而 Lovart 从一开始就是为创意设计服务的垂直 Agent,目标就很清楚:不是让你玩 AI 画图,而是让你像接乙方一样,把一整套设计交给它。
两者在体验上的分歧也很明显。
先说 lovart。它的正式使用的界面不复杂。右边是聊天框,左边是一块无限画布。
你直接输入prompt:
“Generate 5 Pinterest-style square graphics for ‘muji’.Themes:1 product flatlay,1 behind-the-scenes studio shot,1 customer quote,1 moodboard collage,1 tip or value graphic.Use warm, editorial color palette with pin-friendly layout.”
prompt:” Create 5 fashion/beauty-style covers using bodies that usually aren’t represented in ads: aging skin, disabled limbs, post-pregnancy belly, scars, or fat folds.
Treat each body with dignity and softness, not dramatized.
prompt:” Create mock packaging designs for fictional emotional snacks:
“Anxious Puffs,” “Happy Syrup,” “Regret Cans,” and “Stability Crackers.”
Design each package as if it were on a real shelf, but with surreal warning labels like “May cause emotional disintegration.” Visual style should be colorful, ironic, and shelf-ready. ”
最终生成出来的东西,真的像能摆进便利店里。
· 概念化表达能力
这一次,我们让它生成一组“数字灾难艺术馆”的视觉内容。把“bug”变成“展品”。
prompt:”Turn 5 everyday tech failures into visual museum exhibits.Use screenshots of errors like 404 pages,software crashes, dead pixels, or systemreboots.Each visual should present the glitch as an artifact under glass, with label tagslike ‘Exhibit 1: Frozen Memory’ or ‘404 – Lost Connection.’ Style should feel archival, poetic, and strange.”
prompt:”Create a 5-poster series titled ‘You’ ve Been Googled.’Each visual should depict a different person entirely made out of browser UI elements – search bars, autocomplete text, cookie banners, location prompts,or loading spinners. Every figure should represent a different data-driven emotion (curiosity, shame, distraction,compliance, overload). Use halftone textures, retro red backgrounds, and bold comic composition.”
prompt:” Create a 9-panel comic strip titled “The Day I Became a PNG.”
The protagonist, after being dragged into a clipboard, loses their third dimension and exists only as a transparent cutout. They get reused in memes, covered by watermarks, misidentified by AI. By the last frame, they shrink into a pixel. Draw in vintage editorial comic style.”
但实测感觉 Agent 对 prompt 的要求比一般画图类 AI 要高一点。它不太接受那种 “堆概念、罗列名词” 的提示,它需要的更像是一个 creative director 写给设计团队的 brief: