From 3d4bd10035e6baa8dc40e05c09beaffd5fe27fc7 Mon Sep 17 00:00:00 2001 From: KATO Kanryu Date: Tue, 19 May 2026 09:03:14 +0900 Subject: [PATCH] Feat: Add sample workflow for Text-Driven Multi-Prompt Batch Generation --- examples/text_driven_batch_sample.md | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+) create mode 100644 examples/text_driven_batch_sample.md diff --git a/examples/text_driven_batch_sample.md b/examples/text_driven_batch_sample.md new file mode 100644 index 0000000..e1f8d13 --- /dev/null +++ b/examples/text_driven_batch_sample.md @@ -0,0 +1,19 @@ +# SAMPLE WORKFLOW: Text-Driven Multi-Prompt Batch Generation + +* Multi-Prompt Batch Execution: + Demonstrates how to generate multiple completely distinct scenes in a single + execution without increasing the latent batch size. + +* Decoupled System Architecture: + Separates the fixed "Character Asset" (managed via Text Line Selector) from + the variable "Action/Staging Plans" (sequenced via Prompt Line). + +* The "Batch Size = 1 but 2 Images" Mechanic: + - "Prompt Line" extracts multiple staging rows simultaneously (via max_rows >= 2). + - The extracted rows are combined with the character base and fed into "Join Strings". + - ComfyUI's KSampler detects the multiple text conditioning inputs and automatically + generates one unique image per prompt, even with a latent batch_size of 1. + +* Native SDXL Resolution Driving: + "Resolution Selector" simultaneously controls the dimensions for both + "Empty Latent Image" and "CLIP Text Encode SDXL" to prevent aspect ratio distortion. \ No newline at end of file