Running new image and video tests! Quite happy with the results so far. vid2vid models are getting increasingly accurate at applying & replicating motion => TikTok / Reels or basically any shortform production can be fully automated! Cause people keep asking: Currently working on new workflows with Nano Banana Pro, Kling and custom checkpoints on Cloud GPUs (SDXL based checkpoints and WAN LORAs). So it's a combo of public APIs and self-hosted models. I'm not using ComfyUI, all my workflows are fully automated with custom suite (mostly Python backends) to scale faster.