The Art of prompting

Prompting, particularly within the context of working with AI models, is an intricate blend of technical understanding and artistic expression. It’s an art form in itself, as it requires not only a deep comprehension of the AI model’s capabilities but also the ability to creatively convey concepts and intentions. A well-crafted prompt should be precise, succinct, and use accurate vocabulary to effectively guide the AI system in generating the desired output.

Formulating an effective prompt calls for a clear grasp of the objectives and needs of the project, combined with a proficiency in language that enables accurate expression of those ideas. It involves a certain degree of experimentation and finesse, as subtle changes in phrasing or keywords can significantly affect the output. As such, prompt crafting is a unique and valuable competency that combines technical skill, creativity, and a deep understanding of both AI systems and language.

Creating a good prompt for generating accurate architectural images requires a clear understanding of architectural terms and concepts, as well as attention to detail. Here are some tips to help create effective prompts:

1. Precise Descriptions: Use specific architectural terms when describing elements such as materials, styles, periods, or building types. Be precise when mentioning colors, textures, shapes, and sizes.
2. Contextualize the Scene: Specify the setting of your architectural subject. Is it urban, rural, or suburban? Include details about the surrounding environment, such as trees, streets, or other buildings.
3. Lighting and Atmosphere: Mention the time of day, weather conditions, and mood you want to convey. Words like « sunny », « cloudy », or « rainy » can help set the right atmosphere.
4. Viewpoint: Clarify the perspective from which you want the image to be generated. Is it an aerial view, an exterior shot, or an interior view?
5. Style References: If you have a specific architectural style in mind, include its name. Refer to existing buildings, architects, or design movements for more precise results.
6. Attention to Detail: Include small but significant elements such as people, furniture, or landscape features to create a sense of scale and realism.
7. Avoid Ambiguity: Make sure your prompt is clear and unambiguous. Avoid vague or confusing language that might lead to misinterpretation.
8. Review and Revise: Once you’ve written your prompt, review it to ensure it captures all of your requirements. Revise it as necessary to fine-tune your request.

Remember, crafting an effective prompt is an iterative process that might require some trial and error. Experiment with different phrasing and keyword combinations to get the best results.

Example

Generate a photorealistic rendering of a Bauhaus-inspired, two-story villa situated in a serene suburban neighbourhood. The villa should showcase the iconic features of the Bauhaus style, including clean lines, geometric shapes, and a minimalist aesthetic.

The façade should primarily be composed of white concrete, complemented by large glass windows arranged in a grid-like pattern. Incorporate wooden accents in the form of window frames and doorways for a touch of warmth.

Showcase the villa surrounded by a lush, well-manicured lawn with indigenous trees providing shade and greenery. Include a small garden pathway leading to the entrance, which should feature a prominent, cantilevered overhang in the style of the Bauhaus movement.

The perspective should highlight the harmony between the modernist architecture and its natural surroundings, captured under a lightly overcast sky, providing soft and diffuse lighting for the scene. Include minimal furniture and a mid-century modern car parked nearby to enhance the ambiance and give a sense of scale.

Output (Flux)

Flux is Black Forest’s (https://blackforestlabs.ai/ ) Black Forest Labs is a German start up that create one of the most effective image creator.

Output (Dall-E)

Dall-E is now integrated to Bing (https://www.bing.com/images/create )

Leonoardo (Phoenix)

Leonardo is a powerful online app, that has many options.

Using LLM (Large Language Model)

ChatGPT

As ChatGPT is the most famous LLM, but not the only one. ChatGPT can be a very helpful companion as it can describe images. This function will be very helpful to initiate a prompt.

In this example, it is asked to describe the

image:

Answer:

The image appears to be a hand-drawn architectural sketch. The sketch is highly dynamic and expressive, rendered in black ink with loose and energetic lines. It seems to depict an urban or architectural environment with:

• Foreground elements: People sitting or walking, suggesting a public space like a plaza or park.
• Central structure: A curved building or structure with multiple levels and balconies, surrounded by greenery and landscaping.
• Background details: Trees and possibly other architectural elements.
• Overall composition: A strong perspective, with vanishing lines drawing the viewer’s eye towards the central structure, creating depth.

This type of sketch is typical for conceptual architectural presentations, capturing the essence of a design rather than precise details.

For description to prompt to image

A description can help to create a prompt

(source image : https://archikino.com/2021/03/26/gentilclassique-mechantmoderne/)

Answer:

This image shows a modernist-style building with clean, geometric lines and minimalist design. The structure features:

• Facades: Smooth, light-colored walls with a combination of rectangular and circular elements, including two prominent round windows on the upper level.
• Windows: Large glass panes in black frames on the ground floor, creating a transparent and open feel.
• Garden: A carefully designed modern landscape with curving pathways, neatly trimmed shrubs, and decorative elements like a metallic fish sculpture and small planters.
• Context: The building is surrounded by greenery, likely trees, giving it a suburban or park-like setting.
• Human presence: A man in a suit is standing at the entrance, holding a piece of paper or notebook, adding a sense of scale and interaction to the scene.

The overall aesthetic is functional and artistic, evocative of mid-20th-century modernist architecture.

Certainly! Here’s a detailed prompt you can use in Flux or similar AI tools to generate an image inspired by the one provided:

« Create a photorealistic image of a modernist-style house in a suburban garden setting. The house should feature clean, geometric lines with smooth, light-colored walls. Include architectural elements such as two large circular windows on the upper level and large rectangular glass panels with black frames on the ground floor. The surrounding garden should have a modern design with curving pathways, neatly trimmed shrubs, and decorative sculptures, including a metallic fish-like piece. Place a man in a suit standing at the entrance, holding a notebook or paper, to add scale and context. The scene should have soft, natural lighting, with trees and greenery in the background, evoking a tranquil and artistic atmosphere reminiscent of mid-20th-century modernist design. »

Let me know if you need help adapting or refining this!

Here is the photorealistic rendering of a modernist-style house inspired by the reference. It reflects the clean lines, circular windows, and carefully landscaped garden, staying true to the modernist aesthetic. Let me know if you’d like any adjustments!

Other LLMs

Pi

Pi, https://pi.ai/talk, does not allows to upload images, but offers very good prompts after a good description of what is needed.

DeepSeek

DeepSeek, https://www.deepseek.com/, helps to create very accurate prompts. It’s not possible to upload images for a description, but the resulting prompts work well.

.

To be continued part06