凤希AI - 角色加载器

Sometimes your character references aren't in the pack's managed library - they're just a folder of renders you dumped out of a previous session, and you want to pick a few without hand-renaming them. FxAiCharacterBatchLoad (凤希AI - 角色加载器) is the node for that. Point it at any folder, tell it which images you want via a compact index syntax, and it loads them into one IMAGE batch plus a grid preview, scaled down to whatever your VRAM can afford.

How it works

It scans 图片文件夹路径 (image folder path) for image files (jpg/png/webp/bmp/gif/tiff), sorts them, and then interprets 图片索引 (image index) - the clever part. It accepts:

• -1 - every image in the folder.
• 0,2,4 - specific indices, comma-separated.
• 0:3 - a range (start:end, end exclusive, so 0:3 is images 0,1,2).
• Any mix of the above.

Indices are normalized mod the total count, so a stray 9 on an 8-image folder wraps rather than erroring - and duplicates are de-duplicated, so 0:3,2 doesn't load image 2 twice. Each selected image is downscaled by 缩小倍数 (shrink factor) and the batch is concatenated into a single IMAGE tensor, which is what downstream nodes actually want: one tensor, N frames.

Outputs

• 图片列表 - the selected images as a batched IMAGE.
• 网格拼接大图 - a grid composite (auto column count, so it can grow with the batch) also downscaled. Feed this to a reference-aware model or use it as a visual check.
• 总数量 - count of what loaded.

When it beats the managed path

The pack's other character loaders (FxAiCharacterAssetsLoad) assume assets live in ComfyUI/fxai/image/ with subdir/filename paths. This one takes any absolute folder path, which makes it the right tool when your references are elsewhere, when you're prototyping, or when the folder is the source of truth (say, a render farm output). The index syntax is also genuinely useful for iterating: same folder, change 0:3 to 4:7, and you're testing the next batch of poses without touching anything else.

Install

Part of fxai-toolkit (凤希全能节点包), MIT-licensed, by 凤希AI. ComfyUI Manager (search "fxai") or:

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

Pack auto-installs soundfile and psutil; no models, no ffmpeg needed.

Gotchas

• The folder must exist - a bad path raises a clear "folder not found" RuntimeError, not a silent empty batch.
• Range end is exclusive (0:3 = 0,1,2). The one off-by-one thing to remember.
• Everything is labeled in Chinese: 图片索引 = image index, 缩小倍数 = shrink factor, 总数量 = total count.
• Loading 40 images at full res with 缩小倍数 1 is how you run out of VRAM before the sampler even starts. Downscale early - that's what the factor is there for.