Text to Pose (Batch)

If Text to Pose is the single-shot node, Text to Pose (Batch) is the contact sheet. Same idea, one difference: instead of one pose from your prompt, it generates batch_size variations of that prompt and stacks them into a single IMAGE batch. It's a convenience wrapper - no new mechanism, no new model, just a loop under the hood. Nothing magical, and it knows it: three impressions on the search side says almost nobody lands here.

You use it when a pose prompt is a family of interpretations rather than one answer. "a person dancing" could mean a dozen different things, and a batch lets you eyeball the whole spread at once instead of rolling the dice one seed at a time. It's also handy if you want a handful of similar-but-distinct poses to drop into a pose-to-pose or animation workflow in one go.

The inputs that differ from the single node

Everything matches Text to Pose - t2p_model, prompt, width, height, seed, and the same two temperature sliders - plus:

• batch_size - 1 to 16, default 4. The tooltip says it best: "Number of different pose variations to generate."

Two details worth knowing. The batch node defaults its temperatures to 0.3 instead of 0.1, so it ships tuned for variety - leave them there unless you want nearly-identical frames. And it derives each variation by stepping the seed (seed, seed+1, ...), which means the batch is reproducible as long as you keep the same starting seed.

What you get out

Just one output, pose_images - a batch IMAGE. Note the difference from the single node: there's no pose_keypoints output here. If you need the raw keypoints for each variation (say, for a scripted workflow), run Text to Pose once per pose instead. But for the common case - feed the batch to a viewer, pick the pose you like, then regenerate that one at full resolution with the single node - this is the right tool.

It needs the same setup as the rest of the pack: a T2P Model Loader feeding it, the install.py step done (a bare git clone won't cut it - that's covered in the T2P Model Loader article), and a patient first run while the ~150 MB model and the CLIP text encoder download.

The honest take: this node is the pack's least necessary piece, a convenience you'll reach for occasionally and forget the rest of the time. But when you're staring at a vague prompt and want to know what the model thinks "dancing" means before you commit pixels to it, the batch view is a genuinely nicer way to explore than clicking re-roll. Install it with the rest of the pack via ComfyUI Manager (search "Text-to-Pose") and you've paid nothing extra for it.