凤希AI - 场景数据加载器V2

FxAiSceneLoadV2 is the audio-aware upgrade of the pack's per-scene loader, built to match FxAiSceneManagerV2. Where V1 returned a scene's duration, this one returns the scene's audio start and audio duration as separate floats - the two numbers you need when your soundtrack doesn't line up with scene boundaries at second zero. It's the node that hands a loop one complete, audio-timed scene at a time.

The mechanism is a dict lookup with loop-friendly defaults. It takes 场景数据 (LIST, forced from the manager), 行索引 (INT), and 循环复用 (loop reuse, default -1 - and this is the big behavioral change from V1: -1 means "wrap the index to the total scene count," so the loop cycles automatically). It reads 提示词文本 (wrapped in the optional 通用提示词/尾部通用提示词), 音频开始, 音频时长, 音频索引, 图片索引, 尾帧位置, and 转场. The default fallback is worth knowing: any failure - empty list, out-of-range index - returns (行索引, 通用提示词+尾部通用提示词, 行索引, 0.0, 15.0, 行索引, -1, True) and prints a cheerful "[凤希AI场景] 已加载默认值" message to the console. So an out-of-range scene silently becomes a 15-second empty one rather than a crash.

The eight outputs are the full scene schema: 场景行索引 (INT), 提示词 (STRING), 音频文件索引 (INT), 音频开始(秒) (FLOAT), 音频时长(秒) (FLOAT), 图片索引 (INT), 尾帧位置 (INT), 启用转场 (BOOLEAN). Note there's no 时长 output - in V2 the audio duration is the scene duration, which is the whole point: your segment length is derived from the audio, keeping picture and sound in lockstep across a long render.

The one trap is version mismatch. The V2 loader reads V2-manager fields (音频开始, 音频时长) that simply don't exist in V1-manager output - if you feed it a V1 scene list, every scene falls into the default branch and you get 15-second silent segments with prompts that are nothing but the globals. Match the versions: V2 manager → V2 load, V1 manager → V1 load. And remember the default wrap behavior means a broken loop runs forever instead of stopping - the pack assumes you want the loop to keep cycling, which is usually right, but keep an eye on the console message when debugging.

It's under 凤希AI/场景管理, part of fxai-toolkit (凤希AI / fxai.site):

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

or ComfyUI Manager → search "fxai-toolkit", then restart. If your long video is audio-driven, this is the loader to build the loop around.