loras compatibility
is the model actually compatible with other loras based on anima base? i found most loras cant perform well and are poor-consistent when applying to this model,and mostly important the checkpoint changed the embedded style of character loras,to a style i think more inclined to cartoon or anime screenshot.
Yes, LoRAs made for the original base will load and run on this checkpoint without errors — the architecture and parameter layout are identical. But loading successfully is not the same as working well.
The important part is this: this checkpoint went through heavy post-training, and that has pushed the weights a long way from the original base. A LoRA is really just a set of learned directions relative to a specific point in weight space — the point its training started from. Apply those same directions from a model that is now far away from that point, and they simply don't point where they used to. That's exactly why most LoRAs feel weak, inconsistent, and lose character identity here: the model's internal representations have shifted, so the LoRA's edits no longer land the same way.
The style change you noticed — character LoRAs now reading flatter, more cartoon / anime-screenshot-like — is the same effect seen from the other side. This model's baseline aesthetic prior has moved. Since a character LoRA carries not just identity but also the style of whatever it was trained on, that embedded style gets re-interpreted through the new prior, and it comes out looking different from what you'd get on the original base.
So this isn't really a "compatibility" issue you can tune away — it's a weight-space distance issue. LoRAs trained on this model, or at least calibrated against it, will behave far better, because they're learned relative to where this model actually sits. With original-base LoRAs you can sometimes get usable results by lowering the LoRA weight and adjusting prompts, but expecting the same behavior as on the original base will keep leading to disappointment.