引導式測驗 · 中等 等級

編寫有效的影像生成提示測驗

測試影像模型的提示結構,為什麼否定失敗和否定提示有效,以及提示權重、指導尺度和種子形狀如何產生。

問題 1 的 8

根據指南,在結構良好的圖像提示詞中,哪個元素通常應該放在最前面?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. 根據指南,在結構良好的圖像提示詞中,哪個元素通常應該放在最前面?
    • The aspect ratio
    • The subject and what it is doing
    • The quality tags
    • The negative prompt
  2. 為什麼「a beach with no people」(沒有人的海灘)可能會產生一張有人物的圖片?
    • Image models associate words with visual features and do not reliably process negation
    • The model deliberately ignores the second half of every prompt
    • Beaches are always generated with people by default
    • Negation only works in languages other than English
  3. 在 AUTOMATIC1111 風格的 Stable Diffusion 介面中,(red scarf:1.3) 有什麼作用?
    • Generates 1.3 scarves
    • Shrinks the scarf to 1.3 percent of the image
    • Increases the influence of the phrase "red scarf"
    • Tells the model to render 3 images
  4. 負面提示詞在無分類器引導(classifier-free guidance)下如何影響生成?
    • It deletes matching pixels after the image is finished
    • It is appended to the main prompt
    • It changes the random seed
    • It replaces the empty-prompt baseline so the image is pushed away from those concepts
  5. 在調校提示詞時,固定隨機種子(seed)的主要目的為何?
    • It makes each image generate faster
    • It keeps the starting noise the same so you can see the effect of changing one element
    • It improves the resolution of the output
    • It prevents the model from reading the negative prompt
  6. What happens to DALL-E 3 prompts typed in ChatGPT, according to the guide?
    • They are rewritten by the chat model before reaching the image model
    • They are truncated to five words
    • They are converted into Midjourney syntax
    • They are sent unchanged to the image model
  7. In Midjourney, why would you write "hot:: dog" instead of "hot dog"?
    • To generate a sausage in a bun at higher quality
    • To change the aspect ratio
    • To treat "hot" and "dog" as separate concepts, such as a dog that is hot
    • To exclude dogs from the image
  8. Which statement about quality tags like "masterpiece, 8k, best quality" matches the guide?
    • They are required for every modern model
    • They double the image resolution
    • They are the most important part of any prompt
    • They helped some older community models but do little on many modern ones