LoadImageExtended

ComfyUI's built-in LoadImage node gives you an image and a mask. LoadImageExtended gives you that, plus a directly encoded LATENT, the filename, some info text, and the width and height - which means for an img2img workflow it can replace LoadImage and a separate VAEEncode node, saving you a wire and a node.

How it works

You pick an image the same way you would with the stock loader - from the dropdown of whatever's in ComfyUI's input folder, or by uploading through the node itself. Where it diverges is the optional vae input: connect a VAE and the node will encode the loaded image straight into a LATENT for you, ready to feed an img2img sampler. Leave vae disconnected and you just get the image, mask, and metadata - fine if you're only doing something with the pixels (compositing, upscaling, captioning) and don't need a latent at all.

The inputs and outputs that matter

• image - required, the file picker/dropdown.
• vae - optional. Connect it if you want the LATENT output populated for img2img; skip it if you don't need that output.

Outputs: IMAGE and MASK work the same as the stock loader - the mask comes from the image's alpha channel if it has one. LATENT is the VAE-encoded version, only meaningful if you wired a VAE in. filename hands back the loaded file's name as a string, handy if you want to base an output filename on the input one. image Info is a text string - treat it as a quick descriptive readout rather than something to parse programmatically, since the pack doesn't document its exact format. width and height come back as plain INTs, useful for matching a resize or an EmptyLatentImage to the source image's dimensions without doing the math by hand.

How to install it

• ComfyUI Manager: search "ComfyUI-Chibi-Nodes", install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/chibiace/ComfyUI-Chibi-Nodes/, restart.

No models to fetch and no extra dependencies - it reads from ComfyUI's standard input directory the same way the built-in loader does. If you don't see your image in the dropdown, drop the file into ComfyUI/input/ (or use the node's own upload button) and it'll appear after a refresh.

Worth keeping in mind for the pack generally: the author's own README calls these "experimental nodes" and jokes that version 2 arrived "with more bugs." Take that as a signal that this is a small, personally maintained utility pack rather than a heavily tested one - reasonable for something this mechanically simple, but don't expect much beyond what the node itself does.

Common issues

The one real gotcha is the LATENT output silently depending on the optional vae input. If you wire LATENT downstream into a sampler but forgot to connect a VAE up top, you'll hit an error at that node rather than at LoadImageExtended itself, which can be confusing to trace back. If your img2img run errors out on a latent-shape or "expected tensor" complaint, check this node first for a missing VAE connection before assuming the problem is further down the graph.

Beyond that, this behaves like any other image loader: if the picture looks wrong on load (wrong colors, unexpected transparency treated as a mask you didn't want), check whether the source file actually has an alpha channel - LoadImageExtended, like the stock node, will read one and hand it back as MASK whether you intended to use it or not.