training a lora model VS using motion control grayscale depth reference video
A user on Reddit is seeking advice on improving the consistency of AI-generated product photography videos using less expensive models. They are comparing two potential methods: training a LoRA model or using more accurate grayscale depth map reference videos. The goal is to achieve better object consistency, especially during object rotations, which current cheaper models like Seedance 1.5, Grok Imagine, or WAN 2.2 fail to do effectively. AI
IMPACT Users are exploring methods to improve AI video generation consistency with more accessible models.