🌀 BV LUT Loop End

If BV LUT Loop Start is the loop's top, this is the bottom: the node that takes whatever your current job produced, hands it to the next job as its input image, and - when the plan is done - hands you the accumulated final image. Without it the loop is just a stack that never returns anything.

You use it in the middle of the regional LUT loop pattern: Start → Loop Job Resolver → BV Apply LUT → Loop End → (back into the loop) → final image out. The interesting thing is that nothing about this node's inputs says "LUT." It takes a flow signal and a processed processed_image, and everything LUT-specific lives in the loop state it's carrying along. That's the design: the loop is generic, the jobs are LUT-flavored.

How it works

Under the hood this node is a tiny piece of graph programming. When it runs, it expands itself into three hidden prototype nodes:

1. BV LUT Loop Advance (Prototype internal) - takes the current loop state and the processed image, validates the image shape is unchanged, and bumps the job index. It outputs the next state and a continue_loop boolean (true while there are jobs left).
2. BV LUT While End (Prototype internal) - checks that boolean. If the loop should continue, it re-runs the contained subgraph with the new state; if not, it passes the state through.
3. BV LUT Loop Result (Prototype internal) - pulls the final image out of the finished state.

You'll see these three appear in the node graph if you enable debug/legacy views, but you never wire them yourself. Loop End is the single human-facing surface.

There's one shape rule that will bite you eventually: processed_image must keep the same B, H, W, C shape as the loop's current image. The advance node checks this and errors if your Apply step resized the image. Keep your grade resolution-stable across loop iterations.

The inputs and outputs that matter

• flow - connect the flow output from BV LUT Loop Start. It's marked with rawLink, meaning the node reads the actual upstream node to find the loop state - one more reason to keep this wiring exact.
• processed_image - the output of your Apply LUT (or whatever processing sits inside the loop body).
• final_image - the single output: the accumulated result after every job in the plan has run. This is what you preview or save.

How to install

Ships in BV Node Pack:

cd ComfyUI/custom_nodes
git clone https://github.com/BlackVortexAI/bv_nodepack.git

Restart ComfyUI and hard-refresh. Or ComfyUI Manager → "BV Node Pack". No models to fetch.

Common issues

• "requires processed IMAGE with unchanged B,H,W,C shape" - your loop body resized the image between iterations. Either don't resize inside the loop, or restructure so the resize happens after Loop End.
• "BV LUT Loop ended before every job was processed" - this comes from the hidden Result node when the loop collapsed early; usually a sign the plan or flow wiring got disconnected mid-loop. Re-check the Start ↔ End flow link.
• No final image at all - the flow input isn't connected to the Start node. It's easy to wire only processed_image and wonder why nothing returns.

This is one of those nodes where the "ah, I get it" moment is realizing the loop isn't a feature of ComfyUI - it's a feature this pack builds for you. Loop End is where it seals the deal.