Picker T2I Prompt Source

Let's get the honest part out of the way first: Picker T2I Prompt Source, Smart T2I Prompt Source, and Unified T2I Prompt Source are the same node. The code registers all three names against one class object. Different menus, identical behavior, and Picker's schema is literally copied from the original Unified node. So if you've read the article on either of those, you know everything about this one - and the only real question is which name you type into the node search box.

That said, this is still a genuinely useful node if you're working within comfyui-prompt-library-nodes. It builds a positive prompt from a JSON prompt library: quality prefix, up to two roles with outfits, then appearance, pose, scene, camera, and prop snippets, plus freeform additions and a manual prompt.

Why "Picker" is the right mental model

The "picker" framing is about the six picker_* dropdowns: picker_appearance, picker_pose, picker_scene, picker_camera, picker_prop, picker_clothing. Each is populated from a library folder, and when you select an entry, the frontend extension does two things - it appends an alias: Title line into the additions box (e.g. scene: Old Observatory), then resets the dropdown to its placeholder so you can keep picking. It's a rapid-fire way to try variations: swap the scene, swap the pose, queue, repeat, all without ever hand-editing a prompt string.

The additions box is where the picks land and where you can go further - @1.2 applies a weight, a line starting with - removes a term from the final prompt, and no.1girl ... / no.2girl ... scope an addition to a specific subject. manual_prompt appends one-off details that don't deserve a library entry yet.

The inputs and outputs

Everything lives in widgets; the node has no input sockets. prefix picks the quality preset, role_1/role_2 + outfit_1/outfit_2 choose subjects and clothing, seed controls any (random) selections, and selected_terms_preview / prompt_preview are read-only boxes refreshed by the frontend so you can check the composition before queueing.

Outputs: positive (STRING) into your positive CLIPTextEncode, and library_path (STRING) showing which JSON library was loaded.

One thing to know, and it applies to this node as much as its twins: the negative_preset and negative_terms widgets visible in the schema are dead - the node ignores them and never emits a negative. Use the pack's Negative Prompt Source for that. And note the output sanitizer strips Chinese characters and dedupes terms, so content that survives to the final prompt is English-tag-flavored regardless of what's in the library.

Installing it

Zero dependencies. ComfyUI Manager → search "comfyui-prompt-library-nodes" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/2223290690z-hue/comfyui-prompt-library-nodes.git

Then find Picker T2I Prompt Source under Prompt Manager.

Should you use it?

Here's my take: if you're starting fresh, reach for Unified T2I Prompt Source v2 - it's the maintained evolution, adds a third role slot and a lighting picker, and drops the dead negative widgets. Pick this one only if you're working from a workflow that already references it, or if you genuinely prefer the "pickers only" mental model. There's no functional reason to prefer Picker over Smart or vice versa - they're the same node wearing three different name tags, and knowing that up front saves you the confusion of thinking you're missing some version-specific feature.