Hollow Worn Clothing (Sensual Poses) — Open Skill

Hollow Worn Clothing (Sensual Poses)

Sensual Poses

@ExploreFly

Other

0 0 1

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3656543/2026-10-14/aswrjkv3.jpg

# Hollow Worn Clothing — Sensual Pose Generation

Generate **fashion-editorial** images where garments keep a **3D worn silhouette** with **no visible person** inside: a hollow shell shaped by chest, waist, hips, and pose. Combine that core look with **alluring adult poses** and camera angles the user asks for.

There is **no structured project state**; progress lives in chat and canvas assets.

## When to use

- User mentions invisible wearer, hollow clothes, ghost mannequin, “clothes as if worn,” empty outfit shell, or references similar art.
- User wants **sensual / sexy / alluring** **pose and composition** while the body stays **implicit** (only fabric volume).
- User provides a **reference image** (sketch, flat garment, existing hollow-clothing art) and wants refinement or new pose.
- User wants **English prompts** assembled from the vocabulary file.

## References (read on demand)

Do **not** load everything at once. When building or refining prompts, call `read_skill_file` for:

| File | Read when |
|------|-----------|
| `references/prompt-vocabulary.md` | Composing T2I/I2I prompts, choosing pose/camera/lighting blocks, negatives |

## Model channels (mandatory)

Always pass **`model` as a channel** on `generate_image`. Never use physical model names.

| Mode | Channel | When |
|------|---------|------|
| Text-to-image | `t2i` | No suitable reference image, or user wants a wholly new outfit/scene from text |
| Image-to-image | `i2i.hd.lite` | Reference provided; fast iteration, pose/lighting tweaks, moderate fidelity needs |
| Image-to-image | `i2i.hd` | Reference provided; default **quality** I2I for hollow-shell fidelity and fabric folds |
| Image-to-image | `i2i.hd.gpt` | Reference provided; complex multi-clause prompts, stubborn structure, or user explicitly asks for strongest I2I interpretation |

**I2I rules**

- Put the user’s reference (canvas asset, upload, or preset) in `input_urls` (first image = primary structure/clothing anchor).
- Keep the **hollow-worn** and **pose** intent in the prompt; I2I still needs explicit shell and camera language (see vocabulary file).
- If `list_models` shows a channel missing, fall back in order: `i2i.hd` → `i2i.hd.lite` → `i2i`, and tell the user briefly.

**T2I rules**

- Use `model='t2i'` only (no `t2i.hd*` unless user explicitly wants non-NSFW HD quality and content stays clearly non-explicit fashion editorial).

## Workflow

1. **Clarify briefly** (only if missing): garment type(s), pose mood (e.g. arch back, bent forward, hip emphasis), background (studio vs scene), orientation (`landscape` / `portrait` / `square`), and T2I vs I2I.
2. **`read_skill_file(name='hollow-worn-clothing', path='references/prompt-vocabulary.md')`** — pick blocks: core shell, pose, camera, garment, lighting, style, negatives.
3. **If reference image exists**: optional `recognize_image` to list garments, folds, and viewing angle; merge with user intent.
4. **Assemble one cohesive English prompt** (comma-separated or short sentences). Order: **core hollow effect → pose → camera → garment details → lighting → style → quality tags**. Append **negative prompt** block from vocabulary when the tool accepts a single prompt string (negatives as “avoid: …” tail) or as documented in vocabulary.
5. **Generate**
   - T2I: `generate_image(prompt=..., model='t2i', name='Hollow outfit — <short pose>', entity_type='characters', orientation=<user choice>)`
   - I2I: `generate_image(prompt=..., input_urls=[<reference>], model='i2i.hd' | 'i2i.hd.lite' | 'i2i.hd.gpt', name=..., entity_type='characters', orientation=...)`
   - Default I2I channel: **`i2i.hd`**; use **`i2i.hd.lite`** for quick variants; **`i2i.hd.gpt`** when prompt is long or structure must be preserved under heavy pose change.
6. **Reply**: summarize pose, outfit, and angle in plain language; do not paste long URLs.

For multiple variants, call `generate_image` once per variant with a distinct `name`; `batch_size` is not available.

## Content policy