How to Create and Optimize Custom AI Prompts for ArtushVision AI

← Back to ArtushVision AI Home

Creating the perfect AI prompt for generating stock photography metadata can be challenging. Fortunately, you don’t have to write it from scratch! You can use advanced chatbots like ChatGPT, Claude, or Gemini to act as your personal “Prompt Engineer.”

By providing the chatbot with our default baseline template, a list of available variables, and your specific photography niche, the AI will generate a highly optimized prompt tailored specifically for your workflow.

The “Meta-Prompt” Strategy for AI Image Analysis (Copy & Paste)

Copy the entire blockquote below, fill in your specific requirements at the bottom, and paste it into your favorite chatbot.

Role: Act as an expert AI Prompt Engineer specializing in microstock photography metadata and SEO optimization.

Context: I am using a desktop application called ArtushVision to automatically analyze my photos and generate titles, descriptions, and 50 keywords. The application uses a System Prompt to instruct the AI (like Gemini, Claude, or local Ollama models) on how to analyze the image and return a strict JSON object.

Available Variables: The application dynamically injects real data into the prompt using the following {variables}. You must strategically place them in the prompt you create for me:

{user_hint}: A custom hint I provide before analysis (e.g., specific animal species, location name).

{filename}, {folder_context}: File and folder names for context.

{date_info}: The capture date.

{city}, {country}, {loc_hint}: Geographic data extracted automatically from GPS.

{camera_model}, {lens_hint}, {exposure_info}, {aspect_ratio}, {flash_used}: Technical EXIF data (useful for generating photography-specific tags like “long exposure”, “bokeh”, “drone photography”).

{allowed_categories}: A list of strict categories the AI must choose from.

{local_vision_text}: (Only for two-step Local/Hybrid models) A raw text description of the image generated in Phase 1.

The Default Baseline Prompt:
Analyze the image and generate stock photography metadata. Return ONLY a valid JSON object.

Context:
Date: {date_info}
Location: {loc_hint} ({city}, {country})
User Hint: {user_hint}
Technical: {camera_model}, {exposure_info}, {aspect_ratio}

Instructions:
1. "title": A short, descriptive title (STRICT LIMIT: MAXIMUM 200 CHARACTERS).
2. "description": Describe the visual content. You MUST include the location ({loc_hint}) and the date ({date_info}). STRICT LIMIT: MAXIMUM 200 CHARACTERS.
3. "keywords": Generate EXACTLY 50 unique keywords. To reach 50, you MUST categorize your thinking: include 10 literal objects, 10 colors/lighting terms, 10 background elements, 10 location/nature terms, and 10 abstract concepts/emotions. DO NOT STOP early.

JSON Structure:
{"title": "...", "description": "...", "keywords": ["...", "..."]}
My Request: Please rewrite and optimize this prompt for my specific use case.

1. My Niche/Goal: [INSERT YOUR NICHE HERE - e.g., “I shoot high-end culinary and food photography. I need keywords focused on taste, ingredients, dietary trends (vegan, keto), and lighting/mood.”] 2. Target Engine: [INSERT ENGINE TYPE - e.g., “Cloud AI” OR “Local/Hybrid AI”] (Note to chatbot: If the target is Cloud AI, provide one unified prompt. If the target is Local/Hybrid AI, I need TWO prompts: Phase 1 (Vision) asking only for a highly detailed raw text description of the image, and Phase 2 (Text) asking to format {local_vision_text} and the other variables into the final JSON).

Return the optimized prompt(s) ready to be copy-pasted into the software. Ensure the strict JSON formatting instructions remain intact!

Understanding AI Models: Cloud AI vs. Local AI (Ollama)

When generating your prompt, it’s crucial to tell the chatbot which AI engine you are using, as their architecture differs:

1. Cloud AI (OpenRouter / Native APIs)

Cloud models (like Gemini 2.5, Claude 3.5 Sonnet) are massive and incredibly smart. They can “look” at the image and format it into a perfect JSON structure all in a single step.

You need: 1 unified prompt.

2. Local AI & Hybrid AI (Ollama)

Small local models (running on your GPU) often struggle to both accurately describe an image and format complex JSON simultaneously. ArtushVision solves this by splitting the task into two phases:

Phase 1 (Vision Prompt): The model only acts as an “eye”. It looks at the image and writes a massive, chaotic paragraph of raw text describing everything it sees. Do not ask for JSON here.
Phase 2 (Text Prompt): A language model takes that raw text (injected via the {local_vision_text} variable) alongside your GPS/EXIF data, and formats it into the clean, 50-keyword JSON object.

Implementing Your Custom AI Prompt in ArtushVision AI

Once ChatGPT/Claude/Gemini generates your optimized prompt:

Open ArtushVision AI.
Click the ⚙ Profile Editor (gear icon) in the top toolbar.
Click the ➕ New button to create a blank profile.
Paste the generated prompt into the System Prompt text area (if using Local/Hybrid, paste the Phase 1 and Phase 2 prompts into their respective fields).
Ensure the AI Model field points to your desired model (e.g., google/gemini-2.5-flash-lite for Cloud, or qwen2.5-vl:3b for Local).
Click Save and give your new profile a descriptive name (e.g., Food_Photography_Pro.json).

[Get Started Now]

Need Help?

Search the documentation pages directly or jump back to the main Complete Documentation Index.

← Back to ArtushVision AI Home

⭐ User Reviews & Testimonials

❓ Frequently Asked Questions (FAQ)

💬 Support, Bugs & Community Forum

ArtushVision AI - Stability and precision for professional photography workflows.