凤希AI - 影视剧场数据解析

FxAiScreenLoad is the pack's 影视剧场 (cinematic theater) data parser - the node that unpacks one "scene" of a short-drama or film-explainer workflow into everything the segment needs at once: the dialogue line, the footage images, the frame count, the voice reference audio, and the full audio track. Where the scene-loaders just hand you a prompt and indices, this one actually loads the assets. If you're making those AI short-drama videos - a narrator, a couple of shots, matching voices per character - this is the node that assembles one shot's worth of material.

It reads from 场景数据 (a LIST built by the scene manager) at 行索引. The scene dict's 台词 (or 提示词文本 as fallback) becomes the dialogue, wrapped in the optional 通用提示词/尾部通用提示词. The 素材 field is a comma-separated list of image paths resolved against ComfyUI/fxai/image/ - each path loads via the pack's image loader and lands in the 素材 IMAGE output. Then two genuinely clever things happen.

First, frames: it computes the shot length at 24 fps from the scene's 时长 field, then aligns the frame count to the MiniMax H3 frame grid (x + (5-x) % 17 - the "17k+5" pattern the pack's own README documents). That's not numerology; H3's latent works on that grid, and feeding it a misaligned frame count is how you get dropped frames and drift on a multi-segment render. Second, voices: for each image path it asks the character-profile manager which character that avatar belongs to, and if that character has a voice file, it loads the full audio (the 音频 output) and a sliced reference - the 参考音色 output - cut to 音色剪切时长 seconds, or to the scene's own duration if that's 0. That reference slice is the voice-identity sample you feed a TTS or voice-clone node so every line from that character sounds like the same person.

The five outputs are the whole shot: 台词 (STRING), 素材 (IMAGE), 帧数 (INT), 参考音色 (LIST), 音频 (LIST). Each wires into the appropriate part of the H3/MiniMax pipeline.

Gotchas from the source. Image paths are split as subdir/filename against ComfyUI/fxai/image/, so a missing file just logs "[凤希] 图片不存在" and moves on - your 素材 output can be empty without any hard error, which is a quiet failure mode. Voice loading depends entirely on the character-profile manager having a voice registered for that avatar; no profile, no audio. And out-of-range indices return an empty scene rather than throwing, so keep 总行数 in sight. There's a companion FxAiScreenManager for building the scene data this reads, and the pack ships full "影视剧场" example workflows for MiniMax H3 that show the whole chain.

It's under 凤希AI/影视剧场, 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 goal is voiced short-drama at scale, this node is the reason the pack exists.