Fails to follow prompts to generate complex sequential actions

Hello lvmin,

First, thank you for developing this fantastic tool! I’m really excited about the possibilities it offers for video generation.

I’ve been experimenting with prompts that describe multiple actions in sequence, but I’ve noticed that the generated video tends to ignore everything after the first action and simply loops that initial movement. I was hoping the model would perform each step in order, but it seems to fall back to repetition.

Prompt Used:

The woman dressed in green took a sip of water, then stood up and picked up the picture frame on the table and looked at it, leaving tears in her eyes.

Observed Output:

• The video only shows the woman taking a sip of water on repeat, without ever standing up or picking up the frame.

So Is there a recommended way to structure multi-step prompts so that FramePack follows each action in sequence? Or are there any known limitations or upcoming features planned to improve narrative coherence across frames?

Thank you in advance for your time and assistance. I appreciate any guidance you can provide to help me achieve more accurate, sequential action in my generated videos!

The issue names no files, tests, or entry points. Start by reproducing the supplied prompt and reviewing the project’s prompt and video-generation paths to determine whether sequential actions are supported. Done would require a clearly defined fix or documented limitation, but the issue currently asks for guidance and future plans rather than a scoped change.