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.