TOO Simple Image Loader 🖼️

Stock ComfyUI's LoadImage only sees one folder: ComfyUI/input. Every reference image you keep on a second drive, in a giant refs/ folder, or downloaded off Civitai has to be copied in first. TOO Simple Image Loader kills that step - give it any path on your machine and it loads the file, shows you a preview without running the workflow, and digs the prompt, seed, and model out of the image's own metadata. It's one of a handful of nodes I'd install a whole pack for.

The idea is borrowed directly from VideoHelperSuite's Load Image (Path), and the README says so - the author just added a preview and metadata extraction on top. Nothing here calls an API or needs a key; it's all local file reading.

How it works

You type an absolute path into the img_path box. On execution the node opens the file with Pillow, converts it to a ComfyUI tensor, and - this is the useful part - reads the embedded metadata. From a PNG it pulls the A1111-style parameters chunk and the ComfyUI workflow/prompt chunks. From JPEG/WebP it reads the EXIF UserComment and Make/Model tags. That's the whole "the image is the project file" convention, but available as plain outputs you can wire somewhere instead of dragging the PNG back onto the canvas.

The preview is handled by a bundled web extension plus a small server route, so the thumbnail appears the moment you type the path - no Queue Prompt needed. That's the show_preview toggle.

Inputs and outputs that matter

Only three inputs, and you'll touch two of them:

• img_path - the direct path to an image file. The only field you really set.
• show_preview (default on) - kills the preview if you find it noisy.
• image - an IMAGE input for when the pixels come from somewhere upstream instead of a file.

Outputs: IMAGE, MASK, metadata, and workflow. The MASK output is the alpha channel of an RGBA file (inverted to ComfyUI's mask convention), or a do-nothing mask if the image has no transparency - handy if you're loading PNGs with masks from Krita or an inpaint tool. metadata and workflow wire straight into TOO's own Image Metadata editor or Smart Image Saver, which is the intended loop: load a file, fix its A1111 metadata, save it back out.

One sharp edge worth knowing: if you feed the image input instead of a path, the pixels flow through but metadata and workflow stay empty - those are only ever extracted from img_path.

Install

Two ways, same result:

cd ComfyUI/custom_nodes
git clone https://github.com/tetsuoo-online/Comfyui-TOO-Pack

then restart ComfyUI. Or open ComfyUI Manager, search "Comfyui-TOO-Pack" (it's registered on the Comfy Registry), and install from there. The requirements are torch, numpy, Pillow and piexif - piexif is the one you might be missing; a normal desktop ComfyUI already has the rest. No model downloads, nothing heavy.

Common issues

• Metadata comes back empty. The number one cause: the file doesn't carry it. PNGs made by ComfyUI's own SaveImage have the workflow; JPEG/WebP only have it if whoever saved them wrote an A1111 string into EXIF. Reddit-hosted or recompressed images have nothing left at all - re-encoding deletes the chunks, period.
• Path not found. The box is a plain text field with no file browser, so paste the real absolute path. Windows users: escaped backslashes or forward slashes both work.
• No valid source. If both img_path is empty and image is unwired, the node throws instead of guessing. That's the intended behavior.

One honest caveat: this is a small pack by a solo author (Tetsuoo, active on Civitai), and it has basically zero community footprint - you won't find troubleshooting threads. Keep your master files as PNGs, and it'll behave.