🐳随机批次姿势

When you're batch-generating - a character reference sheet, a pose study, a run of variations for a dataset - the bottleneck isn't the sampler, it's coming up with varied prompts. PoseSelectorBatch from 🐳 Pond Nodes (comfy_Pond_Nodes) automates the variety: tell it how many batches you want, how many tags per batch, and it rolls the dice across a weighted set of pose categories, spitting out up to ten distinct pose-tag strings at once.

Display name "🐳随机批次姿势" - random batch pose generator. That's the honest job description: it's a randomizer with a seed, not a curated picker.

How it works

The node holds the same pose-tag pool as the pack's other pose selectors, organized into categories (综合, 姿态, 手势, 视线, 整体, 上半身). For each batch it draws tags_per_batch tags, choosing the category first by your weights, then picking a random tag within it. Every batch gets the ensure_tags list prepended - by default "look at viewer" - so your outputs stay on-brief while the rest varies.

The inputs that matter

• batch_count (1–10) - how many pose strings to produce. Each becomes one of the batch_1 … batch_10 outputs (unused outputs come back empty).
• tags_per_batch (1–10) - how many random tags per batch.
• category_weights - the string that controls the dice. Format is 分类:权重, 分类:权重 with the default 综合:0.3, 姿态:0.2, 手势:0.2, 视线:0.1, 整体:0.1, 上半身:0.1. The weights don't strictly need to sum to 1 - they're relative - but keeping them sane means general poses don't get drowned out by gesture tags.
• ensure_tags - tags every batch must include, comma-separated.
• seed - -1 for a fresh roll every run; any other value makes the output reproducible.

Ten outputs, batch_1 through batch_10, each a comma-joined STRING you can wire to separate prompt branches or feed a batch sampler.

Installing it

Same as every node in this pack - Manager search comfy_Pond_Nodes, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Pondowner857/comfy_Pond_Nodes
cd comfy_Pond_Nodes
pip install -r requirements.txt

Restart, under 🐳Pond/text. Pure Python standard library; skip the pack's heavy deps. (And file away the README's comfyui_HiDream-Sampler conflict warning for the day your console floods.)

The gotcha

The random tags aren't filtered for sense - you can get "salute" next to "spread legs" in one batch, and no node is going to police coherence. For dataset work that's often fine (you read each batch before running), but if you're going unattended, tighten ensure_tags and drop tags_per_batch so the randomness has less room to wander. Also worth knowing: the weights are parsed from a text string, so a typo'd category name silently contributes nothing - if a category never appears in output, check the spelling, not your luck.