AUN Prompt Cycler

Batching out a folder of variations means changing the prompt between runs, and the old way - editing a text box between every generation - is how you spend a whole afternoon. AUN Prompt Cycler automates it: feed it a list of prompts, and every run advances to the next one (or a random one, or one from a filtered subset). It's a direct descendant of tenitsky's prompt cycler, extended with search, ranges, and titled prompts, and it's the quiet workhorse behind a lot of AUN example workflows.

How it works

It keeps a running position internally, and each execution returns the current prompt plus its title and index. Modes:

• Sequential - step through in order, wrapping around. The default, and the one you'll use for "generate all 10 of these."
• Random - pick any prompt each run. Great for exploration batches.
• Manual - manual_index picks a specific prompt (1-based). The "give me #4" mode.
• Range - a comma-separated list of indices or ranges like 1,2,4-8,11. Think "skip the ones I've already seen."
• Search - filter the list by keyword before cycling. mountain sunset means both words (AND), mountain, forest means either (OR), and you can combine: mountain sunset, forest is (mountain AND sunset) OR forest. Case-insensitive.

Prompts go into custom_prompts, one per line, and you can give each a title with Title: Prompt text - the part before the colon becomes prompt_title and the rest becomes the prompt. Lines without a title get auto-generated ones like "Prompt 1." That title output is the sleeper feature: AUN's compact-mode overlays and the AUN_prompt_cycler_selected WebSocket event use it, and you can wire it to a text display to see which prompt is live without reading the whole thing.

The outputs

• prompt (STRING) - goes to your CLIP Text Encode node.
• prompt_title (STRING) - the title of the active prompt.
• index (INT) - the 1-based position. This is the hook that makes the AUN ecosystem click: wire it into a Random Multi-LoRA Model Loader's prompt_index and each prompt automatically pulls its matching LoRA set and trigger words. The README's flagship example is exactly that chain.

Why you'd reach for it

If you're doing any kind of prompt-ab test or batch exploration, this replaces a pile of manual edits with one queue run. The combination that really shines is cycler → multi-LoRA loader: 20 prompts, each with its own LoRA stack, generating in one go. For that alone the pack is worth installing.

Install

Part of AUN ComfyUI Nodes (loz2754):

cd ComfyUI/custom_nodes
git clone https://github.com/loz2754/AUN-ComfyUI-Nodes

or ComfyUI Manager → search "AUN ComfyUI Nodes" → install → restart. Deps (piexif, opencv-python-headless, imageio-ffmpeg, requests) come via Manager; no model downloads.

Gotchas

The internal position is session state - it remembers across runs in the same ComfyUI session, so don't expect a fresh random start unless you restart or reset the node. And because it holds state, two cyclers in one workflow each advance independently, which is usually what you want but occasionally confusing. If a prompt has a colon in it and you want a title, that's ambiguous - the first colon wins as the title separator, so keep titles clean.