Feedback on Video Generation and ComfyUI

Dear SeaArt Team, as a new user, I find your platform's integration of advanced tools like ComfyUI very compelling. Your use of the Hailuo engine is a key feature; while it delivers impressive photorealism, its known issues with prompt fidelity for non-everyday scenes present a significant challenge, often resulting in an accumulation of wasted credits on failed generations.

My experience so far suggests that the current API integration with Hailuo might be limited, and the ComfyUI environment for video may have stability issues. I came to this conclusion after attempting to build a custom ControlNet workflow to improve prompt fidelity. During this process, I made two key observations:

1.     Certain generation artifacts, like incorrect character scaling, humans glitching through animals or furniture, etc., seemed more frequent in SeaArt than on Hailuo's native platform.

2.     More critically, key nodes within the provided ComfyUI video templates (e.g., model selection dropdowns in the SVD workflow) appeared unresponsive, which made it laborious and ultimately prohibitive to debug or build a working solution.

A "quick win" for all involved would be a more robust API connection and a stabilized ComfyUI backend for video. This would allow creators to mitigate Hailuo's prompt fidelity limitations while leveraging the true power of your platform.