mmuuai全能节点

This is the node that makes the mmuuai pack worth installing. MmuuaiUniversalSelector ("mmuuai全能节点" in the menu - the "all-rounder node") collapses a whole media-input chain into one box: load up to 20 images and one video, resize everything to a target resolution, and get back a ready-to-use IMAGE batch, individual image outputs, a native VIDEO, an AUDIO track, and a pile of pass-through parameters. If you've ever built the five-node upload-and-resize chain by hand, this replaces the whole upstream half of it.

What it actually does

Every file it touches comes from ComfyUI/input (upload through the node's own gallery, which also handles drag-and-drop, paste, thumbnails, delete, and drag-sorting). Then three jobs, in order:

• Compute target dimensions from 分辨率 (short-edge preset: 480p / 720p / 1080p / 1440p（2k） / 2160p（4k） / 4320p（8k）, default 720p) times 画面比例, rounded to 尺寸倍数 (size multiple, default 16). Set 画面比例 to 自动 (auto) and it keeps the source's own aspect ratio - the right default for most img2img work.
• Resize the pixels with one of five 缩放方式 (fit / stretch / black pad / green pad / center crop) and one of six 插值算法. The default is 兰索斯 (Lanczos, "high quality"), which routes to ComfyUI's own Lanczos implementation rather than a naive nearest-neighbor blowup - a genuinely sane default. The other modes (nearest, bilinear, bicubic, area, nearest-exact) are there for speed or pixel-art.
• Handle video through ComfyUI's native VIDEO interface: decode, skip the first 跳过视频秒数 seconds, resample to 帧率, resize, and produce an AUDIO track (silence is generated to match if the source has none). 时长 (duration) is worth understanding: 0 means "use whatever's left after the skip," while a positive number resamples to exactly that length - cropping overflow and padding with the last frame if the source runs short.

Inputs worth your attention

• 分辨率, 画面比例, 尺寸倍数 - the sizing trio; 自动 + default multiple covers most cases.
• 帧率 (1–240) and 时长 (0–600s).
• 图片1–图片20 and 视频文件 - the media slots.
• JPEG压缩 (0 = off) - re-encodes through real JPEG to simulate loss. The README is explicit that this does not shrink ComfyUI's VRAM footprint; it just makes the output look compressed, which is useful for previewing what an upload pipeline will do to your images.
• 提示词 (multiline prompt) plus the 选择项 / 整数 / 浮点数 / 布尔开关 pass-throughs, which just echo back out - handy when you want one node to carry a workflow's control values instead of six primitives.

Outputs

图片批次 (IMAGE batch - one tensor of everything you loaded, ideal for batch-aware nodes and IPAdapter), the individual 图片1–图片20 outputs (the frontend hides the ones with nothing loaded, a nice touch), 视频 (VIDEO), 音频 (AUDIO), the ints 宽度 / 高度 / 帧率 / 视频总帧数, 文本 (the prompt), and 选择项 / 整数 / 浮点数 / 布尔值. That's a lot of sockets, but they're organized well enough that the node reads as a single media source rather than a wall of ports.

Install & requirements

Same pack, same install - Manager search "mmuuai-resolution-selector" or clone into custom_nodes and restart. Zero Python dependencies, no model downloads, no keys. One real requirement: the README states the node needs a ComfyUI build that provides comfy_api.latest (the native VIDEO API), so on an older install the video path fails with a 视频读取失败 error. And a genuine trap: don't install the separate mmuuai-universal-node repo alongside this pack, or you'll get duplicate node registration.

The gotchas

The pack ships hard resource limits: 500MB video files, 600s source, 6000 output frames, 200M source pixels for images, 300M output pixel-frames. Those are RunningHub's backend protections - this pack was clearly built for RunningHub's hosted ComfyUI, and the checks run in Python so they fire locally too. If you hit a "超过平台保护上限" error on a big batch, lower resolution, fps, duration, or image count rather than fighting the limit. Everything here is local file handling - no API calls, no credentials - which, given this ecosystem's malware history, is worth knowing before you run any small unknown pack. Expect Chinese labels everywhere.