Stable Diffusion overview in Prompt Engineering / GenAI - Model Metrics & Evaluation

Metrics & Evaluation - Stable Diffusion overview

Which metric matters for Stable Diffusion and WHY

Stable Diffusion is a model that creates images from text. To check how well it works, we look at how realistic and relevant the images are. Common metrics include FID (Fréchet Inception Distance) which measures how close the generated images are to real ones, and CLIP score which checks if the image matches the text description. These metrics matter because they tell us if the images look good and fit the text prompt.

Confusion matrix or equivalent visualization

Stable Diffusion does not use a confusion matrix because it is a generative model, not a classifier. Instead, we use visual examples and scores like FID and CLIP to evaluate quality.

Example FID scores:
Real images vs Generated images
Lower FID = better quality

Example CLIP score:
Text prompt: "A cat sitting on a chair"
Generated image matches prompt well = High CLIP score
    

Precision vs Recall tradeoff with concrete examples

For Stable Diffusion, precision means how clear and detailed the images are, while recall means how diverse and varied the images can be for the same prompt.

If the model focuses too much on precision, images look sharp but may be very similar (low diversity). If it focuses on recall, images vary a lot but may be blurry or less accurate.

Example: For a prompt "a red apple", high precision means every apple looks very realistic and red. High recall means apples might look different shapes or styles but still red.

What "good" vs "bad" metric values look like for Stable Diffusion

Good: FID below 30 means generated images are close to real images. CLIP score above 0.3 means images match text well. Images look sharp, colorful, and relevant.

Bad: FID above 100 means images look very different from real ones. CLIP score below 0.1 means images do not match the prompt. Images may be blurry, strange, or unrelated.

Common pitfalls in evaluating Stable Diffusion

• Overfitting: Model may memorize training images, producing less diverse outputs.
• Data leakage: Using test images in training can falsely improve metrics.
• Ignoring diversity: Only checking image quality but not variety can mislead about model performance.
• Misinterpreting metrics: Low FID alone does not guarantee good text-image match; use CLIP score too.

Self-check question

Your Stable Diffusion model has a FID of 25 but a CLIP score of 0.05. Is it good?

Answer: No, because while the images look realistic (low FID), they do not match the text prompts well (very low CLIP score). The model needs improvement to better understand and generate images that fit the text.

Key Result

Stable Diffusion quality is best judged by FID for image realism and CLIP score for text-image relevance.

Practice

1. What is the main purpose of Stable Diffusion in AI?

easy

A. To translate languages automatically

B. To analyze financial data

C. To create images from text descriptions

D. To detect spam emails

Solution

1. Step 1: Understand Stable Diffusion's function

   Stable Diffusion is designed to generate images based on text prompts.

2. Step 2: Compare with other options

   Other options describe different AI tasks unrelated to image generation.

3. Final Answer:

   To create images from text descriptions -> Option C

4. Quick Check:

   Stable Diffusion = image generation from text [OK]

Hint: Remember: Stable Diffusion = text to image [OK]

Common Mistakes:

• Confusing Stable Diffusion with language translation
• Thinking it analyzes data instead of creating images
• Mixing it up with spam detection tools

2. Which of the following is the correct way to give a prompt to Stable Diffusion?

easy

A. "A sunny beach with palm trees"

B. generate_image(sunny beach palm trees)

C. image.create('sunny beach')

D. createImage: sunny beach, palm trees

Solution

1. Step 1: Identify proper prompt format

   Stable Diffusion accepts text prompts as strings describing the image.

2. Step 2: Check options for correct syntax

   Only "A sunny beach with palm trees" uses a simple text string suitable as a prompt.

3. Final Answer:

   "A sunny beach with palm trees" -> Option A

4. Quick Check:

   Prompt = plain text string [OK]

Hint: Prompts are plain text descriptions in quotes [OK]

Common Mistakes:

• Using code-like syntax instead of plain text
• Omitting quotes around the prompt
• Mixing function calls with prompt text

3. Given the prompt "A cat sitting on a red chair", what kind of output should Stable Diffusion produce?

medium

A. A text description of a cat on a chair

B. An image showing a cat sitting on a red chair

C. A list of cat breeds

D. A video of a cat on a chair

Solution

1. Step 1: Understand prompt to output relation

   Stable Diffusion generates images based on text prompts.

2. Step 2: Match prompt to output type

   The prompt describes a scene; the output is an image of that scene.

3. Final Answer:

   An image showing a cat sitting on a red chair -> Option B

4. Quick Check:

   Text prompt -> image output [OK]

Hint: Text prompt means image output, not text or video [OK]

Common Mistakes:

• Expecting text output instead of image
• Confusing image generation with video creation
• Thinking it lists information instead of creating visuals

4. You gave the prompt "A futuristic cityscape at night" but the output image is blurry and unclear. What is a likely cause?

medium

A. The input text was too long

B. The model does not support night scenes

C. Stable Diffusion only creates black and white images

D. The prompt was too simple or vague

Solution

1. Step 1: Analyze prompt clarity impact

   Simple or vague prompts can cause unclear images because the model lacks detail to generate sharp visuals.

2. Step 2: Evaluate other options

   Stable Diffusion supports night scenes and color images; prompt length is not the main issue here.

3. Final Answer:

   The prompt was too simple or vague -> Option D

4. Quick Check:

   Clear prompts = better images [OK]

Hint: Use detailed prompts for clear images [OK]

Common Mistakes:

• Assuming model can't create night scenes
• Thinking Stable Diffusion only makes black and white images
• Blaming prompt length instead of prompt detail

5. You want to create an image of a "red apple on a wooden table" but the generated image shows a green apple. What should you do to fix this?

hard

A. Add more detail to the prompt like "a bright red apple on a rustic wooden table"

B. Use a shorter prompt like "apple table"

C. Change the model to one that only creates fruit images

D. Remove color words from the prompt

Solution

1. Step 1: Understand prompt specificity effect

   Adding more descriptive details helps the model focus on the correct colors and objects.

2. Step 2: Evaluate other options

   Shorter or vague prompts reduce clarity; changing models unnecessarily or removing color words won't fix the color issue.

3. Final Answer:

   Add more detail to the prompt like "a bright red apple on a rustic wooden table" -> Option A

4. Quick Check:

   Detailed prompts improve image accuracy [OK]

Hint: Make prompts detailed to get correct colors [OK]

Common Mistakes:

• Using vague or too short prompts
• Ignoring color details in the prompt
• Switching models without reason