WarmBloodAban

/

Singularity-LTX-2.3_OmniCine_V1

Like 110

i find that it tends to fight with other LoRA's. There are times where four legs are produced when movement is involved and the movement is wonky as well. Too many glitches in it for my liking.

Amazing work already! The motion quality is super good, and I can run this without any supporting motion enhancement LORA. I'm currently testing with animations/anime style. In order to have the good motion quality, I need to run it at higher strengths (0.7-0.9), but at those strengths it does override some of the character look and quality, and occasionally produce disformation of bodyparts. I'm still testing with various combinations and samplers so it's early days. So far I have simply used:

• Str 0.7-0.9
• Euler A
• Eros10 model (due to better i2v prompt and motion)
• distilled 1.1 lora (the one recommended for Eros10 workflows)
• Loras: Crisp enhance details, ID lora talkvid, VBVR
  I've been playing with the strengths, but the main thing I notice is, having issues when I want the best motion quality from Omnicine I have to have high strength and it does screw up the look of the character most of the time

Anyway, I will keep testing. Good job so far! The motion quality is AMAZING! Once the rest is fixed and proper LORA stack and setting figured out, this will be a game changer

Hi,

Thank you for the detailed testing and for your kind words – I really appreciate it! I'm thrilled that you're enjoying the motion quality and that it works well even without extra motion enhancement LoRAs.

Regarding the issues you mentioned at higher strengths (0.7–0.9): I do notice that the character look and body parts can sometimes degrade. That's definitely on my radar. Right now, I'm still in the data collection phase for the next round of training, and at the same time I'm closely watching the progress of open‑source video generation models. If better techniques or architectures emerge, I'll gladly invest more time into exploring them.

Thanks again for your thorough feedback – it's still early days, but your insights are incredibly helpful.

Best regards