A hanfu character only reads as authentic when the prompt pins three things: the dynasty (Tang, Song, and Ming each have a distinct silhouette), the palette (Tang trends vermilion-gold, Song trends soft pastels, Ming trends jewel tones with embroidery), and a scene detail that fits the era. Plain “Chinese style” produces fusion costume from no specific period. The 12 prompts below force all three. One thing changed since 2025 that matters more than any wording trick: the model you pick. As of June 2026, native bilingual Chinese models (Qwen-Image 2.0, Seedream 5.0) parse garment tokens like 马面裙 and 褙子 and hold dynasty context far better than Western-trained models, which still drift Western on faces and read mixed Chinese tokens unevenly. For keeping one face consistent across a series, see AI consistent character images.
TL;DR
- Pin dynasty + canonical garment name + per-dynasty palette + one period scene detail in every prompt. Vague “ancient Chinese” anchors on nothing.
- Use a native Chinese model (Qwen-Image 2.0 or Seedream 5.0) for the best costume accuracy. They understand
齐胸襦裙,飞鱼服,凤冠霞帔semantically; Midjourney V7 looks great but trends Western on faces and handles Chinese tokens less reliably. - Always add an East-Asian face cue (
East Asian face structure, single-fold eyelid); diffusion models default Western otherwise. - Lock aspect ratio per use:
9:16portraits,16:9scenes,3:4formal. On Midjourney use--ar 9:16(no decimals).
Which model to use for hanfu
| Model (June 2026) | Chinese token handling | Face default | Best for |
|---|---|---|---|
| Qwen-Image 2.0 (Alibaba) | Native bilingual; reads garment names semantically | East Asian | Costume accuracy, embedded Chinese text on signage |
| Seedream 5.0 (ByteDance) | Native bilingual; strong on Hanzi and regional aesthetic | East Asian | Photoreal portraits, cultural context |
| Midjourney V7 | Mixed Chinese tokens parse unevenly | Western (needs override) | Painterly stylization, mood, lighting |
If a Western model is your only option, keep the dynasty and garment in both Chinese and pinyin (Tang dynasty qixiong ruqun 齐胸襦裙) so at least one token lands, and force the face cue every time.
Best for
- Cultural illustration and museum or heritage content
- Character design for period games and webnovels
- Book and album cover art with a specific dynasty setting
- Xiaohongshu / Douyin cosplay reference imagery
1. Tang dynasty lady (vermilion-gold)
beautiful young Chinese woman, Tang dynasty 齐胸襦裙 in vermilion red and gold, peony hairpins, full court makeup with花钿 forehead ornament, courtyard with blooming peonies, soft afternoon light, East Asian face structure, painterly Chinese aesthetic, 9:162. Song dynasty scholar (soft minimal)
young Chinese scholar in Song dynasty 褙子 robes, soft beige and pale green, holding folded fan, standing by ink-painting landscape, soft mist, minimal palette, East Asian face structure, painterly literati style, 16:93. Ming dynasty court lady (jewel tones)
Ming dynasty noblewoman, ornate 马面裙 with phoenix embroidery, jade hair ornaments, intricate updo, palace garden setting, golden hour light, East Asian face structure, painterly oil portrait style, 9:164. Tang dynasty horsewoman
Tang dynasty noblewoman on horseback, riding 胡服 in cream and crimson, leather boots, tall ponytail with gold bands, open steppe with distant mountains, dawn light, painterly Tang frescoes style, 16:95. Song dynasty market scene
Song dynasty young merchant in everyday 短褐, indigo top and grey trousers, carrying woven basket of vegetables, narrow city alley with wooden storefronts and red lanterns, overcast soft light, genre painting style, 3:26. Ming dynasty general (armored)
Ming dynasty general in lamellar armor over red 飞鱼服, helmet with red plume, holding 戚家刀 sabre, military camp at dusk, banners in background, East Asian face structure, painterly historical illustration, 9:167. Wei-Jin literati (loose flowing)
Wei-Jin period scholar in loose flowing 大袖衫, ivory robe with wide sleeves, hair in topknot with wooden 簪, drinking wine under pine tree, bamboo grove behind, ink-wash aesthetic, painterly literati style, 16:98. Qing dynasty Manchu lady
Qing dynasty Manchu noblewoman in 旗装 with 花盆底 platform shoes, embroidered with chrysanthemum motifs, 两把头 hairstyle with hairpins and flowers, palace interior with red lacquer screen, soft window light, East Asian face structure, painterly portrait style9. Tang dynasty dancer
Tang dynasty 胡旋舞 dancer, swirling 帔帛 silk scarves in red and gold, embroidered short jacket and long skirt, hair in双鬟望仙髻, palace hall with lanterns, motion blur on scarves, painterly Tang court style, 9:1610. Ming dynasty wedding portrait
Ming dynasty wedding portrait, bride in red 凤冠霞帔 with phoenix crown and embroidered shawl, groom in red official robe with rank badge, seated formally facing the viewer, red lacquer interior, candlelight, painterly traditional portrait style, 3:411. Tang dynasty Buddhist devotee
Tang dynasty female devotee in pale lavender 襦裙, holding lotus, standing before a 敦煌 mural of bodhisattvas, soft cave lighting, halo of dust motes in the air, painterly Dunhuang fresco style, 9:1612. Modern hanfu street portrait
modern young Chinese woman wearing accurately reconstructed Ming dynasty 马面裙 with simple white 立领 top, walking through old town street with grey brick walls, natural overcast light, candid documentary photo style, 35mm lens, 3:2How to refine
Name the dynasty explicitly and the canonical garment in pinyin or Chinese (Tang dynasty 齐胸襦裙, Song dynasty 褙子, Ming dynasty 马面裙). The model anchors far better on a specific garment than on a vague era. Add an East-Asian face cue (East Asian face structure, soft features, single-fold eyelid) since diffusion models trend Western by default. Lock the palette per dynasty rather than picking colors at random, and add one period-correct scene element (peony courtyard, ink landscape, palace garden) so the era is felt, not just stated. On Midjourney V7 the aspect ratio goes at the very end as --ar 9:16 and cannot contain decimals; on Qwen-Image and Seedream set it in the UI or as a size preset.
Common mistakes
- Mixing dynasties (Tang silhouette + Ming embroidery) reads as cosplay, not period
- Modern accessories slipping in (sneakers, watches, modern earrings)
- Western face proportions when the era and clothing are Chinese
- Generic palette; each dynasty has its own canonical colors
- Calling the garment “robe” instead of naming
襦裙,褙子,马面裙,旗装is too vague to anchor on
FAQ
Which AI model gives the most accurate hanfu? As of June 2026, native Chinese bilingual models are best: Qwen-Image 2.0 (Alibaba) and Seedream 5.0 (ByteDance) parse garment names semantically and default to East-Asian faces. Midjourney V7 produces the most painterly, polished mood but needs an explicit East-Asian face cue and handles Chinese tokens less reliably.
Should I write the prompt in Chinese or English? Keep the bulk in English for broad model compatibility, but write the dynasty + garment name in Chinese (or Chinese plus pinyin) because that single token does the heavy lifting on era accuracy. On native Chinese models you can write the whole prompt in Chinese.
Why do my faces keep coming out Western? Most diffusion models are English-trained and default to Western proportions. Always add East Asian face structure, single-fold eyelid and consider a native Chinese model, which defaults to East-Asian features without the override.
How do I keep the same character across many images? Lock a seed and reuse identical face wording, or use a character-reference feature (Midjourney --oref, or an image reference on Qwen/Seedream). The full workflow is in AI consistent character images.
Is hanfu a single style or many? It is an umbrella for Han Chinese dress across dynasties, each with its own silhouette: Tang 齐胸襦裙, Song 褙子, Ming 马面裙, plus Qing Manchu 旗装 (technically not hanfu but often requested). Always name the specific era and garment rather than “hanfu” alone.
Related
- Chinese aesthetic character prompts — broader Chinese aesthetic library
- Chinese ink landscape prompts — companion landscape language
- East Asian beauty portrait prompts — portrait-style variants
- AI consistent character images — character-series consistency tutorial
Tags: #Chinese-style