🎨 Auto Prompt Combiner

Auto Prompt Combiner is the pack's answer to the "which artist style suits which character?" question - the one that's genuinely painful to answer by hand. You give it a character file and a style file, tell it how many of each to use, and it runs a nested loop: for every character you picked, every style you picked, producing char_count × style_count full prompts.

That's 6 characters × 10 styles = 60 prompts in one run, and you finally get to see Reimu in the style of flooxyfloox next to the style of nikukyu without typing either prompt. For character-sheet exploration or style studies on Illustrious - where artist tags like style by pnya are first-class citizens the model actually honors - this is exactly the workflow you'd otherwise build by hand in a spreadsheet.

How it works

The mechanics are in nodes/prompt_combiner.py and they're refreshingly boring. It parses both TXT files (same tags<TAB>name format as the rest of the pack), then:

• char_start_index and char_count pick the character window
• style_start_index and style_count pick the style window
• for each character, for each style: quality tags + style tags + character tags + random action + random background + random camera + custom positive

Each prompt gets its own fresh random action/background/camera (seeded by seed for reproducibility). One hard cap worth knowing: char_count × style_count is clamped to 100 - the node returns an error string if you exceed it, which is the author protecting you from a 10,000-prompt accident.

The inputs that matter

• character_file and style_file - two dropdowns, both populated from the pack's prompts/ folder. The samples are sample_1girl_v1.txt (characters) and style_names_v1.txt (67 style by … artist lines).
• char_count / style_count - the two dimensions of your grid. Mind the product cap.
• preset - defaults to dynamic here (the loader defaults to standard), which is a deliberate choice: for style combos the author figures you want motion. Swap to none if you want the raw style by X tags to do the talking.
• random_action / random_background / random_camera - same toggles as the other nodes, on by default.

Outputs: prompts (a list of char_count × style_count strings) and negative (one shared negative, preset + your custom_negative).

The gotcha is identical to the batch node

prompts is a list, and a plain CLIP Text Encode will only ever see the first element. The author's combiner_workflow.json demonstrates the same pattern as the batch workflow - preview the list, sample slot 0. To actually render all 60, you need a list-aware consumer (Impact Pack ForEach or similar). The good news: because the list is ordered character-major (all styles for char 1, then all for char 2), you can read the preview and know exactly which row of your grid you're looking at.

Install

ComfyUI Manager: search "Auto Anime Prompt", restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/jluo-github/comfyui-auto-anime-prompts

No dependencies beyond what ComfyUI ships, no models to download, MIT license. Python 3.10+.

Verdict

This is the pack's best idea, slightly undercut by the list-consumption friction in stock ComfyUI. If you already have list-handling machinery in your graph, it's a genuinely powerful little grid generator for Illustrious artist-style exploration. If you don't, expect a ten-minute detour wiring up iteration before it pays off - then it pays off every single time you'd have hand-built a comparison grid.