Blue Iris Version 6.0.1.24 Now Supports Custom Models

[jrbeddow]

I assume none of the older ipcam models that were used with CPAI will work with the BIv6 built in AI.

Hmm...are you only referring to the truly "old" versions (that go back 2-3+ years), or have you been testing the "new" .onyx versions published roughly one week ago by @MikeLud1 (see post #1 Github link above)? The built-in BI AI only supports ONYX versions (with some specific version limitations as well, for now).

Edit: While I was writing my response above, it seems that @MikeLud1 has clarified the situation.

 

[Tinman]

They should be the same models I just renamed the files

weird, I downloaded the V5 from your github and they worked? Just for fun I renamed one of my saved model files and noticed it was not the same size. I then tested the renamed file and sure enough, it did not work.

 

[Tinman]

Hmm...are you only referring to the truly "old" versions (that go back 2-3+ years), or have you been testing the "new" .onyx versions published roughly one week ago by @MikeLud1 (see post #1 Github link above)? The built-in BI AI only supports ONYX versions (with some specific version limitations as well, for now).

Edit: While I was writing my response above, it seems that @MikeLud1 has clarified the situation.

yes I did. I did not see any ipcam-combined files on that link, but I had them saved, but they were the newer .onnx models.
This is what I used:

 

[Tinman]

Well, it's working great now. I'll just dump the Ipcam models for the newer renamed models.

 

[jrbeddow]

Well, it's working great now. I'll just dump the Ipcam models for the newer renamed models.

View attachment 236906 View attachment 236907

Thanks, those are impressive analysis times: which NVIDIA card is that? I'm making the assumption that no onboard (motherboard) GPU could achieve those times.

 

[Tinman]

Thanks, those are impressive analysis times: which NVIDIA card is that? I'm making the assumption that no onboard (motherboard) GPU could achieve those times.

NVIDIA RTX A2000 12GB

 

[Vettester]

@MikeLud1

Hey Mike, according to the release notes for 6.0.1.25 Ken has switched out the models used for plate detection and plate OCR. Are these the same models you previously recommended? If not, have you done a comparison of the two?

 

[MikeLud1]

@MikeLud1

Hey Mike, according to the release notes for 6.0.1.25 Ken has switched out the models used for plate detection and plate OCR. Are these the same models you previously recommended? If not, have you done a comparison of the two?

Let me check.

 

[MikeLud1]

Are these the same models you previously recommended?

The new models are the best combination.

The plate model is mine and it is the one in this post

Thread 'Blue Iris Version 6 License Plate Custom Model'

Dec 29, 2025

After using the built-in ALPR I found the model used to detect the license plate did not perform well at night. I trained a new plates.onnx model that performs well now during the day and at night. Below is the link to the new model. To use the new model save the model in folder C:\BlueIris\AI\models. Then after saving the model restart Blue Iris service.

Recommended min confidence setting
Day 80%
Night 85%

https://github.com/MikeLud/Blue-Iris-Custom-AI-Models/raw/refs/heads/main/Custom-YOLOv8-11/plates.onnx

• MikeLud1
• Replies: 20
• Forum: What's New

• 

The OCR model is the one that is discussed in this post.

Thread 'Fine Tune BI6.0 AI OCR'

Jan 16, 2026

I noticed that BI6 uses the global_mobile_vit_v2_ocr.onnx...... for LPR OCR. Is there a way to fine tune that using the previous model?
global_mobile_vit_v2_ocr Model Config · Issue #75 · ankandrew/fast-plate-ocr Seems like they don't support the old format and only the cct ones. Any guide or help would be appreciated

• adventuredphoto
• Replies: 13
• Forum: AI

 

[Vettester]

Thanks Mike!

 

[killerdan56]

since when was there a PACKAGE model. is that new for v6 or was the file available for use before with cpai
?

 

[MikeLud1]

since when was there a PACKAGE model. is that new for v6 or was the file available for use before with cpai
?

It was just added in the latest release, you need to download the models again from the main AI settings

 

[CrazyAsYou]

Downloading from Blue-Iris-Custom-AI-Models/Custom-YOLOv5 at main · MikeLud/Blue-Iris-Custom-AI-Models Is bi-general-v5.onnx the one with detection only for "person" and "car" like ipcam-general.pt was for CPAI?

 

[Tinman]

Just for fun I tested bi-command-v5 model against yolov10m on my demo running the onboard UHD Graphics 770. Little over twice as fast. 6.0.1.25



Kudos to Ken and Mike for making this so easy to run AI compared to the early days of CPAI!

 

[MikeLud1]

Just for fun I tested bi-command-v5 model against yolov10m on my demo running the onboard UHD Graphics 770. Little over twice as fast. 6.0.1.25

View attachment 236989View attachment 236990

Kudos to Ken and Mike for making this so easy to run AI compared to the early days of CPAI!

To be fair try comparing my model against yolov11s because my model is trained using the small model.

 

[Tinman]

To be fair try comparing my model against yolov11s because my model is trained using the small model.

Sure, where do I find that model? I see the included 10s model.

 

[MikeLud1]

Sure, where do I find that model? I see the included 10s model.

Use the yolo8s.onnx model that came with Blue Iris model download, it should be about the same.

 

[Tinman]

Here is a comparison of models using the onboard Intel UHD 770. bi-combined-v5 to yolov8s.

 

[The_Penguin]

Seems fine now so far with BI 6.0.1.27 and the ipcam-animal that appears to have downloaded with the "download models" link inside BI.

 

[sumps]

I've been running Blue Onyx for AI and using it's built in IPcam-Combined.onnx model.. Updated to 6.0.1.27 and switched over to using the built in AI and installed the new Bi-combined-v5.onnx and ran a clip to through each .. I repeated the same clip several times and got the same results. Blue onyx had slightly slower analysis times, but was able to pickup a person along with car in the clip.. Using BI built in AI I was unable to pick up the person, it only picked up the car.
I copied the IPcam-Combined.onnx model from Blue Onyx to the Blue Iris model directory and switched over to that model and still was unable to pick up the person, only the vehicle .. So why in this situation is Blue onyx more accurate than the built in AI when using the same model? what am I missing?