由 ByteDance 出品

Seedream 5.5 提示詞

字节跳动的生产级图像模型——长链条编辑像素级稳定、原生透明输出,光线、皮肤与文字更自然。

image
A cinematic romantic still photograph captured on a 35mm motion picture camera — a slightly low-angle medium three-quarter shot of an extraordinarily attractive, movie-star-caliber young East Asian couple embracing on a tropical beach at golden-hour sunset, capturing the joyful moment as the man effortlessly lifts the woman high into the air against a pastel pink and gold twilight sky. The strikingly handsome, tall, broad-shouldered man in his mid-20s possesses sculpted aristocratic bone structure, a sharp chiseled jawline, a high straight nose bridge, deep expressive eyes, and effortlessly styled tousled jet-black hair. Wearing an unbuttoned relaxed black linen shirt with rolled sleeves, his strong arms wrap securely around her thighs to lift her upward as he gazes up at her with a warm, captivating smile. The breathtakingly gorgeous, slender woman features an exquisite delicate side profile, graceful swan neck, luminous rosy-flushed porcelain skin, and a sculpted bare back. Wearing a lightweight white backless spaghetti-strap sundress tied with a large silk ribbon bow at her lower back, she rests her hands on his shoulders and looks down into his eyes with an enchanting, shy smile. Captured mid-motion, her lustrous dark chestnut hair and trailing white dress ribbons sweep dramatically through the sea breeze. The setting is a serene tropical coastline with soft-focus palm tree silhouettes along the shore beneath a dreamy sky blending from twilight blue into warm pink and glowing sunset gold. Strong golden backlighting wraps her flying hair, bare back, and sheer flowing fabric in radiant warm rim light, while floating sun-kissed bokeh particles shimmer in the hazy sea air. Captured with a wide-latitude cinema package and a 50mm vintage anamorphic lens at f/2.0, producing creamy oval bokeh, soft organic halation, and gentle lens flare. Flattering, fine, even real human skin with warm subsurface scattering, strand-by-strand windblown hair physics, natural linen and chiffon textures, and fine 35mm theatrical film grain — photographed on real film, never CGI, never plastic, never 3D render.
Ultra-realistic cinematic street portrait of a beautiful young woman with short, messy dark brown bob hair, natural freckles across her nose and cheeks, soft facial features, expressive hazel eyes, and a calm, slightly melancholic expression. She is wearing an oversized, chunky knitted beige wool turtleneck sweater with a thick folded collar covering part of her chin. She is gently holding a fluffy, light gray Persian cat in her arms. The cat has a round face, dense soft fur, small ears, long whiskers, and striking bright amber-orange eyes, looking directly into the camera. Environment: A cold, foggy city street at night, with a cinematic urban atmosphere. Blurred pedestrians in the background, soft glowing streetlights, and an out-of-focus orange neon sign behind her. Subtle mist creates depth and atmosphere. Composition: Vertical portrait, medium close-up framing, subject centered, the woman looking directly into the camera while holding the cat close to her chest. The cat is prominently positioned in the lower center of the frame. Photography: Shot on an 85mm portrait lens, f/1.8 aperture, shallow depth of field, extremely detailed skin texture, realistic freckles, individual hair strands, highly detailed cat fur, natural facial proportions, soft diffused lighting, cinematic film grain, muted earthy color grading, warm amber highlights mixed with cool greenish-gray shadows. Photorealistic editorial fashion photography, authentic candid street photography, atmospheric cinematic lighting, subtle vintage film aesthetic, professional photography, ultra-detailed, 8K quality. Aspect ratio: 9:16 portrait.

關於 Seedream 5.5

Seedream 5.5 是字节跳动的下一代图像模型,面向专业视觉生产,而不是一次性实验。前代 Seedream 5 靠双语提示词理解和高保真质感立住口碑;5.5 保留这套底子,把力气集中砸在真正吃时间的那一段——编辑。

核心是编辑稳定性。未改动的区域按像素保留,色偏被压制,每一轮输出都能作为下一轮输入继续改,而不会在长链条里慢慢崩坏。另外两项升级对生产同样关键:原生 RGBA 输出支持全透明与半透明——省掉抠图环节;光线、材质与皮肤更自然,密集小字的还原也更忠实。

这些能力改变了提示词该写什么。你不必每一轮都重新描述整张图,而是点明一处改动,并说清什么必须活下来:身份、光线、姿态、背景。上方画廊收录了经过验证的 Seedream 5.5 提示词及真实输出,你可以从一个经得起多轮编辑的结构起步。

如何寫好 Seedream 5.5 提示詞

  1. 1多轮编辑时,每轮只改一件事,并把上一轮输出作为参考带进来。模型天生会保留你没要求改动的部分;一次只给一个清晰的增量,它才接得住。
  2. 2明确什么必须保留:「人脸、表情、光线和背景保持原样,只改外套」。把边界说出来,好编辑才会变成可复现的编辑。
  3. 3当图片是要直接使用的素材时,要求透明背景,并描述边缘:头发、羽毛、蕾丝、玻璃、半透明。原生 RGBA 意味着 alpha 通道是生成出来的,不是事后抠的。
  4. 4密集文字要逐字引用,说明语言,并交代它在版面中的位置。Seedream 5.5 对又小又密的文字远比前代忠实,但前提是你把字原样给它。
  5. 5让角色跨场景出现时,先列出定义他的特征——脸型、发型、标志性服装——再说明哪些可以随姿态变化。把不变量写清楚,角色参考才稳。
  6. 6第一条提示词按「主体 → 材质与质感 → 光线 → 构图 → 风格」来搭。具体名词胜过形容词,第一帧够强,后面就能少走几轮。

常見問題

Seedream 5.5 最擅长什么?

长链条编辑。未改动区域保持像素级稳定、色偏被压制,你可以对同一张图连续精修很多轮,而不必推倒重来。它还能直接输出原生透明素材,密集文字的还原也更忠实。

Seedream 5.5 和 Seedream 5 有什么区别?

升级集中在编辑稳定性、原生 RGBA 透明和真实感——光线、材质、皮肤与密集小字。如果你的工作是一次性生成,差距没那么大;如果你反复打磨同一份素材,这是「能用的链条」和「越改越飘」的区别。

Seedream 5.5 支持透明背景吗?

支持。它一次生成就产出带全透明与半透明的 RGBA 图像,头发、羽毛、蕾丝这类精细边缘也能处理——拿到的是可直接使用的素材,不需要另做抠图。

在哪里可以试用这些 Seedream 5.5 提示词?

Seedream 5.5 可在字节跳动的即梦(Dreamina)和豆包中使用,开发者可走火山引擎 API。从本页复制任意提示词粘贴过去,替换成你自己的主体或参考图即可。