NAI 批量提示词任务列表

This node is the part of the pack that treats a paid image API like a for loop. NAI 批量提示词任务列表 (Batch Prompt Tasks) doesn't generate anything itself - it builds a list of tasks for the NAI 批量文生图任务 executor to chew through. You define the fixed parts of a prompt once and a column of things to swap in, and the node expands that into a set of complete prompts, all in the same base style.

The input surface is small and deliberately shaped:

• base_positive / base_negative - the fixed NAI_POSITIVE_PROMPT / NAI_NEGATIVE_PROMPT objects from the pack's prompt nodes. This is your {base}: "same style, same character, same lighting."
• task_lines - the list itself, one task per line, in Name::prompt form:

  角色1::1girl, red hair
  角色2::1girl, blue hair
  

  The :: splits a readable name from the prompt fragment. Max 100 tasks, and the node refuses to build more so you can't accidentally queue a hundred paid requests.
• positive_template - how base and item combine, default {base}, {item}. Want the item before the base, or wrapped in parens? Edit the template.
• negative_lines - optional per-task negatives, one per line, aligned by line number to the tasks. Leave blank and every task inherits base_negative.

Outputs: 批量任务 (the NAI_BATCH_TASKS object the executor consumes), 任务数量 (an INT, useful for sanity checks or wiring into a display), and 任务摘要 (a string naming the first 12 tasks so you can eyeball what you built before spending).

The point of the split

The task-list node is deliberately separate from the executor because it's cheap and free - it does no API calls, just text expansion. So you can build, inspect, reorder and re-run the list as many times as you want, and only the executor (the node that actually generates) costs you anything. That's a good design: all the fiddling happens at $0, and you commit Anlas only when the task list actually looks right.

It also composes with the prompt processor: because the tasks are built from typed prompt objects and can carry per-task negative lines, a randomized or chunked base prompt (via the processor) flows through to every task.

Install and the gotcha

Same pack install: search ComfyUI-NovelAI-GENYTOOLS in ComfyUI Manager, or git clone https://github.com/XTOGENY/ComfyUI-NovelAI-GENYTOOLS.git into custom_nodes + pip install -r requirements.txt, token in settings.

The gotcha is the :: separator being the one true format. Forget it and write 1girl, red hair on a line with no name, and the parser won't split what you meant - check 任务摘要 before you queue. And remember: this node is the cheap half. The expensive half is one node away, and it's where the max_total_anlas guard lives.