NF Preview Selector (continue)

This is the other half of the Continue button. When you pick images in NF Preview Selector and hit Continue, your node is replaced in the queued prompt by this one - same node id, same three outputs, but no link inputs and no gallery. It loads the candidate batch that was saved to disk and hands the selected images (and latents, if you connected them) to the downstream half of your graph.

You don't add it. You can't usefully add it - it's flagged dev-only, so it won't even appear in the Add Node menu unless dev mode is on. But it's the node whose two inputs explain almost every weird thing you can hit when reviewing batches, so it's worth ten minutes.

Why a second node instead of a mode on the first

Because "run only the second half of the workflow" is a completely different job from "display a gallery and block".

If Continue just re-queued the normal selector, the whole upstream branch would execute again - new candidates, wrong images, expensive. A node with no link inputs has no upstream branch to execute. The frontend also narrows the job to just the output nodes that depend on the selector, so the run consists of the decode-to-save part of the graph and nothing else. The images come back from a file on disk instead of from the sampler.

Which is exactly why candidates die when ComfyUI restarts: they're in the temp folder, and that gets emptied. This node's job is to read a file that the previous run wrote, and if that file is gone, there is nothing to read.

Inputs and outputs

Two required inputs, both plain strings:

• batch_id - which stored candidate batch to load. A 32-hex-character id (a monotonic nanosecond timestamp plus random hex) that the gallery writes into the prompt when you press Continue.
• selection - the indices to keep, comma-separated, e.g. 0,2. This is the same value that comes back out of the normal node as selection_indices.

Outputs are identical to the normal node's: selected_images, selected_latents, selection_indices. Because the class swap preserves the node id, your existing wires to those outputs keep working - that's the point of the swap.

What its error messages mean

Its errors are prefixed [NF Preview Selector], and the codes are worth recognising because they're unambiguous once you know them.

• BATCH_NOT_FOUND - "The candidate images are no longer available (ComfyUI was restarted or they were cleaned up). Generate again." That's the temp-folder rule, stated plainly by the node itself.
• INVALID_SELECTION / "No image selected" - the selection string was empty. Continue with nothing ticked does this.
• "Image index 7 is out of range (batch has 4 images)" - the selection refers to a batch that isn't the one on disk anymore. Usually means you pressed Generate again after picking.
• "Node 12 is not in the prompt (muted or bypassed?)" - the selector itself was muted or bypassed before Continue.

A raw batch_id value that isn't a valid 32-hex id also resolves to BATCH_NOT_FOUND rather than a crash, which is friendly, but it means a hand-edited workflow won't tell you the id is malformed.

Setup

Part of the same pack - install it once and get both nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/NyaFuP/ComfyUI-NF_Suite.git

Restart. Zero Python dependencies, frontend prebuilt in web/, requires ComfyUI 0.22.0+ and frontend 1.44.19+.

The gotcha worth knowing

If you wire this node's outputs into your own Save Image and then press the normal Queue, the graph now has a route to an output that passes through the source node - so ComfyUI will execute it with whatever batch_id is sitting in its widget, which on a fresh workflow is an empty string. You'll get the "candidate images are no longer available" error instead of an image, and it'll look like the node is broken.

The fix is to understand the split: the normal node is the one in your saved workflow, and the source node only ever exists for the duration of a Continue job. Wire your downstream half to the normal selector's outputs, and let the packaging handle the swap.