首页
资产
我的作品
模型
音色库
开发者
博客
简体中文
价格
登录
AI 生图
首页图片工具AI 生图

图片工具 · Nano Banana · GPT-Image-2 · Seedream 5.0 Pro

AI 生图

一句话生成图片,哪不对改哪

将文件拖放到此处以上传

仅支持 .jpg, .jpeg, .png, .webp 文件

HOW IT WORKS

如何生成与编辑图片

写下想要的画面,或者放张参考图;改的时候只动你框住的地方,别处保持原样。

  1. 01

    先挑模型

    Nano Banana、GPT-Image-2、Seedream 5.0 Pro 都在同一个编辑器里,同一句提示词想换个模型再试,在这里切一下就好。

  2. 02

    说清楚,或者给它看

    写明主体是什么、要什么质感;光靠文字说不准的,放最多 5 张参考图。

  3. 03

    一次只改一块

    框住不对的地方,说清楚应该是什么样。画面其余部分不动,修一处不会把已经满意的地方一起弄没。

  4. 04

    在版本之间来回退

    每次生成、每次修改都单独存一版。可以互相对比、退回任意一版,并下载原图。

查看更多示例
A vintage educational wall-chart infographic titled 'THE SOLAR SYSTEM — A COMPARATIVE GUIDE' at the top in stately serif caps. Eight planets — Mercury, Venus, Earth, Mars, Jupiter, Saturn, Uranus, Neptune — rendered in painterly detail in a horizontal row at approximate relative scale. A labeled callout box below each planet lists: planet name in bold, diameter in km, distance from Sun in million km, number of moons, surface temperature range, one curious fact in italic. Top-right inset: the Sun with its data. Bottom: compact legend, distance ruler, attribution 'Prepared by the Interplanetary Survey Office · Public Domain · Revised 2026'. Aged cream background, thin hairline rules, serif body with sans display headings. Retro aerospace science aesthetic. All numbers and labels perfectly legible.A square 4-panel black-and-white Japanese-style comic strip titled '深夜のデバッグ' (Late Night Debugging). Screentone shading, clean inked lines, uniform panel gutters. Panel 1 top-left: a tired young engineer at a desk, a wall clock shows 2:47 AM, speech bubble in Japanese 'もう少しで終わる…'. Panel 2 top-right: extreme close-up of a monitor showing code with a red error 'SyntaxError: Unexpected token', big sound effect 'ガーン!!' in jagged kana. Panel 3 bottom-left: the character slumped in despair, thought bubble 'セミコロン…忘れた…'. Panel 4 bottom-right: the character smiling, monitor now green 'All tests passed OK', narration box at top '...そして朝が来た。', sound effect 'キラーン☆'. Classic clean comic style, high contrast. All Japanese rendered accurately.A vertical chalkboard menu in a cozy cafe, shot straight on. White and pastel chalk handwriting on a dark slate board with chalk dust. Header 'LUMINARA CAFE — TODAY'S MENU'. Four sections with headings: ESPRESSO BAR listing Espresso 2.50, Cappuccino 3.50, Latte 4.00, Flat White 4.00, Cortado 3.80. SPECIALTY DRINKS listing Iced Vanilla Latte 4.50, Mocha Hazelnut 5.50, Caramel Macchiato 5.00. SWEETS listing Classic Tiramisu 6.50, Vanilla Cheesecake 5.50, Chocolate Brownie 4.00. SANDWICHES listing Ham and Cheese 8.50, Tomato Mozzarella 7.50, Tuna Melt 8.00. Small chalk doodles of coffee beans, a steaming cup, and a croissant in the margins. Warm, imperfect, human handwritten feel. Every item and price clearly legible.
Q&A

AI 图片生成常见问题

01

可以用哪些模型?

Nano Banana、GPT-Image-2 和 Seedream 5.0 Pro,在同一个编辑器里随时切换,各自出的结果并排留在版本历史里,方便直接对比。

02

能只改图片的一部分,而不是整张重新生成吗?

能,这正是这个编辑器的重点。框住一块、说清楚要改成什么,只有框内会被重新生成。这次修改会存成新的一版,不覆盖上一版。

03

AI 图片生成器是什么?

把一段文字描述,或者你上传的一张照片,变成一张新图。你负责说清楚要什么,画和拍都交给它。

04

能拿自己的照片当底图吗?

可以。最多上传 5 张参考图,支持 JPG、PNG、WebP,生成时会照着它们走。

05

能出多大、一次几张?

从 1:1 到 21:9 的多种画幅;三档质量;一次可以出 1 张、2 张或 4 张。

06
A portrait 6-panel comic book page in a vintage action-adventure style: bold inked outlines, dotted screentone shading, vibrant flat colors. A jagged yellow caption banner at the top reads 'CHAPTER ONE: THE AWAKENING'. Panel 1 wide top: a thunderstorm over a stylized futuristic city skyline, caption box 'METRO CITY — 3:17 AM'. Panel 2: close-up of glowing eyes opening, sound effect 'KRAKOOM!!'. Panel 3: a cloaked hero leaping between rooftops, speech bubble 'I can feel them... every heartbeat.' Panel 4: a mysterious figure in shadow, caption box 'But he was not alone tonight...'. Panel 5: lightning arcing between them, sound effect 'ZAAKT!!'. Panel 6 wide bottom: both facing each other in the rain, narration box 'And so it begins.' Proper speech-balloon tails, hand-drawn sound effects, dynamic panel borders.
A 2:3 glossy fashion magazine cover, photorealistic. Masthead at top in massive tall red serif: 'MODE'. Dateline 'MAY 2026 · $9.99'. Cover subject: a striking high-fashion model with avant-garde makeup wearing architectural haute-couture, direct confident gaze, studio lighting. Cover lines arranged around the model in varied sizes: large left 'THE FUTURE OF FASHION: 50 DESIGNERS RESHAPING 2026', mid right 'IN CONVERSATION WITH DIRECTOR AYA MORI', bottom left 'INSIDE PARIS HAUTE COUTURE WEEK', tiny teaser 'Plus: The Zero-Waste Closet · p. 74'. Barcode in bottom-left '04 > 7 25274 00123 5'. Issue number top-right 'Vol. 178 · No. 5'. Crisp commercial editorial polish.
A 2:3 portrait layered paper-cut craft poster, warm natural light with soft shadows between paper layers. Composition: overlapping silhouettes of a tree and a minimalist city skyline cut from rich forest-green and cream cardstock, a rising sun in terracotta orange behind. Large bold modernist slogan wrapping the composition: 'LESS, BUT BETTER'. Smaller justified body paragraph below: 'A manifesto for conscious living. Choose fewer things. Choose them well. Let what remains matter more. — Published by the Sustainable Futures Initiative, 2026.' Bottom row of three small icons with captions: 'REDUCE · REUSE · RESTORE'. Palette: forest green, cream, terracotta, soft white. Visible paper grain texture. All text razor-sharp.

生成的图能商用吗?

个人和商业用途都可以,具体以你所在套餐的最新条款为准。

可以用来做什么
07

只修那一处不对的地方

整张图都挺好,只有手、字或颜色不对——不用重新生成,框住那块改掉。

08

封面和缩略图

视频、播客、帖子要的 16:9 或 1:1 配图,按最终展示的画幅生成,省掉事后裁切。

09

给已有照片换个样子

把自己的图当参考传进来,保留原来的构图,换成另一种风格。

10

同一个想法比几个模型

同一句提示词换个模型再跑一次,两边结果并排留在版本历史里对比。

ListenHub

一站式 AI 内容创作平台,从想法到音频、图片、PPT 和视频。

音频
AI 播客文字转语音多角色配音音频转文字音色克隆AI 声音
视频
AI 视频解说视频口型同步广告大片营销成片动作模仿
图片
AI 生图PPT
资源
定价发现音色库博客API 文档MCP Server 指南Agent Skills 指南
公司
关于我们加入我们联系我们服务条款隐私政策
© 2026 MarsWave