NAI 图片保存器

This is the node that actually writes your image to disk, and in this pack it's more than a formality. ComfyUI's built-in Save Image only writes PNG and stuffs the whole ComfyUI workflow into the file. NAI 图片保存器 (Image Saver) writes real PNG or WebP, embeds NovelAI-style metadata - prompt, negative, model, seed, sampler, characters, vibes, parameters - and, critically, keeps the ComfyUI workflow out of the file by default. For a paid API where every render cost you Anlas, that's the difference between a shareable artifact and a file that hands your exact pipeline to anyone who opens it in an editor.

It's an output node (it always runs, and ComfyUI's cache works backward from it), and it passes the image through on its 图像 output so you can chain a preview or another saver behind it.

The inputs that matter

Most are the pack's standard wiring - image, NAI_PIPE, positive, negative, core and advanced settings, plus strength/noise and request_kind (文生图 / 图生图 / 局部重绘) so the metadata records which kind of run this was. Then the saver-specific ones:

• filename_prefix - default NovelAI/GENYTOOLS, and it behaves like ComfyUI's prefix: subfolders via slashes, counter suffix appended.
• format_mode - 跟随高级参数 (follow the advanced card's image_format), or force PNG / WebP here. Default follows the card, so the NAI 高级参数 node's format choice is what lands.
• webp_quality / webp_lossless - 95 and lossless by default. NovelAI metadata rides in WebP's EXIF/XMP; lossless WebP keeps quality and still beats PNG on size.
• include_comfy_workflow - the privacy switch, default off. Flip it on if you explicitly want the graph embedded like a normal ComfyUI save.

seed_list is a JSON array of seeds (default [0]) matched to the image batch, so the metadata records the actual seeds that produced each image, not a placeholder.

Why you should just use it

Two concrete wins over the default saver. First, real WebP - the built-in node can't do it at all. Second, transparent backgrounds: a V5 transparent-background PNG keeps its RGBA alpha channel here, which the naive SaveImage path can flatten. And on the privacy side, the default-off workflow embedding is the pack deliberately choosing not to leak your token-bearing graph into every output - the metadata section of the ecosystem docs spends a whole section warning people that the default is disclosure; this node flips that default.

Install

Search ComfyUI-NovelAI-GENYTOOLS in ComfyUI Manager, or git clone https://github.com/XTOGENY/ComfyUI-NovelAI-GENYTOOLS.git into custom_nodes + pip install -r requirements.txt. Output lands in the normal ComfyUI/output/ folder.

One quirk: it needs a standard ComfyUI IMAGE batch and will error if you hand it something else, and it expects the typed prompt objects wired in (it reads the actual prompt text for metadata from them). Leave a socket dangling and you'll get a "needs standard IMAGE batch" style error rather than a silent save - annoying, but it beats writing a file with empty metadata.