Is multi-reference / R2V generation supported in the open-source inference code?

[Falkonar]

Hi, thank you for releasing LingBot-Video.

I noticed that the project website shows a demo where multiple reference images appear to be used for a single video generation prompt. For example, in the “Robot and human jogging together” demo, the UI shows three reference images above the text prompt.

However, in the current open-source repository and inference documentation I could only find the following modes:

• t2i
• t2v
• ti2v

The local CLI also seems to expose only a single image input through --image, which appears to work as a first-frame condition for ti2v.

My questions are:

1. Is the multi-reference workflow shown on the website currently supported in the open-source inference code?
2. If yes, what is the correct CLI/API usage for passing multiple reference images?
3. Is there a specific prompt_json schema for multi-reference character/object conditioning?
4. If this feature is not currently released, is it planned for a future update?

I am especially interested in the use case where several reference images define different subjects/characters, and the model generates a video that preserves those identities while following a text instruction.

Example from the website:
Robot and human jogging together

Thanks.