What is this?
A local recipe (agent skill) for generating NSFW and pose-controlled images with Qwen-Image 2.1, combining an author-level prompt formula, a mandatory NSFW LoRA, and a 136-pose 3D doll library for precise pose control.
A local recipe (agent skill) for generating NSFW and pose-controlled images with Qwen-Image 2.1, combining an author-level prompt formula, a mandatory NSFW LoRA, and a 136-pose 3D doll library for precise pose control.Is this a ComfyUI workflow, a standalone app, or something else?It's a recipe built on top of ComfyUI: prompt templates, Python scripts (scripts/qwenposetest.py, scripts/runvnccsui.py) and a pose library, meant to be driven by AI coding agents or used manually.
It's a recipe built on top of ComfyUI: prompt templates, Python scripts (scripts/qwenposetest.py, scripts/runvnccsui.py) and a pose library, meant to be driven by AI coding agents or used manually.How do I get started replicating the author's results?Install the prerequisites listed in install/setup.md, then run a single command (e.g. scripts/qwenposetest.py --mode t2i) to reproduce the author's shower sample image in about 5 minutes.
Install the prerequisites listed in install/setup.md, then run a single command (e.g. scripts/qwenposetest.py --mode t2i) to reproduce the author's shower sample image in about 5 minutes.What do I need installed before I can run this?You need ComfyUI plus the qwenimage2.1int8convrot base model (the same checkpoint the referenced Civitai author used) and the NSFW LoRA (default femaledVaginus v1.3b); full dependency details live in install/setup.md.
You need ComfyUI plus the qwenimage2.1int8convrot base model (the same checkpoint the referenced Civitai author used) and the NSFW LoRA (default femaledVaginus v1.3b); full dependency details live in install/setup.md.Can I skip the LoRA or shorten the prompt to speed things up?Not without a warning: the recipe explicitly forbids silently weakening results — if you drop the LoRA, shorten the prompt, or reduce steps, the user must be told the trade-off beforehand.
Not without a warning: the recipe explicitly forbids silently weakening results — if you drop the LoRA, shorten the prompt, or reduce steps, the user must be told the trade-off beforehand.Which coding agents can drive this skill?Claude Code, Codex, Cursor, and OpenClaw can all use it as an agent skill, and it also supports plain manual use by a human.
Claude Code, Codex, Cursor, and OpenClaw can all use it as an agent skill, and it also supports plain manual use by a human.What is Qwen-Image 2.1 in this context?It's the base diffusion checkpoint (qwenimage2.1int8convrot) this recipe runs on, the same model the referenced Civitai creator used, sourced from Hugging Face's Comfy-Org/Qwen-Image-2.1.
It's the base diffusion checkpoint (qwenimage2.1int8convrot) this recipe runs on, the same model the referenced Civitai creator used, sourced from Hugging Face's Comfy-Org/Qwen-Image-2.1.How do the three generation routes differ?Route 1 (T2I) creates new NSFW images from text with the LoRA always on; Route 2 (Edit) changes an existing character's pose while keeping face and clothing; Route 4 (VNCCS PoseStudio) applies one of 136 exact 3D poses via a MakeHuman/three.js skeleton-rotation renderer.
Route 1 (T2I) creates new NSFW images from text with the LoRA always on; Route 2 (Edit) changes an existing character's pose while keeping face and clothing; Route 4 (VNCCS PoseStudio) applies one of 136 exact 3D poses via a MakeHuman/three.js skeleton-rotation renderer.Where do the 136 poses come from?They come from four public Hugging Face repositories (Totemistyk/GeneralPoses, Totemistyk/BalletPoseStudio, MIUProject/VNCCSPoseLibraryMain, MIUProject/MySuperPoseRepo) plus 55 hand-picked additions.
They come from four public Hugging Face repositories (Totemistyk/GeneralPoses, Totemistyk/BalletPoseStudio, MIUProject/VNCCSPoseLibraryMain, MIUProject/MySuperPoseRepo) plus 55 hand-picked additions.What sampler and resolution settings does the recipe use?Euler + beta sampler, CFG 2.5, 50 steps, at a 1152x1728 portrait resolution, with the NSFW LoRA always applied at weight 1.0.
Euler + beta sampler, CFG 2.5, 50 steps, at a 1152x1728 portrait resolution, with the NSFW LoRA always applied at weight 1.0.What are similar projects?shuohao-skills, codex-ppt-skill, threejs-architecture-effects, xxg-portrait-rebuild-light, skills, linkedin-skills, pua, GPT-Image2-Skill
shuohao-skills, codex-ppt-skill, threejs-architecture-effects, xxg-portrait-rebuild-light, skills, linkedin-skills, pua, GPT-Image2-SkillWhich topics is it in?openclaw-skill, qwen-image, comfyui, text-to-image, threejs, codex-skill, diffusion-models, claude-code-skill, agent-skill, openclaw, qwen, claude-skills
openclaw-skill, qwen-image, comfyui, text-to-image, threejs, codex-skill, diffusion-models, claude-code-skill, agent-skill, openclaw, qwen, claude-skills