Organize optimized workflows by reusable type

This commit is contained in:
Morpheus
2026-09-23 08:47:12 +00:00
parent a321551449
commit 7ff0b7547a
30 changed files with 88 additions and 22 deletions
@@ -0,0 +1,18 @@
# Qwen Image 2.1 — character-sheet edit workflow
Accepts any character image as `character_reference.png` and produces a nine-panel sheet at exactly 1920×1088: full-body front, three-quarter, left/right profiles, back, three-quarter back, and three close-up head shots.
The input is resized and padded to a 1024×576 16:9 generation canvas before it is passed as a Qwen Image 2.1 reference image. The finished result is resized to 1920×1088 before saving to fit the available GPU memory. The prompt treats the input as authoritative, so it can be an illustration, render, photograph, or existing character sheet and does not assume gender, clothing, or visual style.
- API workflow: `workflow.json`
- WebUI version: `ui-workflow.json`
- Input filename: `character_reference.png`
- Output: `fluxExperiments/qwenImage21/edit_character_sheet_*.png`
- Canvas: 1920×1088
- Steps: 24
- CFG: 4.0
- Sampler/scheduler: Euler / simple
The API workflow uses ComfyUI’s flattened dynamic input key `images.image_1`, which is required when queueing the reference image through `/prompt`.
The server’s exact installed stack is the same as the text-to-image workflow: `qwen_image_2.1_int8_convrot.safetensors`, `qwen3vl_8b_w4a8.safetensors`, and `qwen_image_2.1_vae_bf16.safetensors`.