गाइड से जुड़ी प्रश्नोत्तरी · मध्यम स्तर
छवि निर्माण प्रश्नोत्तरी में नकारात्मक संकेत
परीक्षण करता है कि कैसे नकारात्मक संकेत क्लासिफायर-मुक्त मार्गदर्शन में बिना शर्त इनपुट को प्रतिस्थापित करते हैं, जब वे मदद करते हैं, और क्यों लंबी सूचियां प्रतिकूल प्रभाव डाल सकती हैं।
प्रश्न 1 का 8
In classifier-free guidance, what does a negative prompt replace?
प्रत्येक प्रश्न का पूर्वावलोकन करें8 प्रश्न
जैसे ही आप प्रश्नोत्तरी लेते हैं उत्तर और स्पष्टीकरण दिखाई देते हैं।
- In classifier-free guidance, what does a negative prompt replace?
- The positive prompt
- यादृच्छिक बीज
- The empty unconditional prompt
- The VAE decoder
- Why might writing 'no cats' in the main prompt actually produce cats?
- Text encoders handle negation poorly, so the word 'cats' still pulls the image toward cats
- The model deliberately ignores instructions it disagrees with
- The seed already contains a cat
- Negation automatically doubles the guidance scale
- Why does classifier-free guidance roughly double the compute per step?
- It uses two seeds at once
- It decodes the image twice at the end
- It upscales the image at each step
- The model predicts noise for both the positive prompt and the negative or empty prompt
- According to the guide, which negative term is most likely to have a strong, reliable effect?
- 'bad anatomy'
- 'watermark'
- 'ugly'
- 'extra fingers'
- What happens when you use a negative prompt with a guidance-distilled model that skips the unconditional pass?
- It works twice as strongly as usual
- It is treated as a second positive prompt
- It is ignored unless the tool turns true CFG back on
- It causes the model to crash
- What is a likely downside of a very long negative prompt list?
- Less varied images with stiff compositions or distorted colors
- Faster generation
- Guaranteed correct hands
- Higher output resolution
- In ε = ε_neg + s · (ε_pos − ε_neg), what does s represent?
- The seed
- The guidance scale
- The number of steps
- The sampler type
- How is a negative prompt different from a post-processing filter?
- It is applied only after the final image is produced
- It removes objects using a hand-drawn mask
- It affects only the file's metadata
- It influences every denoising step during generation