fxai-toolkit
ComfyUI - 凤希全能节点包
Nodes (101)
Slice your soundtrack into equal chunks so a long video never OOMs
This node fetches the right file for the current chunk
Point it at a folder of audio and it plans every cut for you
The frame-count math that keeps audio and video from drifting apart
Hear every audio segment before you render them all
A file manager for your soundtrack folder, parked right inside ComfyUI
Glue every segment back into one soundtrack with a single ffmpeg call
The audio splitter that keeps your long video on the rails
The segmenter that knows MiniMax H3 doesn't like your frame count
Cut the exact slice of audio your current loop iteration needs
Queue a long batch before bed, and let the PC switch itself off after
A visual library for your character reference images, no graph needed
Turn a character asset list into a ready-to-use image batch and a contact sheet
Load a whole folder of character shots in one pass, with an index selector
Pull one frame out of a character image batch by index
Pick character reference images from the library right on the node
Name, portrait, voice, and bio per character
One node that hands you your character's face, voice, and bio at once
The no-audio frame calculator for when you know seconds, not samples
Build the first/last-frame pair that bridges two video segments
The first/last-frame node that stops chained videos from jumping cuts
The conductor node that tells your segmented-video loop how many times to run
Turn a folder of images into a comma-separated list of paths, no Python required
Stitch two image lists into one — the smallest node in the pack, and surprisingly useful
Load a folder of images as a clean batch — square canvas, masks included
Downscale an image by an integer factor without breaking downstream nodes
Pull one frame out of a batch — with its mask and dimensions attached
Load an image from a Base64 string — for when your image arrives as text
Stretch a short image sequence into a loop of any length — for seamless repeats
Upload, browse, reorder, and auto-number your assets
Overwrite control, custom numbering, and deletion
A preview node that actually handles this pack's Python-list image format
Pick one image from a list by index — and unlike its sibling, it refuses out-of-range picks
Scale any image so its longest edge hits your target — then snap it to a friendly multiple
Force a whole frame sequence to one resolution — white-letterboxed, not stretched
A float input node — boring, but it's the 'define once, wire everywhere' glue
Define a number once and wire it to every widget that needs it
One wire, both numbers out
Swap one frame's worth of compressed data for a reference
Ask a video latent how many frames it holds — with one format caveat
Pull a single frame out of a latent without decoding anything
The audio loader that plays nice with ComfyUI loops
Batch image loading for people who hate duplicate nodes
Load the right prompt for the right shot, without editing the graph
Stack LoRAs by the dozen, with per-LoRA fade-in, inversion, and trigger tags
A custom LTX 2.3 sampler that locks the face and syncs the lips at once
Plant keyframes anywhere in an LTX clip, not just the first and last
A ratio calculator that gives you model-friendly sizes, every time
The ratio node with half-dimensions baked in, for split-layout workflows
Cut the audio for segment N of a long MiniMax H3 render, frame-aligned
Reuse the DiT blocks you've already computed — faster H3 at a quality cost
Stop hand-computing H3 frame counts — let the audio length do it
Build a MiniMax H3 AV latent from prompt, keyframes, refs, and audio
Keyframes, refs, and audio with fewer knobs
Turn H3 output frames and audio into an mp4 — and hand the tail back to the loop
Every frame in, one transition frame out
Pick one of up to seven audio tracks, by number, without rewiring
One node, a stack of prompts, and a counter to walk them
Turn a paste-in scene list into a loop-ready batch, with a loop count
Format your scene list into one multi-panel 'grid' prompt
FxAiMultiLineTextLoad
FxAiMultiPromptEditor
Turn a prompt timeline into frames, text, and a dict — the timeline query node
FxAiMultiViewGenerator
FxAiPhotoSizeConfig
FxAiPhotoSizeConfigV2
Polish your prompts with a local Ollama model, no cloud, no key
FxAiPromptManager
FxAiPromptRelayEncode
FxAiQwenEditEnhanced
The clean Qwen image-edit encoder, minus the character-list complexity
FxAiQwenEditMultiView
The stripped-down three-view generator that skips the face magic
FxAiRandomSeed
The end-of-run housekeeper that stops long batches from dying on you
The per-scene unpacker that feeds one segment at a time
FxAiSceneLoadV2
FxAiSceneManager
Scene management, now with audio-start control and a duration list
Dialogue, footage, voices, all in one pass
Turning a screenplay list into scene data a loop can actually eat
The frame math behind 'generate in chunks' — done for you
Cut the audio to match the video segment — loudness, sample rate, and stereo in one pass
The most boring node in this pack, and that's why you want it
A loop controller that keeps your segment loop honest
Your shot list, turned into a loop the graph can walk
Pulling one shot's prompt and props out of the storyboard
The storyboard reader, slimmed down for character assets
V2, the sequel that's literally the same movie
The tiny node that unblocks a whole pipeline
A print() statement for your graph
Turn your generated frames into an actual mp4 on disk
The video generator that made the transition frame an explicit citizen
The generator that finally made audio a required input
The generator that auto-derives your transition frames
Load a video as frames without a Python video library in sight
The node that tells you where the pack actually saved your videos
Stitch your segments back into one video, fast
Preview an mp4 path in ComfyUI — the terminal node for your video pipeline
Chop a long video at exact seconds — with an optional title-card head
视频转VR actually just re-wraps your flat video for a headset — here's the honest version
不知道怎么操作的进Q群:775649071 不想加群的,那就去B站看视频吧:https://space.bilibili.com/1242183138
ComfyUI - 凤希全能节点包 产品说明书
一、产品概述
1.1 产品定位
ComfyUI - 凤希全能节点包是ComfyUI生态专属、主打低配设备友好的AI长视频全流程解决方案。工具集内置全套自研节点,核心解决行业通用痛点:原生ComfyUI生成长视频时显存爆满、内存溢出、程序闪退、低配电脑无法运行长帧任务等问题。
本工具通过视频分段拆解、循环迭代生成、帧数据轻量化处理、分段合并渲染核心技术逻辑,将超大时长视频任务拆分碎片化执行,从底层降低硬件资源占用,同时一站式覆盖图像、音频、提示词、场景、分镜、自动化运维全链路功能,是适配个人创作者、自媒体、AI工作室的全能型ComfyUI节点插件合集。
1.2 核心解决痛点(行业刚需)
- 硬件门槛高:原生AI视频生成仅支持高端显卡,16G以下显存电脑无法生成10秒以上长视频
- 资源溢出报错:长帧序列生成过程中,显存、内存持续堆叠,极易出现OOM报错、工作流中断、软件闪退
- 成片连贯性差:手动分段生成画面断层、人物变形、场景跳转,过渡生硬
- 流程繁琐零散:原生生态音视频处理、分镜管理、提示词批量配置无成套节点,需拼接多款插件,兼容性差
- 批量量产效率低:无自动化调度、批量处理能力,人工重复操作多,无法规模化出片
1.3 适配用户群体
- C端用户:AI短视频创作者、自媒体博主、剪辑爱好者、低配电脑个人玩家、动画制作新手
- B端用户:AI内容工作室、短视频量产团队、电商内容制作团队、个人独立开发者、ComfyUI工作流定制师
二、核心技术原理(核心竞争力)
本工具区别于普通ComfyUI插件,核心优势为长视频轻量化生成架构,专属技术逻辑如下:
- 长视频分段拆解机制:自动将任意时长长视频,拆解为多段短帧任务,单次仅加载少量帧数据,避免显存一次性满载
- 循环迭代生成逻辑:支持多轮循环续帧生成,每段视频承接上一段画面特征,保障人物、场景、画风高度一致性
- 潜变量轻量化处理:内置潜变量清空、替换、提取节点,实时释放无效缓存,杜绝内存堆积溢出
- 智能过渡帧优化:帧生成器专属优化,自动补全段落过渡帧,彻底解决分段拼接画面断层、卡顿问题
- 全流程资源管控:生成过程实时监控资源占用,搭配自动任务调度、闲置资源释放,低配设备稳定运行不崩图
三、全功能模块详细说明
本节点包包含八大核心功能模块,覆盖AI视频创作全流程,所有节点相互适配、无缝联动,无需额外插件支撑。
3.1 长视频核心生成模块(主打功能)
专为解决长视频生成痛点研发,多版本迭代升级,兼顾稳定性、连贯性和画质。
- 多版本视频生成器(v1/v2/v3):v1基础稳定生成;v2优化人物一致性、画面比例适配;v3强化过渡帧效果,彻底解决长视频人物变形、画面跳变问题,支持超时长视频循环生成
- 帧生成器(v1/v2):AI智能补帧、续帧,修复长视频生成过程中的画面断层、帧率不稳定问题,适配分段循环生成逻辑
- 视频分段帧计算节点:自动计算总帧数、分段帧数、起止索引,精准拆分长视频任务,无需手动算帧
- 视频合并与渲染节点:自动拼接所有分段视频,无缝合成完整长视频,保留原画质、帧率、音效
- VR视频专属生成:支持普通AI视频一键转VR视角,拓展长视频创作场景
3.2 轻量化资源优化模块(低配设备核心保障)
从底层优化显存、内存占用,是低配电脑跑长视频的核心支撑功能。
- 潜变量系列节点:支持潜变量清空、替换、帧提取、帧数统计,实时清理无效缓存,杜绝内存、显存堆积溢出
- 图像缩放降分辨率节点:智能缩放素材尺寸,自定义画质阈值,在不影响成片效果的前提下,大幅降低资源负载
- 数据类型转换节点:整型浮点自动适配转换,优化模型运算逻辑,减少无效资源消耗
3.3 全维度音频处理模块
配套长视频生成,实现音频同步分段、剪辑、合并,音画完全同步。
- 多模式音频加载:单音频加载、批量多音频加载,适配多镜头、多分段视频配音需求
- 音频智能分割:按时长均等分割音频,匹配视频分段生成逻辑,音画时长精准对应
- 音频帧读取与预处理:单独处理音频帧数据,修复长视频配音卡顿、音画不同步问题
- 音频合并合成:自动拼接分段音频,适配长视频成片,支持多音轨叠加
3.4 批量图像素材管理模块
适配图文转视频、帧动画、素材批量预处理场景,支撑长视频批量量产。
- 图像批量/单张加载:高速加载本地素材,支持批量导入、统一管理,适配多镜头创作
- 双版本图像管理器(v1/v2):v1基础素材管控,v2优化批量处理逻辑、修复素材加载BUG,提升批量出片稳定性
- 图像尺寸自定义配置:预设多类视频尺寸模板,支持自定义分辨率,适配短视频、长视频、横纵屏各类场景
3.5 提示词智能运维模块
解决长视频多镜头提示词繁琐配置问题,支持批量、分段、动态提示词调度。
- 多提示词编辑器/加载器:支持多组提示词批量导入、分段匹配不同视频镜头,无需手动反复修改
- 提示词优化节点:智能润色、适配AI生成模型,提升长视频每一段画面的画质统一性
- 提示词中继编码:优化提示词传输逻辑,避免长工作流提示词丢失、失效
- 多行文本加载:支持长文案、分镜文案批量导入,适配故事机长视频创作
3.6 场景与分镜故事板模块
主打长视频结构化创作,支持多场景、多分镜连贯生成,杜绝画面场景错乱。
- 双版本场景加载/管理器:统一管控视频场景风格、画面色调、场景元素,保障长视频全程场景统一,避免频繁跳变
- LTX场景收集优化:适配主流AI视频模型,解决长视频多场景生成错乱、元素丢失问题
- 故事板创建与加载:支持自定义分镜脚本,按分镜自动生成连贯长视频,适配剧情类、动画类长视频创作
- 影视剧场流(场景管理/数据解析):流式管理影视级多场景数据,逐场景解析台词、素材、角色音色与音频,实时打印执行进度,适配短剧、影视解说长视频创作
3.7 自动化运维辅助模块
适配批量量产、长时间渲染任务,解放人工,提升自动化效率。
- 可配置自动关机功能:长视频批量生成完成后,自动触发设备关机,节省设备能耗,适配通宵批量渲染场景
- 生成全局控制器:统一调度帧生成、视频渲染、音频合成全流程,精准控制任务起止索引,避免漏帧、错帧
- 视频预览与纠错:实时预览分段成片,提前排查画面BUG,减少无效渲染,提升长视频成片率
3.8 MiniMax H3 本地音视频生成模块
专为本地 MiniMax H3(Hailuo)模型深度定制,无需第三方云端即可完成提示词驱动、首尾帧锚定、多参考音视频注入的图生视频全流程。
- MiniMax图生视频(v1/v2):提示词 + 首尾帧关键帧(几何锚定)驱动 H3 生成;参考图片/视频/音频按官方 <Picture/Video/Audio N> 编号注入,支持外置音频作音色参考、原音频锁定或系统生成
- MiniMax视频保存(v1/v2):H3 输出帧序列 + 音频直接合成 mp4,过渡帧与过渡帧音频同步输出,供下一镜头循环衔接;v2 为固定帧率/目录的精简版
- MiniMax帧数计算器:按音频时长自动对齐 H3 的 17k+5 帧网格(24fps),输出视频/音频潜变量时长,杜绝逐段舍帧累积误差
- MiniMax音频分段加载:长音频按分段时长切出当前段并对齐 H3 帧网格,供长视频逐段循环生成
- MiniMax块缓存加速:采样早期缓存 DiT 块输出,后期复用,显著降低显存占用、提升生成速度
四、产品核心优势(差异化卖点)
4.1 硬件门槛极低
彻底打破高端显卡限制,8G、10G、12G低显存显卡,甚至低配笔记本,均可稳定生成1分钟、3分钟、10分钟以上超时长AI视频,无OOM报错、无闪退。
4.2 长视频成片质量高
通过循环续帧、过渡帧优化、场景锁定、人物一致性算法,解决市面上分段生成工具普遍存在的人物变脸、场景跳变、画面卡顿、画风割裂等问题,全程画面连贯统一。
4.3 全流程一体化闭环
区别于单一功能插件,本节点包实现「素材加载-提示词配置-分镜设计-分段生成-音画合成-成片导出-自动化收尾」全链路闭环,无需搭配其他插件,兼容性100%适配ComfyUI原生环境。
4.4 支持批量量产运营
所有节点支持批量调度、循环执行,适配工作室规模化AI长视频量产,大幅降低人工成本,提升出片效率。
4.5 持续迭代优化
工具持续更新迭代,不断修复已知BUG、优化生成逻辑、新增功能节点,版本持续升级,稳定性和功能性持续提升。
五、适用创作场景
- AI超长剧情动画、连续帧动漫视频制作
- 自媒体、短视频账号批量长内容量产
- 低配电脑AI视频创作、个人居家创作
- 分镜故事类、剧情解说类长视频生成
- VR全景AI长视频、沉浸式视频创作
- 图文转长视频、批量素材剪辑合成
六、品牌官方定位与宣传Slogan
6.1 官方正式名称(技术/商业/B端使用)
ComfyUI - 凤希全能节点包 定位:ComfyUI低配友好型 AI长视频全流程节点合集
6.2 市场推广名称(引流/C端/运营使用)
ComfyUI - 凤希全能节点包 定位:低配电脑也能跑的全能AI视频创作节点工具
6.3 通用Slogan
核心版:长帧循环迭代,突破显存限制,低配畅跑AI长视频 通俗版:告别显存爆满,低配电脑轻松做超长AI视频
七、配套服务
- 技术社群答疑:官方QQ交流群 775649071,提供安装指导、工作流搭建、故障排查等全方面技术支持
- 视频教程支持:官方账号持续更新新手教程、高阶工作流教学
- 免费迭代更新:持续优化节点功能、适配最新ComfyUI版本与AI视频模型
授权说明 License
开源协议
本项目基于 MIT License 开源。
免责声明
本软件按现状提供,开源版本不提供任何商用担保与技术保障,如需技术支持请联系作者。