Hi @Francois_MCD ,
We have not yet been able to confirm your "next to the lake" or "in the middle of the water" issues. However, your concerns have been noted and will be considered as we work on improving the AI Visualizer from an architect's prompt perspective. As we aim to simplify the prompts, we will also evaluate adding more controls to decrease the need for very detailed and descriptive texts, ensuring better image generation outcomes.
Meanwhile, I am grateful to read that you will be conducting more tests. To get better architectural engineering responses from AI Visualizer (powered by Stable Diffusion) prompts, consider the following areas, including the use of the creativity slider:
1. Specificity and Detail
- Architectural Style: Be specific about the architectural style (e.g., modern, Gothic, Baroque, Brutalist).
- Materials: Mention materials (e.g., glass, concrete, wood) to give the model context.
- Structural Elements: Include details like arches, beams, and columns.
2. Scale and Proportion
- Size and Scale: Indicate the size and scale of the structures (e.g., a small cabin, a skyscraper).
- Proportions: Include proportions if relevant (e.g., height-to-width ratio).
3. Function and Use
- Purpose of the Building: Specify whether it's residential, commercial, industrial, etc.
- Specific Requirements: Mention any specific requirements (e.g., energy efficiency, accessibility).
4. Context and Environment
- Surrounding Environment: Describe the environment (e.g., urban, rural, waterfront).
- Landscape Elements: Include gardens, courtyards, parks, etc.
5. Design Elements
- Key Design Elements: Detail features like large windows, open floor plans, rooftop gardens.
- Aesthetic Preferences: Mention whether you prefer minimalist, ornate, etc.
6. Lighting and Atmosphere
- Desired Lighting: Indicate the desired lighting (e.g., natural light, ambient lighting).
- Atmosphere: Describe the atmosphere you want to create (e.g., cozy, futuristic).
7. Cultural and Historical Influences
- Cultural/Historical Influences: Reference influences if applicable (e.g., Renaissance architecture, traditional Japanese design).
8. Visualization Style
- Visualization Style: Specify whether you prefer photorealistic, conceptual sketches, or 3D renders.
9. Creativity Slider
- Adjust Creativity Level: Use the creativity slider (0-100%) to control the level of creativity in the output.
- 0% Creativity: The model focuses strictly on the explicit details provided in the prompt.
- 50% Creativity: The model balances adherence to the prompt with some degree of creative interpretation.
- 100% Creativity: The model prioritizes artistic and innovative interpretations, potentially taking significant liberties with the prompt.
We truly appreciate your valuable feedback regarding the prompts for image generation. We understand the importance of precision and accuracy in architectural design, and we acknowledge that our current prompts may not always meet your detailed specifications.
Going forward we intend addressing these Prompt concerns as follows:
-
Backend Prompt Engineering:
- Working on improving the backend of our system to better understand and process architectural terminology and concepts. This includes fine-tuning the AI to better grasp specific styles, materials, proportions, and functional elements that are crucial to your designs.
-
Feedback Integration:
- Collecting and analyzing all input from the architectural community to identify common issues and areas for improvement. This collaborative approach will help us refine the model to better suit your needs. Thanks a lot for your guidance on this effort π
-
Simplifying Prompts:
- One of our key objectives is to reduce the need for long, detailed prompts. We aim to enhance the AIβs ability to generate accurate images based on shorter, more concise descriptions. This will save you time and make the process more efficient.
Appreciate Improvement Process:
-
Detailed Feedback: When you encounter issues, please provide specific examples of where the prompts did not meet your expectations. Detailed feedback on what elements were missing or incorrect will help us pinpoint and address these gaps.
-
Suggestions for Improvement: If you have ideas on how the prompts could be better structured or what additional details could be included, we would love to hear them.
I hope this input will be helpful for those of you testing the AI Visualizer and will motivate you to continue providing feedback in this area. Thank you all for your assistance.
Kenneth Solvik
Director, Product Innovation and Technology, Product Success
