|
|
--- |
|
|
license: mit |
|
|
library_name: prompt-tool |
|
|
tags: |
|
|
- stable-diffusion |
|
|
- comfyui |
|
|
- sdxl |
|
|
- qwen-vl |
|
|
- prompt-extraction |
|
|
- style-extraction |
|
|
- pose-extraction |
|
|
- text-generation |
|
|
- vision-language |
|
|
- gritai |
|
|
- visionharvester |
|
|
model_index: |
|
|
- name: VisionHarvester-v1 |
|
|
results: [] |
|
|
--- |
|
|
|
|
|
|
|
|
|
|
|
# VisionHarvester v1 — Identity-Safe Image Style & Pose Extractor |
|
|
|
|
|
**By GritAI Solutions LLC** |
|
|
|
|
|
VisionHarvester v1 is a lightweight prompt-based extraction tool for creators working with Stable Diffusion, ComfyUI, SDXL, LoRA training, and Qwen-VL workflows. |
|
|
|
|
|
It converts a reference image into clean, modular, identity-safe text blocks you can reuse for character building, style replication, dataset generation, and scene reconstruction — without copying real people. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🚀 What VisionHarvester v1 Does |
|
|
|
|
|
VisionHarvester is built around five reusable components: |
|
|
|
|
|
### 1. Base Identity (Safe & Generic) |
|
|
A neutral description of the subject that includes: |
|
|
- General body type |
|
|
- Hair color and basic hairstyle |
|
|
- Clothing and fabric behavior |
|
|
- Broad, non-identifying facial description |
|
|
|
|
|
No pose, no emotion, no personality. |
|
|
|
|
|
### 2. Pose (Geometry Only) |
|
|
Short, comma-separated fragments describing: |
|
|
- Limb positions |
|
|
- Body orientation |
|
|
- Weight distribution |
|
|
- Head/hip angles |
|
|
|
|
|
No outfit, no style, no emotion. |
|
|
|
|
|
### 3. Outfit & Materials |
|
|
Details about: |
|
|
- Clothing type and cut |
|
|
- Colors |
|
|
- Fabric texture and behavior (matte, glossy, stretchy) |
|
|
- Accessories |
|
|
|
|
|
### 4. Camera & Lighting |
|
|
Information about: |
|
|
- Framing (close-up, half body, full body) |
|
|
- Camera angle / lens feel |
|
|
- Lighting direction and softness |
|
|
- Major shadows and highlights |
|
|
|
|
|
### 5. Style Tags |
|
|
Reusable tags such as: |
|
|
- studio fitness look |
|
|
- clean background |
|
|
- soft cinematic lighting |
|
|
- high-resolution texture |
|
|
|
|
|
These drop straight into Stable Diffusion prompts. |
|
|
|
|
|
--- |
|
|
|
|
|
## 📂 Included in This Repository |
|
|
|
|
|
- README.md — this documentation |
|
|
- LICENSE — MIT license |
|
|
- isionharvester_v1_extractor.prompt.txt — main extraction prompt |
|
|
- VisionHarvester-PoseStyleExtractor.json — ComfyUI workflow (optional) |
|
|
- examples/ — sample images and their extracted outputs |
|
|
|
|
|
Example files: |
|
|
- examples/example_01.png |
|
|
- examples/example_01_output.txt |
|
|
- examples/example_02.png |
|
|
- examples/example_02_output.txt |
|
|
- examples/example_03.png |
|
|
- examples/example_03_output.txt |
|
|
|
|
|
--- |
|
|
|
|
|
## 🟩 Main Extraction Prompt (v1) |
|
|
|
|
|
This is the core VisionHarvester v1 prompt shipped in isionharvester_v1_extractor.prompt.txt: |
|
|
|
|
|
\\\ |
|
|
Extract a clean, neutral description of the woman in the image. |
|
|
|
|
|
Keep it simple: |
|
|
• No pose or body positioning |
|
|
• No emotions or personality |
|
|
• No unique facial identifiers |
|
|
• No NSFW content |
|
|
• Do describe hair, body type (general), clothing, colors, fabrics, and broad facial features |
|
|
|
|
|
Output 2–4 sentences that would work as a Stable Diffusion base identity block. |
|
|
\\\ |
|
|
|
|
|
Use this in: |
|
|
- Qwen-VL custom prompt |
|
|
- Any Vision-LLM |
|
|
- ComfyUI Qwen nodes |
|
|
- Image-to-text or SD prompt pipelines |
|
|
|
|
|
--- |
|
|
|
|
|
## 🖼 Example Outputs |
|
|
|
|
|
Example of the kind of identity-safe text this prompt produces: |
|
|
|
|
|
\\\ |
|
|
An athletic woman with long dark hair, a medium tan complexion, and soft neutral facial features without distinctive identifiers. She is wearing a fitted black sports bra made from matte stretch fabric and high-waisted leggings. Her appearance is clean, simple, and suitable as a Stable Diffusion base identity. |
|
|
\\\ |
|
|
|
|
|
--- |
|
|
|
|
|
## 🔒 Identity & Safety |
|
|
|
|
|
VisionHarvester v1 is designed to: |
|
|
- Avoid 1:1 face cloning |
|
|
- Avoid unique facial markers |
|
|
- Avoid real-person or celebrity references |
|
|
- Avoid explicit or NSFW content |
|
|
|
|
|
It focuses on style, clothing, pose, and scene — not identity. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧩 Use Cases |
|
|
|
|
|
- Character consistency |
|
|
- Pose and outfit reuse |
|
|
- LoRA dataset prep |
|
|
- Style transfer |
|
|
- Scene reconstruction |
|
|
- Visual prompt creation |
|
|
- Multi-lane ComfyUI pipelines |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧱 Author |
|
|
|
|
|
**GritAI Solutions LLC** |
|
|
Robert "BonusLockSmith" Lucyk |
|
|
Lawton, Oklahoma |
|
|
|
|
|
--- |
|
|
|
|
|
MIT licensed. Free for personal and commercial use. |
|
|
|
