Omni Animation API can generate an animation that matches the character’s look.



With this setup the player doesn’t even have to specify the animation parameters. All of the parameters can be retrieved from the physical characteristics of the customized character. Lets take a look at our prototype.



In this prototype there are three different characters: a generic character, a zombie, and a ninja. By default the animations are standard animations that are not meant for any specific character type. With the API layer we could generate animations that match the characters:

The animations in this video were generated with Omni Animation but the API layer does not exist so they were instead pre-generated and retrieved from a server. This video was created with Unity and we used Asset Bundles to pull in the new set of animations which makes it really flexible. There aren’t any limitations to how your animation system is setup.

 

The Omni Animation limited beta will be released on August 14!

​

Omni Animation has been in development for almost three years and we’re excited to announce we’ve reached the point of being able to release a beta!

Our background is in game development tools where we primarily work with indies and small game studios. We’ve noticed that these groups are not always able to find high quality animations at an affordable price that fits their character style. That’s initially what inspired us into thinking there must be a way to generate animations using machine learning, and now with Omni Animation there is.

Limited Beta​

Omni Animation will release on August 14 to a limited number of registrations. We want to ensure all registered users have a seamless experience and also that everything scales appropriately. During the limited beta we’ll be monitoring the length of time it takes to generate an animation, any errors or poor results from the animation generation, as well as overall site performance.

A queue system will be used for granting access to the beta. We will be sending an email in early August to our mailing list that will allow you to be first to reserve your spot in line for the beta. If you are not signed up for the mailing list you can do so on OmniAnimation.ai. Once a space opens up you’ll get an email with a unique code to register. Bonus points – visit us at our booth at devcom this year and you’ll get a registration code that allows you to skip the queue! We’re hoping to open up to all registrations ASAP once we get past this initial beta phase.

Discord Server​

Omni Animation now has a dedicated discord server. This will be the primary discussion method for Omni Animation. You can find our discord server here.

 

On August 11th at 2pm GMT I will be giving a live preview of the Omni Animation website. I'll be streaming this preview at https://www.youtube.com/live/lwCBqADYMOs. I'll be watching the chat to answer any questions as well.

 

Omni Animation will be released this Monday, August 14. We have finalized our release schedule and wanted you to be the first to know about it.

• August 14 at 11am GMT: The website goes live and you can reserve a spot in the limited beta.
• August 14 at 2pm GMT: A live preview of the website will be shown on YouTube.
• August 14 at 4pm GMT: Emails will start to go out to the first set that reserved a spot in the limited beta.
• Week of August 14 and beyond: We will release more spots in the limited beta.

When the website goes live we will send you an email letting you know that you can now sign up. We are really looking forward to the release and seeing how you use Omni Animation!

 

The Limited Beta of Omni Animation has been released! Click here to sign up.

​

After you sign up for the beta you will be assigned a number within the queue. In order to ensure everything goes smoothly we will be limiting the number of sign ups initially. When your slot opens up we will send an email with a unique code that you can then use to register.

Omni Animation has been a ~3 year project and we're just getting started. Included within the initial release are 6 AI models to get you started with your locomotion animation. We also have more than 600 non-AI animations to help complete your animation set. These non-AI animations were created by professional actors and have the same high quality as the AI animations.

 

​

We are at devcom! If you're here in Cologne stop by our booth on Tuesday for a demo of Omni Animation and also a code for instant access to the beta. We are near the snacks.

 

This past week we showcased Omni Animation at devcom in Cologne, Germany. It was our first official event for Omni Animation and it was an awesome experience!

​

We learned so much from speaking with everyone that visited our booth and it's helped us figure out what to prioritize in the coming months. Here are some highlights that we took away from devcom:

Generation Credits

When you hit the "Generate" button it will deduct one credit even if you don't download the animation. While it makes sense because we have get charged for the CPU & GPU usage, it also leads to less experimentation because you don't want to "waste" the credit. We received a lot of feedback on this and are going to be changing it so credits will only be deducted when you download the animation. There will still be an internal generation count to prevent the case where someone is constantly generating animations without downloading anything, but it will be high enough that we don't expect anyone to notice with normal usage.

We'll let you know as soon as this change is in place but we expect it to be early September.

Rig Improvements

Omni Animation uses a humanoid rig that you can then use the bone remapping feature within any game engine or rendering program. We received feedback from several people that they would want to be able to download the animation with their own rig. While the end result will be the same, it will increase productivity because you won't have to go through the remapping steps.

This feature is something that we'll be working on this year. There are a lot of different rig types so it could take some time to implement but it's a priority for us.

Training Data

One of the major topics that came up related to the training data used by Omni Animation. While everyone is happy that there are no copyright issues since we own all of the training data and are giving you a license to use animations in your project, they also want to be able to upload their own data. This is a topic that we've always had on the roadmap, but after devcom we realize it should be implemented sooner rather than later. This topic is going to take a lot of research in order to do correctly. As soon as we have something to show we'll let you know.

Website Changes

While at devcom we also noticed certain questions kept coming up so we'll be redesigning parts of the website to highlight the answers to these. One question kept coming up is the fact that you can continue to use any downloaded animations even if you cancel your subscription. We'll also be showing the price per animation depending on the subscription tier and not just showing it based on number of credits.

While these are a few of the major topics, a lot of minor topics, such as increasing the animation library or potential partnerships, also came out of devcom. Overall it was a very positive experience and we hope to be back next year. Until then, we'll be releasing our first animation drop shortly so be on the lookout! And thanks again to everyone that stopped by our booth at devcom, it was great meeting you!

 

Animation Drop 1 is here! This drop contains 202 new animations and is focused on assault rifle, pistol, and life categories. This video shows a few highlighted animations:

We have also updated the website with the following major changes:

• Generating animations will not cost any credits. We have a check in place to prevent abuse, but in normal circumstances you will not hit this limit.
• Minor artifacts have been fixed with the AI animations.
• The animation viewer will change to a lower quality mode for less powerful machines.

With these changes we have also sent out 200 more invites. We should be able to go through the waiting list more quickly now so if you haven't received an invite code yet it will be coming soon.

We are hard at work for the October update. Later this month we will give a preview of the changes that are coming. There is one feature in particular that we think will be really useful for everyone.

 

The Omni Animation Limited Beta has ended! You can now sign up without needing a registration code.

Thank you to everyone who participated in the limited beta. During this limited period we corrected artifacts within the AI generation process, fixed website errors, and released new non-AI animations.

With the limited beta complete we are excited to open up Omni Animation to a larger audience. We are just getting started - next month's update we will be adding more animations and parameters for both AI and non-AI animations. As soon as we have something to show we'll be sure to let you know.

 

The second animation drop has been released! This drop contains 200 new animations and focused on the melee category. With this drop we now have 1,046 animations within the library. The video below contains some highlighted animations from this drop.

Next week we will be releasing new website features including new parameters on all animations. We'll have more details as soon as this feature set has been released.

Our next animation drop will focus on some community requested animations. If you'd like a particular animation let us know on discord.

 

The latest Omni Animation update has been released! This update includes some exciting changes:

New Parameters​

​

Our current focus is on actively expanding the number of parameters available for both AI and non-AI animations. This is the groundwork required before allowing you to upload your own animations. With this update we've added the following parameters:

• Animation Speed
• Mirror
• Frames

These parameters are available for both AI and non-AI animations. The next website update will contain more advanced parameters.

Subscription No Longer Required​

​

We listened carefully to your feedback and now you can purchase animations without needing an active subscription! Due to the resources required you will need an active subscription in order to modify the parameters or generate AI animations, but this is great for downloading animations straight away without any modification.

More Animations
​

Last month we released 200 additional animations focused on melee movements. This month we will be releasing another set of new animations including parkour, sword & shield, and dual pistol.< As always, we'd love to know what animations you'd like to see and appreciate your feedback!

 

@Tvtig was one of the first Omni Animation subscribers and he has started to create a new Unity tutorial series. For his first video he shows you how you can animate 3D character movement, and for the video he chose to use Omni Animation. Take a look at his video below:

 

why is he showing a mixamo screenshot for skinning (?)

 

some of the animations especially the hands do not work for different body types - being able to upload a character (such as what mixamo offers) is important to adjust for that - or to test if an animation works for the character.

is there a better way to do this than necessarily having to upload the character?

 

your upload character bone rigging needs to be more robust ... a character that worked fine when uploaded to mixamo triggers no root bone found in yours... i recommend doing more than a string search, look specfically for structure

https://omnianimation.ai/support/4

 

I think that he was just showing the different ways to download the animation.

Improved rigging is slated for December. We plan on implementing something like RigNet which will allow for a much more robust rig mapping. With this you will then also be able to download the animations in the rig that was uploaded so there's no need to remap the animation.

 

Coming soon.

 

We just released the landing page for AnimateGPT, our text to animation solution. On the AnimateGPT page you can sign up to be the first invited when the service is released in December!

 

The fourth animation drop has been released! There are 203 new animations in this drop in the agriculture, conversation, pistol, and reaction categories. This drop contains some user requested animations in the agriculture and conversation categories.

 

Hi there, really interesting product, the animation quality looks really great. I've just been trying to evalutate your product though and was left fairly confused about the AI side of things, the marketing material is not translating into a clear understanding, so please excuse my frankness and take this as feedback about someone coming in totally cold.

You've mentioned the service is an AI driven solution, I get you're training some AI models but it's not really clear how the AI models are applied either by youself, a user or what they're needed for. It's clear you mocap record and clean up animations, AI does 'something' and you're reguarly expanding your library of configurable animations which is great. Mixamo has had configurable animations for a very long time, search for an animation, configure various sliders to achieve the speed or style you want, export something unique. I recognise the quality and consistency of your animations look a lot better though, and the quality and kinds of configurable settings in mixamo can be wildly inconsistent animation to animation, plus it's nice to see a fresh animation library that's not 10+ years old. But I'm just unsure how "AI" is a game changer for your product when you compare the actual end result for users and I'm hoping you can fill in the blanks for me so I can understand why your product is better with the use of AI. The current descriptions are a bit ambigious... after animation mocap recording and cleaning: The machine learning team then develops a deep learning model using the recorded data as training data. OK, so they make a trained model and... After the model has been polished it is deployed to the web. The process is then repeated for other animations... and then the model does?

What am I missing here? is it the nature of 'how' you're applying the configurable settings to a base animation that makes use of AI? is this something that's only really benefiting how your tool works at your end on how you edit animations before an end user gets accessed to them? you mention you have 6 models, what does that mean? if you add more AI models, how will that change things for users?

The animation gpt sounds interesting, but is this just going to be a text search matching you with an animation from the library? do you have plans to mix between totally different animations to create animation sequences, instead of just smaller configurable changes to a single animation? eg you tell gpt you want an animation where the character: run at full speed, come to a sliding stop, open a treasure chest, jump back and look surprised.

Basically I'm just trying to cut through the marketing spin and get a clear idea how this all works, and what actually makes this better than something else, thanks.

 

Thanks for the questions, @jason_yak!

We have six purely AI generated animations, such as walk and run. These animations use the parameters to generate the animations frame by frame, allowing you to get the precise result that you're looking for. These animations use an autoregressive AI model to create an animation frame by frame. We only have six of these pure AI generated animations right now but the plan is to keep expanding into an entire locomotion set.

From there we will be releasing AnimateGPT, which you mentioned. AnimateGPT is more than using AI to pick animations from a dataset - it does so by blending the dataset to try to create the animation that you are searching for. This will allow it to create animations that are not directly found in the dataset. Now, this generation capability is limited by the training data. The results a pretty good if it knows what it is. It gets more and more powerful as this dataset size is increased.

Both of those topics relate to the animation library that you see us regularly expanding. These animations that are listed within the library will be used to increase the size of the training dataset. While we are doing this we are also adding both AI and non-AI parameters. Some of these overlap with the parameters offered by Mixamo, but we have a list of 15-20 parameters that we'd like to add.

One of the things that we will be focusing on next is allowing you to upload your own animation and use the parameters with your own animation. Eventually we'll be able to do some really cool things such as use AI to transfer styles onto your own animations. This way if you are an animator you are able to use Omni to assist with your workflow rather than it being purely generation based.

Does that clear things up?

 

Thanks for the info. Yes that helps to get a clearer idea about how you’re making use of AI, thanks. I can definitely see a great roadmap taking shape, lot of potential and I really like the sound of being able to upload our own animations to apply parameters/styles.

 

Awesome. Yeah the uploading your own animations feature is going to be extremely useful. In all likelihood we are still few months away from being to that point but we're getting there. If you haven't done so already you can sign up for our newsletter on the homepage and we'll notify you when there any new major updates.

 

Animation Drop 5 has been released! This drop contains new animations in the agriculture, crafting, pistol, and parkour categories bringing the total to 1598 animations in the library.

Next week we will be releasing AnimateGPT, our text to animation solution. I'll have more info on that shortly

 

Text to Animation is here!​

AnimateGPT, our text to animation solution, has been released! With AnimateGPT you can generate new animations using an AI model that was trained from more than 10,000 animation clips.

AnimateGPT is currently in beta and is in continuous development. The primary purpose for AnimateGPT is to be used as a prototype tool in your animation creation process. The quality of the results generated by AnimateGPT is heavily dependent on whether the prompt is included in the training data. If a topic is not included in the training data the model will try to output a result according to what it knows.

We are looking forward to seeing how everyone uses AnimateGPT as we continue to develop it. Join our discord to discuss AnimateGPT and share your results!

 

AnimateGPT has been released for a week now and there have been some interesting prompts. All total there have been more than 625 prompt generations. We are looking to see how people are using AnimateGPT and are working on improving some of the results generated.

 

On the Omni Animation website we will be soon introducing packs. Packs are exactly what you think - they are a collection of animations. Before we release packs on the Omni Animation website we will be releasing a select few packs on the Asset Store.

The first pack that we are releasing is the Omni Animation - Pistol Pack:
https://assetstore.unity.com/packages/3d/animations/omni-animation-pistol-pack-276060

The Pistol Pack is already integrated with our character controller allowing you to quickly replace the default animations.

 

The sixth animation drop has been released! This drop contains 162 animations focused on pistol, knife, and magic animations. There are now 1,760 animations within the library.

 

Continuing with packs we have released the Omni Animation - Knife Pack! You can see it on the Asset Store at: https://assetstore.unity.com/packages/3d/animations/omni-animation-knife-pack-277864