🔶 Cache Reloader IMG-> SAVE

Simple job, important one: this is the node that closes the loop on Chaosaiart's frame-by-frame animation trick. You put your finished frame - whatever came out of your KSampler for this pass - into this node, and it stashes it so that chaosaiart_reloadIMG_Load can hand it back as the starting point for the next frame. Without this node actually running every single pass, there's nothing to carry forward and every "frame" would just start from scratch.

It's an output node, meaning ComfyUI treats it as a terminal step that always executes, not something that only fires if something downstream needs its result. That matters here specifically because the save is the entire point - if this node got skipped because nothing consumed its output, the whole animation chain would break silently.

Two fields, both required: image is the frame you're caching - plug in whatever your sampler/VAE decode step produced this pass. reloader is the slot number (0–100) this save is filed under, and it has to match the number on whichever chaosaiart_reloadIMG_Load node is supposed to pick it back up next run. If you're running multiple parallel image caches in one workflow (say, caching both a full frame and a cropped detail pass), give them different reloader numbers or they'll overwrite each other.

Outputs: IMAGE, a passthrough of whatever you fed in, so you can keep using the frame downstream in the same pass without a second wire back to your sampler. Info gives you a text confirmation of what got cached, useful for sanity-checking your loop is actually writing something and not silently failing.

Install: search "Chaosaiart-Nodes" in ComfyUI Manager and install it, or use Manager's "Install via Git URL" with https://github.com/chaosaiart/Chaosaiart-Nodes. By hand: cd ComfyUI/custom_nodes && git clone https://github.com/chaosaiart/Chaosaiart-Nodes, then restart ComfyUI. Linux needs pip install opencv-python and pip install tqdm on top of that (inside your venv if you use one) - this node's whole job is shuffling image data between runs, so opencv-python is doing real work here, not just sitting unused.

The failure mode to watch for is the same one across every reload pair in this pack: reloader number mismatch. If your Load node is reading slot 5 and this Save node is writing slot 6, you'll get an animation that never actually animates - every frame comes out looking like the fallback/blank starting image because nothing you saved is ever being read back. There's no error for this, just a workflow that quietly doesn't do what you built it to do, so it's worth double-checking both numbers before you go chasing sampler settings.