凤希AI - 分段场景 - 时间轴提示词管理器

FxAiMultiPromptEditor is the authoring end of the pack's timeline prompt system. Long videos rarely want one prompt for the whole clip - you want "this for the first two seconds, that for the next five, something else after." This node is where you lay those time-stamped prompts out. It parses the JSON you paste in and normalizes it into a clean list of segment objects that the rest of the pipeline can consume.

The input is a single prompts_data field: a multiline STRING holding JSON, defaulting to []. The expected shape is a list of objects, each with 索引编号 (an integer grouping key - think "scene number"), 开始时间 (start time in seconds), 结束时间 (end time in seconds), and 提示词文本 (the prompt text). Each object gets normalized to a fixed schema with a 序号 added: {序号, 索引编号, 开始时间, 结束时间, 提示词文本}. That normalization is the whole mechanism - it's a tolerant parser that defaults missing fields to 索引编号=0, 开始时间=0, 结束时间=15, and empty text rather than erroring. There's an optional 刷新标记 (refresh marker) input, the standard pattern in this pack for forcing a re-parse when you change the JSON.

The single output is 多提示词数据, a LIST of those dicts. That LIST is the currency of the whole timeline system: it feeds directly into FxAiMultiPromptLoader, which is the node that actually turns the timeline into per-segment prompt text. Think of the editor as the database and the loader as the query - this one holds the rows, the loader picks the row for the current index.

Where this gets real: it sits upstream of FxAiPromptRelayEncode, the pack's temporal-relay node, in the long-video flows. The JSON gives you a way to hand-edit the timeline as text (or generate it programmatically and paste it in), then the loader converts seconds to frames at your chosen framerate, and the relay encoder biases each time segment's attention toward its own prompt. If you've ever fought a single-prompt video that drifts, this is the structured alternative.

Gotchas: it's a parser, not a validator - garbage JSON quietly becomes an empty list, and you get zero rows back instead of an error, so check the LIST before it's too late. And the times are seconds as floats, but downstream they're converted to frames using the loader's framerate, so keep the framerate consistent between here and the loader or your segments land on the wrong frames. It's in 凤希AI/场景管理 (Scene Management), part of the fxai-toolkit pack by 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 you're building multi-shot video, this is where the timeline starts.