RloadImageC

RloadImage hands you back a *-typed anything - fine for previewing, useless for anything that insists on a real IMAGE tensor. That's the gap RloadImageC fills: it takes the loaded resource, treats it as a list of image handles, and converts each one into a proper float32 RGB tensor, batches them, and emits images (IMAGE).

Think of it as the adapter between the load side and the pixels. The repo's own example workflow is literally RloadImage → RloadImageC → PreviewImage, and the same chain feeds a VAE encode, a control-image slot, an upscaler - anywhere IMAGE flows. Without this node, the * output is a dead end for anything that renders.

How it works

For every entry in the anything input it runs Image.open, applies ImageOps.exif_transpose (so orientation flags from a phone or editor don't leave the image sideways), handles 16-bit mode-I images, converts to RGB, normalizes to float32 / 255.0, and stacks everything with torch.cat. Feed it one image and you still get a batch of one - [1, H, W, 3] - which is exactly what ComfyUI's image consumers expect. The exif_transpose touch is the sort of detail that tells you this author has actually shipped images; most one-off converter nodes skip it.

Inputs and outputs

Just two sockets:

• anything (*) - the loaded resource, expected to be an iterable of openable image objects. Wire it straight from an RloadImage or Rload output.
• images (IMAGE) - the batched result, ready for PreviewImage or any image consumer.

Install

Same as the rest of the pack: zero extra dependencies, no models. ComfyUI Manager → search ComfyUI-ResourcesLoad → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/MrFrankHobbidy/ComfyUI-ResourcesLoad

Restart, done.

Gotchas

It assumes anything is an iterable of image-like objects. Save a latent or a string with the load side and Image.open will throw - this node is only for image payloads. An empty list crashes on imageg[0], so "nothing loaded, nothing to convert" fails loudly rather than quietly. Keep it wired straight to the load node and both of those stay theoretical.