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 +1,36 @@
# Qwen Image 2.1 — nine-panel character sheet
Creates one square character reference sheet with full-body front, three-quarter, left-profile, right-profile and back views, plus three close-up head-and-shoulders views.
## Server-validated model stack
- Diffusion model: `qwen_image_2.1_int8_convrot.safetensors`
- Text encoder: `qwen3vl_8b_w4a8.safetensors` with CLIP type `qwen_image`
- VAE: `qwen_image_2.1_vae_bf16.safetensors`
- Native nodes: `TextEncodeQwenImage21`, `QwenImage21Cache`
## Use
Replace `[REPLACE WITH A DETAILED CHARACTER DESCRIPTION]` in the `Text Encode Qwen Image 2.1` prompt with the subject description. Describe identity, age, face, hair, body proportions, clothing, accessories, colors, materials, and distinguishing marks. Keep the panel/layout instructions intact.
`workflow.json` is the API prompt format. `ui-workflow.json` is the native ComfyUI web-UI format.
## Starting settings
- Canvas: 1024×1024 square, nine panels
- Steps: 24
- CFG: 4.0
- Sampler: Euler
- Scheduler: simple
- KV cache: automatic device, int8 cache
- Seed: `20260922`
## Prompting notes
Qwen Image 2.1 supports long descriptive prompts and explicit layout instructions. State the grid structure and panel order directly, repeat identity-preservation requirements, and explicitly forbid labels/text if a clean sheet is wanted. The model supports reference images through the same text-encoder node; this workflow is text-only so it can serve as a reusable baseline for any character.
## Sources consulted
- Qwen Image 2.1 model card: https://huggingface.co/Qwen/Qwen-Image-2.1
- Qwen Image repository and release notes: https://github.com/QwenLM/Qwen-Image
- ComfyUI native Qwen node implementation: https://github.com/comfyanonymous/ComfyUI/blob/master/comfy_extras/nodes_qwen.py