BRING YOUR OWN MODEL
Use Any YOLO Model
NobleAIM runs Noble’s own models, models you train yourself and models you find online. If it’s a YOLO model saved as an ONNX file, NobleAIM can load it.
What NobleAIM can load
NobleAIM reads the YOLO family of models, the same family Noble’s own models are built on.
YOLOv5 to YOLO26
Models from YOLOv5, YOLOv7, YOLOv8, YOLO11 and YOLO26, including detect, pose and segment models.
ONNX or TensorRT
An .onnx file loads as it is. NobleAIM also loads TensorRT .engine files, but build those on your own PC rather than downloading them. A PyTorch .pt file needs converting first.
A square input
Export at a square size such as 640 by 640, which is what YOLO uses by default. Note the size, as NobleAIM asks for it when you add the model.
Load a model in four steps
Adding a model happens inside NobleAIM. You need an active NobleAIM licence, the same as for Noble’s models.
-
1
Get the file
Download an .onnx model, export one you have trained, or convert a .pt file (see below). In File Explorer, hold Shift, right-click it and choose Copy as path.
-
2
Add it
Click Manage Models, open the Custom Models tab and choose Add Custom Model. Paste the path, enter the input size it was exported at and click Add.
-
3
Activate it
Pick the model in the Model list on the same tab and click Activate. Scan Folder adds every model in a folder at once.
-
4
Set its classes
If the model has more than one class, open the Model tab. Put any class you don’t want aimed at, such as teammates, in Ignored Class IDs, and use Target Classes to set where it aims on the rest.
Once a model is added, the Build Accelerator tab can build a faster version of it for your graphics card: a TensorRT engine on NVIDIA, or a DirectML version on AMD and Intel. A model’s own page normally lists its classes, for example 0 for enemies and 1 for teammates.
Where to find models
There’s no single library of free game models, so here’s where to look, or how to make your own.
Hugging Face
Search Hugging Face for YOLO and the name of your game, and read each model’s page for its licence and classes. Many are shared as .pt files, so convert them first.
Roboflow Universe
Roboflow Universe has lots of labelled game datasets you can download and train a model on. Its trained models usually can’t be downloaded.
Train your own
Ultralytics Hub trains YOLO models on your own data and exports them as ONNX. Our developers page walks through it.
Convert a .pt file to ONNX
NobleAIM can’t load PyTorch .pt files. If the model was made with Ultralytics, convert it with Ultralytics’ own tool: install Python, run pip install ultralytics, then close NobleAIM and run:
yolo export model=best.pt format=onnx imgsz=640
Swap best.pt for your file’s name and 640 for the size the model was trained at. The .onnx file appears next to it. Opening a .pt file can run code on your PC, so only convert one from a source you trust.
Stay safe with downloaded models
Models from other people come with no checks from us, so a little care goes a long way.
Choose .onnx where you can
A .pt file can run code when it’s opened, so only convert one from a source you trust.
Run nothing else
A model is a single file. Never run an .exe, script or installer that comes with one, and scan downloads with your antivirus.
Use models you’re allowed to
Check each model’s licence. Leaked or pirated models, including copies of Noble’s, aren’t allowed under our EULA.
We don’t check them
Noble doesn’t test or vouch for other people’s models, and our support covers NobleAIM itself. Section 7.4 of our EULA has the full terms.
Results depend on the model
A model trained on PC footage may track differently on a console capture feed, and results vary with the model, your settings and your capture setup.
Noble’s own models are trained for each supported game, and Import model settings sets up their classes for you. Every NobleAIM plan includes our library of base models, covering Apex, Valorant, Warzone, Fortnite, Rainbow Six, Marvel Rivals and more, in the Noble Marketplace.
Using your own models: FAQ
What people ask most often before loading a model of their own.
-
Can I use a YOLO model I found online?
Yes. If it’s a YOLO model saved as an .onnx file, add it with Add Custom Model. You need the right to use it, and you use it at your own risk, as Noble doesn’t check models from other people. -
Can I use a YOLO model made for Apex, Fortnite or Valorant?
Yes, as long as it’s a YOLO model in ONNX format. Models made from PC footage may track differently on a console capture feed, so try one before you rely on it. -
Why won’t my model load?
Usually it’s a .pt file that needs converting, an input that isn’t square, or a model that isn’t YOLO. NobleAIM tells you why it refused a file. -
Do I need an NVIDIA graphics card?
No. Any modern dedicated graphics card can run NobleAIM, though NVIDIA cards give the best experience. Build Accelerator builds a faster version of your model for whichever card you have: a TensorRT engine on NVIDIA, or a DirectML version on AMD and Intel. -
Do I need a NobleAIM licence to use my own model?
Yes. Your models run inside NobleAIM, so you need an active NobleAIM licence, the same as for Noble’s models. -
Who owns a model I bring?
Not Noble. A model you train yourself, or get from someone else, stays under its owner’s rights and licence, even after you load or convert it in NobleAIM. The one exception is a copy of one of Noble’s own models.
Bring your model to NobleAIM
Every NobleAIM plan includes the app, our base model library and Custom Models for loading your own. Questions about a model? Ask in our Discord.