Claude Sonnet 4.6 图像提示生成对比分析

Claude Sonnet 4.6 is way better than I expected. In my workflow, I used to rely on Gemini to reverse-engineer prompts from specific images and generate JSON prompts. But with both Gemini and ChatGPT, I usually had to manually tweak the prompts quite a bit before getting the results I wanted. Today I tried Claude Sonnet 4.6 for image-to-prompt reverse engineering, and the results were surprisingly good. The prompts were immediately usable, and the quality felt very consistent. A lot of people have recommended Claude Sonnet as a replacement for Gemini/GPT, but I couldn’t use it due to regional restrictions. Huge thanks to @ZenMuxAI for launching Claude Sonnet 4.6 yesterday (and making it free to use). I tested it right away—amazing work. Image comparison: Left: generated from Claude Sonnet 4.6’s prompt. I love the perspective here—plus the composition feels more dynamic: the characters in the top and bottom halves are actually facing each other like they’re in an intense match, battling it out. Right: generated from Gemini’s prompt. The main character is larger, but you can see most characters are oriented in the same direction, so the scene feels less “interactive.” I’ll drop both prompts in the comments—click to view if you need them.

Prompt 拆解