引導式測驗 · 中等 等級

AI 影像生成測驗

測試擴散基礎、影像工作流程選擇、視覺約束、最終尺寸檢查、放大限制和來源感知使用。

相關引導路徑人工智慧圖像生成
問題 1 的 8

哪種學習過程描述了擴散圖像模型的特性?

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. 哪種學習過程描述了擴散圖像模型的特性?
    • They learn a denoising process
    • They retrieve a finished stock image for every prompt
    • They copy one training image pixel for pixel
    • They use only hand-written drawing rules
  2. 一項產品結合了圖像生成、編輯、升級與後製。團隊應該評估什麼?
    • Only the prompt field, since later stages cannot change the image
    • The complete workflow and final artifact
    • The model name without inspecting the output
    • Only the first generated preview
  3. 哪些細節可以在提示中協助傳達視覺目標?
    • A guarantee that every tiny feature will render correctly
    • The legal status of the model’s training data
    • Subject, composition, lighting, palette and needed empty space
    • The final output’s provenance after publication
  4. 圖表需要精確的小標籤與重複的幾何圖形。創作者應該怎麼做?
    • Assume the prompt guarantees every label is correct
    • Judge only the overall style
    • Skip review if the resolution is high
    • Inspect the details directly and rebuild unreliable elements as editable shapes or text
  5. 為什麼要在預期的顯示尺寸下檢查圖像?
    • A thumbnail can hide distortions or unreadable text that appear in the final layout
    • Every image becomes accurate when enlarged
    • It proves the depicted event occurred
    • A small preview always has more detail than the full image
  6. What does upscaling guarantee about depicted details?
    • It restores the original object’s physically correct structure
    • Nothing about accuracy; more pixels do not necessarily recover accurate detail
    • It converts every image into a vector
    • It verifies the source and rights of each object
  7. When should an illustration be labeled as synthetic?
    • Only when it is low resolution
    • Never, because realistic images are documentary evidence
    • When the context requires preventing it from being mistaken for documentary evidence
    • Only if the AI tool is named in the file name
  8. What should a team preserve to help reproduce or review a final generated asset?
    • Only the prompt, while deleting the delivered file
    • The model’s training dataset regardless of rights
    • A screenshot of the first preview only
    • The actual final asset and relevant generation settings