凤希AI - 分镜数据解析V2

FxaiStoryBoardLoadV2 ("凤希AI - 分镜数据解析V2") is the second revision of the storyboard loader, and unlike the storyboard generator's fake V2, this one is a real change. It does the same core job - take a storyboard LIST and a row index, return the assembled prompt - but it trades away two inputs and swaps the third output from scene props to character assets. It's the version you want when your story is character-driven rather than prop-driven.

What changed vs. V1

V1 (FxaiStoryBoardLoad) took four required inputs including 循环复用 and 默认场景序号, and output 场景道具 (scene props). V2 drops both of those extra inputs - the required set is just 分镜数据 and 行索引 - and its third output is 角色资源 (character assets) instead of props.

So the "V2" here genuinely is a slimmer, more specialized node:

• Same prompt assembly: 通用提示词 + shot["提示词"] + 尾部通用提示词, with the same two optional STRING globals.
• Same out-of-range behavior: empty storyboard or bad index returns just the globals and an empty character string, no crash.
• But it reads 角色资源 from each shot's dict, and hands it back raw - no comma-normalization, no fallback. That's the point: if a shot lists the characters on screen, this node surfaces them for a character-asset node downstream.

Outputs:

• 提示词 - STRING, assembled prompt for this shot.
• 行索引 - INT, echoed back.
• 角色资源 - STRING, the shot's character references (empty string if the shot has none).

When to reach for it

Character consistency is the eternal problem in long AI video - the pack's README sells "人物一致性" hard, and this node is the plumbing half of that pitch. If your storyboard rows carry character references per shot, V2 hands them straight to the character asset machinery without the prop-conversion layer V1 adds. If your story is about props and scene elements instead, V1's 场景道具 path is still your friend.

The caveat

Because there's no 循环复用 here, per-shot looping falls to your loop node - wrap the 行索引 with a modulo yourself if you want a recurring shot, or use V1 for that. And remember the silent-fallback behavior: an out-of-range 行索引 won't error, it'll just give you a prompt with no character data. In a loop, keep that index clamped (FxAiStartEndIndex) so you never silently render character-less frames.

Install

Same pack, same command:

cd ComfyUI/custom_nodes
git clone https://github.com/fxai666/fxai-toolkit

Restart ComfyUI, or use ComfyUI Manager and search "fxai-toolkit". Pure Python dict access, no dependencies beyond the pack's normal load.