凤希AI - 图像管理器 - 批量获取

ComfyUI's built-in Load Image handles one image. The moment you're doing batch work - a frame-animation pipeline, a folder of character references, a stack of storyboard stills - you want a node that loads a whole folder at once and hands you a usable batch. FxAiImageBatchLoad is this pack's version, and it's built around one opinion: every image comes out as a 1024×1024 square, letterboxed onto a black canvas, so a mixed folder of assets all arrives at the same shape without being stretched.

That opinion is worth understanding before you use it, because it's not hiding anything. If your source images aren't square, they get scaled to fit inside 1024×1024 and centered on black bars. That's the classic "contain" behavior - nothing is cropped away, but you get pillarbox/letterbox bars and the aspect ratio is preserved. For feeding reference frames to a video model that wants a fixed square input, it's exactly what you want.

How it works

Give it 图片文件夹路径 and 图片索引 (same selection syntax as the pack's other loaders - 0, 0,2,5, 0:3, or -1 for everything). Each selected image is loaded, optionally shrunk by an integer 缩小倍数 (1–10), then fitted to the 1024×1024 canvas - note the canvas itself is divided by 缩小倍数, so at 2× you get a 512×512 canvas. Every frame is paired with a matching all-ones mask (nothing is cut out, so the mask is just "fully opaque," kept for compatibility with inpainting-style inputs).

The bonus output is a 网格拼接大图: a contact sheet that tiles all the loaded frames into a single image, laid out as a near-square grid. For a quick visual audit of a big folder - "did all 20 reference shots load?" - one glance at that grid beats scrolling twenty previews.

Inputs and outputs

• 图片文件夹路径 - source folder. Bad path raises a RuntimeError.
• 图片索引 - selection string; -1 = all.
• 缩小倍数 - integer downscale applied to both the images and the canvas.

The four outputs are 图片列表 (the frames), 遮罩列表 (matching masks), 网格拼接大图 (the contact sheet), and 总数量. Wire 图片列表 into anything that takes an IMAGE batch, or into the pack's single-frame picker (FxAiImageGetSingle) when you want just one.

Installing it

It's part of fxai-toolkit (author 凤希AI, MIT). Install via ComfyUI Manager (search "fxai-toolkit") or:

cd ComfyUI/custom_nodes
git clone https://github.com/fxai666/fxai-toolkit
# restart ComfyUI

No models. The pack auto-installs soundfile and psutil at first load. Chinese labels; support via QQ group 775649071 and Bilibili.

Where people get burned

The black bars surprise people who expected center-crop. If you feed these square letterboxed frames into an image-to-video model, you'll see black edges in the output - that's the canvas doing its job, and if you don't want bars you should crop first or use a different sizing approach. Also, the mask output is always fully white (all ones): don't treat it as an alpha channel. And remember the 缩小倍数 applies to the canvas too - at 4× you're working on a 256×256 canvas, which is a great way to make fast rough batches but a bad way to lose detail you needed. Errors on individual files are caught and skipped with a console message, so a folder with one corrupt image doesn't kill the whole load.