Seedream 5 Pro

面向精确编辑、图层感知构图、多语言文字和密集视觉版式的设计智能旗舰模型。

免费试用 Seedream 5 Proarrow_forward
Seedream 5 Pro capability hero — a scene separating into clean translucent layers with precise selection outlines.

Seedream 5 Pro 最擅长什么

精准编辑

使用点选、套索或草图引导针对性的编辑,同时留意图像周围的上下文。用于最终构图前,请检查结果。

图层分离

利用图层分离识别并隔离主体、背景、文字及其他元素,将其纳入结构化修改流程。

原生多语言文字

ByteDance 表示,Pro 支持在图像中使用 10+ 种语言生成文字,包括非拉丁文字。我们的一次生成海报记录了成功之处和错误,因此发布前请校对生成的文字。

密集排版与信息图

规划并渲染复杂的信息版面——图表、示意图、多面板构图、信息密集的海报——它们保持连贯清晰,而不会散作噪点。

五条提示词,五个首轮输出

这五条提示词已固定在页面清单中。每张卡片展示其原始提示词的首个经审核输出,说明仅限于该文件中可见的内容。

午夜之后

一项竖幅虚构时尚企划,场景位于午夜的高架列车下方。

Seedream 5 Pro9:16

一位剃光头、呈成人外观的时尚模特穿着镀铬肩甲和黑色连体衣,站在高架铁路结构下被红青灯光照亮的湿街上。

首个输出:呈成人外观的主体、剃光的头部、雀斑、眉部疤痕、镀铬肩甲、红青双色的湿街照明,以及两只形态自然的手都清晰可见。不透明连体衣在曳地薄纱下遮住胸部和骨盆,但高架结构下看不到列车车厢。

原始提示词

A vertical 9:16 luxury fashion campaign for a fictional independent label. One fictional adult model, age 29, full body, standing beneath a rain-slick elevated train at midnight. She has a shaved head, visible freckles, a small healed eyebrow scar, and wears sculptural chrome shoulder armor over an opaque ink-black silk bodysuit with a long translucent organza train; chest and pelvis fully covered. Both hands fully visible with natural anatomy. Wet pavement reflects deep crimson signage and cold cyan streetlight. 50mm editorial photography, believable skin texture, restrained film grain, confident direct gaze, high-fashion composition, no legible text, no logos, no celebrity likeness, no explicit nudity.

未来之肤

一幅方形艺术肖像,以半透明树脂、欧根纱、身体彩绘和金箔构成。

Seedream 5 Pro1:1

一位带有钴蓝色遮盖、呈成人外观的主体侧身位于波纹树脂和翻卷的象牙色欧根纱后方,背部布有金色轮廓线。

首个输出:一位呈成人外观的主体侧身位于波纹树脂和飘动的象牙色欧根纱后方,带有钴蓝色遮盖、金色轮廓线,以及一只五指清晰可见的手。金色呈现为精细线条而非真正的金箔,第二只手也隐约可见。

原始提示词

A square 1:1 contemporary gallery portrait of one fictional androgynous adult dancer, age 34, posed behind rippled translucent resin and suspended sheets of ivory organza. Bare shoulders, upper back, and legs are visible; opaque cobalt body paint and sculptural fabric panels fully cover the chest and pelvis. Thin gold leaf traces the shoulder blades like a topographic map. Side profile, calm expression, one hand pressed against the resin with five natural fingers visible. Museum-grade fine-art photography, soft hard-edged shadows, mineral palette, tactile skin and fabric detail, mature and provocative but non-sexual, no visible nipples or genitals, no text, no logos, no celebrity likeness.

ORBITAL / 07

一块横幅虚构腕表发布板,包含分解组件和指定排版文字。

Seedream 5 Pro16:9

一块黑铜配色的 ORBITAL 07 腕表发布板,准确的标题和规格文字位于标有核心、表圈和表带的组件旁。

首个输出:所有要求的主标题、副标题、组件标签和规格文字,在清晰的黑铜配色发布板上都准确且清晰可读。装配完成的腕表表壳侧面有三个控制件,显得异常繁复,但产品与分解组件的层级关系仍然连贯。

原始提示词

A premium 16:9 product-launch board for a fictional modular wearable named "ORBITAL / 07". Exact headline: "ORBITAL / 07". Exact subhead: "ADAPTIVE FIELD WATCH". Center a photoreal black-ceramic watch with a luminous copper inner ring. To the right, show three clean exploded components labeled exactly "01 CORE", "02 BEZEL", and "03 STRAP". Along the bottom, include one compact specification strip with the exact text "72H POWER · 10 ATM · TITANIUM" and "AVAILABLE 07.18". Matte obsidian background, burnished copper accents, strict Swiss editorial grid, generous negative space, crisp small text, no logos other than the fictional ORBITAL / 07 wordmark.

噪声大教堂

一个方形专辑封面概念,包含生物机械大教堂、闪电和克制的标题排版。

Seedream 5 Pro1:1

一张标题为 CATHEDRAL OF NOISE 的专辑封面展示了白色闪电下的黑曜石与青铜大教堂,入口处有一个微小的红色人物。

首个输出:准确的标题和艺术家名称围绕着白色闪电下的黑曜石与青铜大教堂,微小的红色人物提供了尺度参照。形态保持为建筑结构而非解剖结构,没有血腥内容,也没有借用其他作品的角色。

原始提示词

A square 1:1 album cover for a fictional experimental record. Exact title: "CATHEDRAL OF NOISE". Exact artist name: "VANTA CHOIR". A colossal biomechanical cathedral rises from a black salt flat during an electrical storm; its arches are made from polished obsidian, bronze resonators, and translucent membranes, architectural rather than anatomical. One tiny red-cloaked figure stands at the entrance for scale. Integrate the title as restrained monumental typography across the upper third and the artist name as a small line at the bottom. Radical contemporary album design, deep blacks, oxidized bronze, violent white lightning, tactile print grain, no gore, no existing logos, no copyrighted characters.

潮汐之城 2089

一个横幅建筑概念:围绕潮汐涡轮建造的气候适应型漂浮城区。

Seedream 5 Pro16:9

一座被风暴环绕的漂浮城区围绕发光的潮汐涡轮,前方是湿润的花园步道、骑行者和两个身穿橙色服装的人物。

首个输出:一台发光的潮汐涡轮位于分层漂浮城区的中心,周围有温室桥、花园、骑行者、湿润倒影,以及恰好两个身穿橙色服装的人物。多个环形建筑带和水带让所要求的三环数量显得不那么直观,而剖面系统只能在小尺度下看见。

原始提示词

A cinematic 16:9 architectural visualization titled only by its visual concept, with no written text: a climate-adaptive floating district in the year 2089, built as three concentric inhabited rings around a luminous tidal turbine. Foreground: a wet public promenade with gardens, cyclists, and two adult engineers in orange pressure suits seen from behind. Midground: a transparent greenhouse bridge crossing left to right between timber towers. Background: a storm wall over the ocean split by one shaft of warm sunlight. Include a coherent cutaway through one ring showing homes, transit, water storage, and public gardens. Physically believable scale, weather, reflections, and structural connections; premium film and architecture concept art; no text, no logos.

一张图像,多种语言

ByteDance 表示,该模型支持在图像中渲染 10+ 种语言的文字,包括从右向左书写的阿拉伯文。这项更广泛的厂商声明并未由这一次生成验证;下文只报告我们生成并检查的六行海报。

2026-07-16,我们使用 Seedream 5 Pro 以 2048×2048 只生成了这张海报一次。海报用六种语言呈现同一句问候;我们没有为了提高质量而重新生成。

这一次实际生成了什么

6 行中有 4 行完全准确:英语、简体中文、西班牙语和法语。阿拉伯文字母正确连写并按从右到左排列,但其中的拉丁字母品牌显示为‘Studio Seedream’,且标点位置错误。印地语将 में 错误渲染为 मे——缺少随韵点(anusvara)——但可见的 स्व 合字仍然清晰可辨。

这只是一次有日期记录的提示词和输出,并不能证明普遍的语言准确性。请把图像内生成的文字视为待校对的初稿。

一次生成的多语言海报:英语、中文、西班牙语和法语准确;阿拉伯语颠倒了拉丁字母品牌的顺序,印地语的 में 缺少随韵点。

五项编辑,直接对比

拖动每个对比滑块,查看获准使用的原图与结果。每张卡片都包含注册表中的原始提示词,以及仅针对可见内容的有限说明。

睁开闭着的眼睛

对一张公园情侣肖像进行的局部修改。

编辑后:公园肖像中女子睁开眼睛编辑前:公园里的情侣肖像,女子闭着眼睛
编辑前
编辑后
drag_handle

女子的眼睛已睁开,而情侣与公园在视觉上仍保持可辨识的一致性。

这项视觉对比并不能证明身份在数学意义上保持不变,也不能证明每处细节都未改变。

原始提示词

In the uploaded group photo, detect every person whose eyes are fully or partially closed and open only those eyes. Preserve each person's original eye shape, eyelid crease, iris color, pupil size, gaze direction, and the lighting on their face. Keep all facial structure, expression, skin texture, pores, hairstyle, head tilt, and camera angle exactly as they are. Do not remove eyeglasses, reflections, or shadows. Do not add, remove, or reposition any person. Do not enlarge, brighten, or sharpen the eyes beyond what matches the rest of the face. The final image must look like a single real photograph taken at the same moment, with everyone's eyes naturally open and consistent with the original scene

移除背景人群

一张街头照片,主体身后是拥挤的人行横道。

编辑后:女子站在合成的空人行横道前编辑前:一名女子站在拥挤的街道人行横道前
编辑前
编辑后
drag_handle

人群被移除,并在原位置合成了一个看起来合理的空人行横道。

重建的人行横道是生成式解释,并非对被遮挡背景的还原。

原始提示词

Remove all people in the background of the uploaded image. Preserve the main subject exactly. Do not change the subject's face, body, clothing, pose, or expression. Reconstruct the background naturally where the people were removed, matching the original location, architecture, lighting, shadows, and depth. Do not crop, blur, or stylize. The result must look as if the photo had originally been taken without people in the background.

添加黄金时刻光线

一幅重新布光并采用更暖色彩氛围的肖像。

编辑后:肖像采用更暖的方向性光线,并改变了色彩对比编辑前:光线较冷且较平的肖像
编辑前
编辑后
drag_handle

结果加入了更暖的方向性光线,同时也改变了饱和度与对比度。

这项对比不能说明光线是唯一发生变化的部分。

原始提示词

Front-side sunlight, fully preserving every original detail of the photo without changes, except for the lighting and the color atmosphere. Enhance loose strands of hair against the backlight. Strengthen facial definition with a high-detail scale enhancement without altering the features. Strict rule: Do not change the face, hair, or background structure of the subject. Do not change the composition.

改变表情

一项要求只改变面部表情的肖像编辑。

编辑后:室内肖像呈轻微的闭口微笑编辑前:室内肖像呈中性表情
编辑前
编辑后
drag_handle

中性表情变为轻微的闭口微笑,而构图、房间、头发和衣着在视觉上保持一致。

视觉上的一致并不构成数学意义上的身份保证;较小的面部细节可能经过重建。

原始提示词

Change the facial expression of the person in this photo to SMILING. Keep the person's identity, age, bone structure, face shape, eye shape, nose, lips, and jawline exactly as in the original. Preserve skin texture, pores, freckles, moles, facial hair, and hairstyle. Adjust only the muscles involved in the new expression — mouth, cheeks, eyes, and eyebrows — so the change looks anatomically natural. If the expression involves a smile or laugh, show teeth only if appropriate and keep them consistent with the person's real teeth shape and color. Preserve the original lighting direction, shadows, skin tone, and color grading. Do not change the head position, pose, gaze direction, clothing, or background. The result must look like a real photo of the same person captured a moment later with a different expression.

修复受损照片

一组竖幅对比,展示带裂痕和折痕的档案照片及其编辑结果。

编辑后:移除裂痕和折痕后的合理修复结果编辑前:一张带有裂痕和折痕的受损肖像照片
编辑前
编辑后
drag_handle

裂痕和折痕被移除,缺失细节被重建为一种看起来合理的修复结果。

“32k”只出现在所引用的提示词中;它不是经测量的输出分辨率,也不是历史准确性的证明。

原始提示词

Restore this photo with period-accurate techniques, addressing any age-related issues it may have, such as blurring, damage, fading, scratches, tears, folds, worn-out areas, or being in black and white. First, analyze the image to identify the approximate era and original photographic process to ensure a historically accurate restoration. Make it look fresh and clear by gently sharpening soft edges and facial features without overdoing it, smoothing out grainy spots or noise if present, and reconstructing missing parts with realistic textures that match the original. If color is added or absent, bring the image naturally and vibrantly but true to the era's photographic technology without looking artificial; balance colors to match natural lighting, adjust brightness and contrast so everything pops nicely, and maintain original tonality. Add subtle details to faces, objects, or backgrounds that might have been lost, like fine lines in clothing, lifelike skin textures, or small elements in the scenery, while keeping the overall feel authentic, preserving natural grain patterns, and not changing the composition. Ensure the whole image is balanced, with no harsh shadows or washed-out areas, remove technical defects with a respecting the nostalg charm and exposure qualities of the time. Finally, upscale it to a higher resolution like Full HD 32k for clarity, outputting in a photo-realistic style that looks like a professionally restored or recent high-qua photo.

常见问题

Seedream 5 Pro 专为设计密集型图像工作打造:结构化版式、图内文字、图层分离,以及由点选、套索或草图引导的精确修改。

可用于广告活动视觉、产品发布板、编辑人像、专辑封面、建筑概念、信息图,以及其他重视层级关系与可控细节的构图。

可以。你可以描述目标修改,也可以用点选、套索或草图进行引导。本页示例会在修改前后结果旁展示所用的精确提示词。

是。我们会公布每个示例的精确提示词,并且只描述该次结果中实际可见的内容。每张展示图仅使用一次生成输出,不为追求质量而重跑。

用 Seedream 5 Pro 创作

注册即赠 75 个免费积分——无需绑卡,永不过期。今天就从这款设计智能旗舰模型开始。

开始创作arrow_forward
Seedream 5 Pro — 具备设计智能的 AI 图像模型 | Seedream Studio