Fix s2v shot: 1080p & multi-person switch mask error

[huochaitiantang]

No description provided.

[gemini-code-assist]

Code Review

This pull request updates rs2v_infer.py to set rs2v.input_info before loading mask latents, preventing errors during repeated generation. It also adds the fixed_area field to SekoTalkInputs in input_info.py. The reviewer recommends also adding the fixed_area field to the RS2VInputInfo dataclass to ensure full compatibility and consistency, as the rs2v task relies on this parameter.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

While adding fixed_area to SekoTalkInputs is correct, the RS2VInputInfo dataclass (defined around line 137) is still missing the fixed_area field, unlike S2VInputInfo (defined around line 106).

Since the rs2v task now relies on fixed_area for mask resizing (as noted in rs2v_infer.py), any workflow that instantiates or parses RS2VInputInfo directly will lose or fail to support the fixed_area parameter.

Please add fixed_area: str = field(default_factory=str) to RS2VInputInfo as well to ensure full compatibility and consistency:

@dataclass
class RS2VInputInfo:
    ...
    # shape related
    resize_mode: str = field(default_factory=str)
    fixed_area: str = field(default_factory=str)  # <--- Add this field
    original_shape: list = field(default_factory=list)
    ...