✍️ Module 4: Prompt Engineering — The Art of Speaking to the Machine

4.1 Fundamental Principles of Prompt Engineering

The prompt is the interface between human intention and automatic generation. A good prompt not only describes what is wanted, but also how it should be, including style, composition, lighting, perspective, and emotional context.

Key elements of an effective prompt:

• Main subject: “a Siamese cat”
• Action or state: “sleeping on a silk cushion”
• Artistic style: “watercolor style, soft strokes, pastel colors”
• Composition: “close-up, blurred background, natural lighting”
• Technical quality: “high resolution, 8K, fine details, no artifacts”

Complete example:

“A Siamese cat sleeping peacefully on a blue silk cushion, watercolor style with soft strokes and pastel colors, close-up with blurred background, natural window lighting, high resolution 8K, fine details, no artifacts”

4.2 Advanced Prompting Techniques

• Negative Prompts: Words or phrases to exclude from generation. E.g.: “deformed, blurry, extra fingers, low quality”.
• Weighting: Assign weights to certain terms using (word: weight). E.g.: (Van Gogh style:1.3).
• Step Conditioning: In some pipelines, the prompt can vary across generation steps.
• Embeddings and Trigger Words: Use of custom embeddings (e.g., generated via Textual Inversion) that activate learned styles or concepts.

4.3 Common Mistakes and How to Avoid Them

• Overly vague prompts: “something beautiful” → provides insufficient information.
• Too many contradictory concepts: “realistic and cartoon at the same time” → confuses the model.
• Lack of structure: Mixing styles, actions, and qualities without hierarchy.
• Ignoring negative prompts: Results in images with artifacts or unwanted elements.