Here's the 100% AI image I ended up w/as my starting point:
"Adorable anime robot dog wearing battle armor, proudly standing in an action pose, line art, dynamic pose, hyper details"
Discussion
Used that as an input first, plus a pic of real Kuma second:
Best result from that combo was:
(I had to specify "--no couch, blanket, pillows" to mostly ignore them from Kuma's pic)
And then as I said, that combo result as the input plus the original Kuma pic again got me to my final result.
So you can trial-and-error your way to forcing Midjourney to keep referencing irl Milou in one or many actual pics that you specify as part of the input.