就今天,看到阿里发布了千问新一代图像生成模型Qwen-Image-3.0。
还记得之前的2.0版本,在文字处理上就很有惊喜。

这一次,据说模型在指令控制和生成画面上的表现,更进一步。
官方介绍说,升级后的Qwen-Image-3.0优势在于能精准把控生成文字,生成高质量图片、长信息图、复杂UI界面,服务真实生产力场景。
我自己用下来,感觉这是国内处理中文内容效果最好的,而且是信息密度最高、文字层次最丰富的图像模型。
带大家一起来看看质感。
01汉字效果最好,没有之一
使用方法非常简单,在阿里云百炼、千问AI平台上申请API通过后,或者打开千问APP都可以体验到。
我这里是从Qwen Studio上体验的。

这一代的Qwen-Image-3.0模型,支持最大 4.5k token 的输入,对长文本、复杂信息和多模块内容的理解能力明显长了不少。
当描述长度增加,我们能直接写出完整结构、排版要求、字体规范、颜色规则、布局逻辑甚至数据内容,这就让图像生成,逐渐接近了结构化的设计工具。

提示词:
瘦金体书法作品:包含下面所有文字:壬戌之秋,七月既望,苏子与客泛舟游于赤壁之下。清风徐来,水波不兴。举酒属客,诵明月之诗,歌窈窕之章。少焉,月出于东山之上,徘徊于斗牛之间。

像我截取的《赤壁赋》里的这一段,用“瘦金体”书法作品呈现出来,一笔一画都法度严谨,就是略显完美刻意了些。
到这里,可能有朋友觉得,似乎和2.0版本,或者跟最强海外模型比,差别没有很大,没什么惊喜。
那不妨看看下面这个,我让模仿米芾的,落款就是仿米芾。
这个提示词我打磨了很久,从做旧风格到气息生动到墨色浓淡,给了不少细节。
效果还是很有惊喜的:

提示词:
Createa single finished image ofa Chinese calligraphy work asa contemporary homage tothe high Song -dynasty literati energy associated withMi Fu, but unmistakably notan actual historical artifact, nota forged antiquity, andnota literal copyofanyknown surviving work. The image should look likea museum -grade photographic scan orcareful archival photograph ofa realhanging -scroll-likesheet: fullvertical sheet visible ortightly cropped tothe paper edge, neutral unobtrusive background, naturallighting, noframe, nolabel, nomodern graphic design. The paper iswarm aged xuan -paper -likematerial ina light tan tohoney -buff tone withsubtle fibers, soft uneven aging, faint handling wear, believable absorbency, andrestrained authenticity; oldandarchival, but nottheatrically damaged, nocracks, nofantasy parchment styling. Ink behavior must be exceptionally convincing andphysically plausible: deep black tosoftened gray -black edges, pressure settling intothe paper, visible entry /travel /exit ofthe brush, line width variation, dry -brush drag, occasional flying -white, feathered absorption, resistance atturns, pivot andre -acceleration, micro -hesitations before somedirectional changes, somestrokes swift andblade -like, others swollen andweight -bearing. The overall hand issemi -cursive dominant withcontrolled cursive transitions, fast but cultivated, tensile, slanted, forceful yet refined, unstable ina controlled way, withstrong literati intelligence andrealcalligraphic structure ineverycharacter.
The composition must be four vertical text columns, read top -to-bottom withineachcolumnandright-to-leftacross the sheet, withbreathing space, controlled asymmetry, slight naturalwaviness inalignment, uneven but intentional spacing, shifting charactersizes, uneven but disciplined line density, compression followed byrelease, anda strong written flow rather than a typeset layout. Everycharactermust remain individually recognizable andgrounded inrealChinese calligraphy; preserve radical logic, internal balance, andreversible structure tostandard inprinciple. Avoid abstraction that destroys readability. The work should feel scholarly, understated, emotionally open, forceful, breathable, anddeeply human inbrush energy, conveying confidence, risk, restraint, flow, cultivated instability, andrecovery. Use very few restrained cinnabar -red seals, oneortwo atmost, placed plausibly inside orcorner positions withliterati restraint, muted naturalred, never dominant.
Render the exact text withnosubstitutions, noomissions, noadditions, noreordered characters, nomerged characters, nocorruption. The four columns must be exactly:
Column1:山雨新歇江气微凉
Column2:展卷对坐心手俱适
Column3:近得古帖数种笔意奇宕
Column4:因临数行觉胸次稍开
An optional smaller side inion orcolophon may appear ina restrained secondary hand andmust read exactly: **90=
月敬仿米芾笔意
The stylistic energy should synthesize the spirit ofSong literati calligraphy: slanted momentum, elastic brush travel, abrupt changes ofspeed, uneven internal tension, compressed structure resolving intoopenness, strong top -to-bottom momentum, andcharacter-to-characterdiversity. Somecharacters should feel airborne andswift, somebriefly anchored andheavy, andinternal timing should feel discovered during writing rather than precomputed. The resultshould be a technically serious, historically sensitive, contemporary Al -era celebration ofChinese calligraphic intelligence, photographed asa realobject.
Negative prompt: unreadable pseudo -Chinese glyphs, malformed characters, missing characters, extra characters, wrong order, modern font look, logo -likecalligraphy, perfect vector smoothness, repeated stroke templates, rubbery curves, mirrored stroke forms, decorative swirls, chaotic illegible cursive, graphic poster layout, wallpaper aesthetie, bright synthetic red seals, excessive seals, fake museum label, overdone antique distress, fantasy parchment, cartoon ink edges, flat digital brush texture, copied famous original composition, ornamental filler calligraphyte

提示词:
一幅中国传统书法横轴长卷作品。画面包含强劲有力的黑色毛笔书法字体,为行草书风格。文字在带有岁月沧桑感的浅褐色复古宣纸/羊皮纸上从右至左竖排书写,纸张带有老旧纹理、墨渍痕迹、微微的水渍和风化边缘。左下角有一枚小巧的红色传统中国篆刻印章。超宽全景视角,真实的历史古朴质感,极致细节,大师杰作
米芾的字体,飘逸俊秀,也不在话下。
应用的场景也很多,比如书法体展示、文学历史教程PPT等等。
以后我的社媒封面,也终于可以有好看的中文书法字体了。
我在想,有没有可能把书法汉字,和古风仙侠海报相结合呢?试了一下。
Automatically generate a high-aesthetic "Silhouette Universe / Collector's Edition Narrative Poster"based on [theme: xxx]. Do notdefault to common containers such asbottles, hourglasses, glass domes, orpocket watches. Instead, let the AI choose the most symbolic, visually strong, andnarratively suitable outer silhouette forthe theme. The silhouette can be an artifact, building, gate, tower, archway, dome, stairwell, corridor, statue, profile, eye, hand, skull, wing, mask, mirror, throne, ring, crack, light curtain, shadow, geometric form, spatial cross-section, stage frame, abstract symbol, oranymore creative contour that best represents the theme.
The goal isnotto simply place a world inside an object, but to make the entire themed universe grow naturally within, around, andthrough the silhouette itself. The silhouette should be elegant, recognizable, andcompositionally dominant. Inside oralong its boundary, generate a rich andlayered narrative world strongly tied to the theme, including iconic scenes, key architecture orspaces, symbols andmetaphors, traces of characters orcivilizations, foreground-midground-background depth, andatmosphere withemotional tension. Elements such asdoors, staircases, bridges, water, smoke, paths, light sources, ruins, mechanical structures, landscapes, abstract forms, creatures, orprops should allfeel unified andorganically integrated rather than pasted together.
The final composition should feel like a premium collecto r's poster with strong design value. The large shapes should feel stable, the main silhouette should be unmistakable, and the inner world should have depth, structure, and breathing room. Details should be rich but not crowded. You may add small human silhouettes, distant buildings, beams of light, doorways, bridges, steps, corridors, reflections, skylight, or far-background structures to enhance scale, story, and epic atmosphere. The overall mood should be quiet, grand, refined, and lingering rather than noisy or overloaded.
Blend the feeling of a collector's edition film poster, high