🔶 Cache Reloader IMG-> LOAD

This is the image-specific version of Chaosaiart's Save/Load cache pair, and it's the node that actually feeds your img2img chain from one generated frame into the next. Where chaosaiart_reloadAny_Load is a generic "remember anything across runs" node, this one is purpose-built for images: it knows how to size a blank starting canvas when there's nothing cached yet, and it has an explicit override for seeding the very first frame with a real picture instead of a blank one.

The mechanism is the same feedback trick the whole cache system runs on: a chaosaiart_reloadIMG_Save node elsewhere in your graph writes an image tagged with a reloader slot number at the end of each run, and this node - tagged with the matching number - hands it back at the start of the next run. Since ComfyUI workflows can't literally cycle, this is how "generate, feed the output back in, generate again" actually happens across repeated queues. Whether it's actually the first run or a later one is decided by the restart input, which needs to come from chaosaiart_restarter_advanced.

On that first run, there's no cached image yet, so you need something to start from. That's what preSave_Image_Mode (aspect ratio: widescreen 16:9, portrait 9:16, or square) and preSave_Image_Size (360p / 480p / HD / Full HD) are for - they define the blank starting canvas size when nothing's cached. If you want to seed frame one with a real image instead of blank, wire it into the optional preSave_image_override - that's the img2video path, where you're animating outward from an actual starting photo rather than pure noise. The optional vae lets the node decode/encode as needed for whatever it's doing internally with the cached data; if your images come out looking off, double-check it's the same VAE your checkpoint expects. reloader is the slot number and, again, has to match whatever's on your paired Save node.

Output is a single IMAGE - either the previous frame's saved output, or your blank/override starting image on the first pass.

Install: search "Chaosaiart-Nodes" in ComfyUI Manager, or use "Install via Git URL" with https://github.com/chaosaiart/Chaosaiart-Nodes. Manually, cd ComfyUI/custom_nodes && git clone https://github.com/chaosaiart/Chaosaiart-Nodes and restart. On Linux you also need pip install opencv-python and pip install tqdm inside your venv - the Windows install script handles this automatically. Given this node deals in actual image data across runs, opencv-python is the dependency doing the heavy lifting behind the scenes.

Common trip-ups: a reloader number that doesn't match its Save counterpart means you'll only ever get the blank/override starting image, forever - every frame looks like frame one. And if your preSave_Image_Mode/Size doesn't match the resolution your checkpoint actually wants (say you picked 360p on an SDXL setup), you'll fight resolution artifacts before you even get to the animation part - pick the size that matches your model's native resolution, not just whatever looks convenient.