Skip to content
TheChatArchiveThe Chat Archive
Claudeclaude-sonnet-4-20250514· AnthropicHugging Face LMArena Human Preference 140K

If you had to make the best prompt for an image generation with the following detail :…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Metallama-3.3-70b-instruct’s answer.
  1. User

    If you had to make the best prompt for an image generation with the following detail : "On a table where only the top is visible, we can see multiple fresh juice cup on the right while the left table space is empty. Background is something natural and freshly tropical representing Madagascar's biome. The background is not blurry." What would it be?

  2. Claude Assistant

    Here's an optimized prompt for image generation:

    "A wooden table surface viewed from above, with multiple glasses of fresh tropical fruit juice arranged on the right side of the table, leaving the left side completely empty. The juices should be vibrant colors - orange, mango yellow, deep red hibiscus, and green. In the sharp, detailed background, show Madagascar's distinctive landscape: red laterite soil, iconic baobab trees with their distinctive bottle-shaped trunks, lush green rice terraces, and native ravenala (traveler's palm) trees. Include bright tropical lighting with rich, saturated colors throughout. The background should be in crisp focus, not blurred, showing the unique biodiversity of Madagascar's highlands and coastal regions. Photorealistic style with excellent lighting and composition."

    Key improvements I made:

    • Specified the table material and viewing angle for clarity
    • Added specific juice types and colors that would be authentic to Madagascar
    • Included iconic Madagascar flora (baobab trees, ravenala palms)
    • Mentioned the distinctive red soil characteristic of Madagascar
    • Emphasized "crisp focus" and "not blurred" to ensure background sharpness
    • Added lighting and style specifications for better results
    • Used more descriptive, specific language that image generators respond well to

    This should give you a much more detailed and authentic representation of what you're looking for.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.