I’ve been testing WAN 2.1 and WAN 2.2 for a while now, and if you’re playing around with image-to-video animations, especially using LoRAs, here’s something worth trying.
WAN 2.1 has a huge LoRA base. Most LoRAs out there were made for it, and they generally apply well. The model also does a decent job at keeping things consistent, especially when you're pushing for stylized effects or specific movements.
WAN 2.2 is a different beast. It’s cleaner, sharper, and better at preserving the original image—especially fine details. It’s split into two versions: high denoise (which acts more like a base model) and low denoise (which works more like a refiner). The downside? A lot of WAN 2.1 LoRAs don’t transfer well to WAN 2.2. Some barely apply, others just add unwanted noise or even erase small details from the original image.
My Solution
Use WAN 2.1 as your base, like you would use WAN 2.2 high-denoise. Let it apply the LoRA effect and drive the animation. Then run the result through WAN 2.2 low-denoise to clean things up and recover fine details.
This hybrid setup gives you:
The LoRA effect and motion from WAN 2.1
The detail and structure from WAN 2.2
A result that stays true to your original image, while still feeling stylized and alive
TL;DR
If you want your LoRA effect to show while keeping the subject clean and sharp, mix WAN 2.1 + WAN 2.2 low-denoise.
Pure WAN 2.1 = decent but not as sharp.
Pure WAN 2.2 = sharp but may ignore older LoRAs.
The hybrid? Balanced and reliable, especially in things that the original model was not trained to do, if you know what I mean.