Silver H3 Line Batch

There's a specific annoyance this node fixes, and it's the oldest one in ComfyUI automation. Your prompts input on the H3 Batch or Staged node is a list, you've got thirty prompts to work through, and the workflow is designed for about five of them at a time. So you copy, paste, run, copy, paste, run. SilverH3PromptsBuilder - displayed as Silver H3 Line Batch - makes that a one-line change between runs.

Paste all your prompts into one multiline box, one per line. Set batch to how many you want rendered per run. Bump index to advance to the next window. That's the whole feature, and if you've ever managed a render queue by hand this will feel like getting a spare hand.

The inputs, all three of them

text is the multiline box: one prompt per line, and everyone's favourite detail - blank lines are skipped, so you can space your list out for readability without polluting the batch. Every non-empty line is one video prompt.

batch is how many prompts per window (default 5, up to 4096). This maps directly onto generation cost: each entry becomes one clip in the H3 nodes, so batch is "how many renders per run" and also, if you're combining, "how long the stitched master will be."

index is which window to emit. The window starts at index * batch - after blank lines are removed - and runs for batch entries. Set index to 1 and you get prompts 6–10 at the default batch of 5. Bump it each run to march through the book.

The output is a single prompts STRING_LIST for this window, and you wire it straight into the prompts input on SilverH3Batch or SilverH3StagedMultishot. One entry, one clip.

The move that makes it actually automatic

Type into index, right-click it, and pick Convert widget to input - no, better: leave it as a widget and look at the little dropdown ComfyUI puts next to every integer widget, control_after_generate. Set it to increment. Now each queued run advances the index by one on its own, and you can hit Queue five times in a row and walk away while it renders prompts 1–5, 6–10, 11–15 and so on.

One warning about that dropdown, because it's the most reliable beginner injury in ComfyUI. The control fires after the run by default, so the number in the box is the one that will be used next, not the one that just ran. On a seed that's how people lose good generations; on an index it means your first run may not be the window you expected. The global fix is Settings → "widget control mode" → Before, which makes the box show the value actually used. Worth doing once on any machine you queue from.

If the whole idea sounds like it belongs in a different program: it does, and the reason it's a node is that ComfyUI gives custom nodes a text box and a counter and nothing else. The pack's authors built the loop out of the two primitives available.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/SilverAndJade/comfyui-silver-nodes

Restart ComfyUI; Manager search "Silver Nodes" works too. No dependencies of its own - the pack's requirements.txt (beautifulsoup4, moviepy, a pinned opencv-python) is for the loader nodes in the same pack, and the H3 nodes want a reasonably current ComfyUI.

Where people get burned

Index times batch, not index. index 2 with batch 5 is prompts 11–15. If your second window looks like it skipped a prompt, recount with blank lines removed.

Walk past the end of your list and there's nothing to emit. Nothing dramatic happens, but a run that renders zero clips is usually an index that drifted too high, not a broken node.

Don't mix it with the other prompt helper and expect them to cooperate. Silver H3 Prompt List is the adapter that turns a string into a list and appends one extra entry; this node is the sliding window over many entries. Feeding one into the other is possible and almost never what you meant.

Re-check your wiring after pack updates. The H3 nodes' inputs have changed over time, and a workflow saved against an older schema keeps stale widget rows that misalign the connected inputs. If a run behaves as though the wrong prompt list went in, recreate the node in the graph - it's not your prompting.