凤希AI - 分段场景 - 时间轴获取器

FxAiMultiPromptLoader is the working half of the pack's timeline-prompt system: the editor (FxAiMultiPromptEditor) holds the time-stamped prompts, and this node queries them for the current segment and renders the answer in every format the rest of your graph might want. If you've got a 多提示词数据 LIST coming out of the editor, this is the node that turns it into something a sampler or a relay encoder can actually eat.

Its job is to filter the timeline down to one 索引编号 (index value - your "scene" key), then convert each matching segment's start/end seconds to frames at the 帧率 you set (default 24). For each segment it emits a line like [12-48]:a cyberpunk street at night (the frame version) or [0.5-2s] a cyberpunk street at night (the seconds version). It also does a genuinely thoughtful little fix: when two adjacent segments end/start on the same frame, it bumps the later one forward by a frame so the segments don't overlap - a real bug source in naive timeline code.

Inputs that matter: 多提示词数据 (LIST, forced from upstream), 索引值 (INT, which scene to select), 帧率 (INT, 24 by default), 默认提示词 (STRING - the fallback when nothing matches), plus the usual 通用提示词/尾部通用提示词 globals and 刷新标记. The four outputs are the interesting part, because they're four different views of the same data:

• 时序按帧数提示词 - STRING, [start-end]:prompt lines with globals, the frame-anchored format for temporal-relay workflows.
• 分段纯文本提示词 - STRING, just the prompts glued together (globals + each segment's text), for when you want one combined prompt.
• 字典数据结构 - DICT: {global_prompt, segments:[{prompt,start,end}], total_frames}. This is the one that feeds FxAiPromptRelayEncode, which needs the structured segments to build its temporal attention mask.
• 时序按秒数提示词 - STRING, the [0.5-2s] human-readable version.

The multi-output design is the pragmatic part: whichever downstream consumer you're using, there's a ready-made port instead of a conversion node in between.

Where it fits: the temporal-relay pipeline for long video - the same "this prompt at this moment" problem the community solved for LTX via Prompt Relay (ecosystems/ltx-video.md). Watch the framerate consistency between this node and the latent's actual frame count - the DICT's total_frames is computed in this node's framerate, and if it disagrees with the latent, the relay mask misaligns. It's in 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 you're building multi-scene video, this is the node that makes the timeline executable.