🗂️ TA Prompt Controller

Once you've got an LLM writing prompts for you, you immediately need a way to turn it off. TA Prompt Controller is the switch: it routes between a manually typed prompt and a generated one, or blends them in either order, and a single dropdown changes the whole behavior of your prompt path.

The two inputs are manual_prompt (a multiline text box, with dynamic prompts syntax supported) and generated_prompt (an optional string input - leave it unconnected and it defaults to empty). The mode dropdown has five positions:

• Manual Only - ignore the LLM, use your text.
• Generated Only - ignore your text, use the LLM's.
• Combine: Manual + Generated - your prompt first, then delimiter, then theirs.
• Combine: Generated + Manual - the reverse.
• Clear / Empty - output an empty string, useful for deliberately blanking a prompt.

The delimiter defaults to ", ", and in combine modes the node only inserts it when both sides are non-empty - so a missing generated prompt degrades to just your manual text rather than a trailing comma. Output is a single final_prompt string that feeds your CLIP Text Encode.

The workflow shape this enables is the classic TA pattern: TA Smart LLM generates a prompt, TA Prompt Controller decides whether it's allowed through, and your manual text is the safety net. During iteration you flip between "Generated Only" (see what the LLM produces) and "Manual Only" (fall back to what you know works), and when the model is behaving, "Combine" layers the LLM's take on top of your base prompt. It's the same mental model as an A/B switch but with a blend mode thrown in, and because the mode is just a widget, you can change it per-run without touching any wiring.

If the LLM backend is down, the graceful path is obvious: generated_prompt arrives empty, so "Generated Only" silently produces an empty prompt - which is exactly when you'd flip to Manual Only. Pair it with the pack's TA_LLM_PromptSelector if you want that fallback handled automatically instead of by hand.

Install. No dependencies; standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/tmode-1960/TA-ComfyUI-Nodes-Pack

or ComfyUI Manager (search "TA ComfyUI Nodes Pack"), restart, under TA Nodes/text.

Gotchas. Three things. "Clear / Empty" is deliberate - you may genuinely want a blank prompt, but don't leave it set by accident and wonder why nothing renders. The combine modes skip the delimiter when one side is empty, which is good behavior but can surprise you if you assumed the delimiter always appears. And the pack's v2.x rename means TA v1 workflows need rebuilding before this node will show up in them.