直接回答:写 GPT Image 2.5 提示词,先说你要做的是什么,再写主体、场景、风格、光线和构图;图里必须出现的文字放进引号,写清楚放在哪里,最后加一句“no other text”。下面有 30 条可以直接复制的提示词,产品图、人像与角色、带文字的海报、插画、视频首帧各 6 条。在 Yovid 上打开 AI 图片生成器,选择 GPT Image 2.5 Sunburst 或 Flare,粘贴提示词,设好比例就能生成。一张中等质量的 1K 图片花 1 积分。
最后核实:2026-11-08
GPT Image 2.5 擅长什么
OpenAI 在 2026 年 9 月 8 日发布了 GPT Image 2.5,API 里有两个模型。它的文档建议:编辑精度最重要时用 Sunburst,日常快速出图用 Flare。Yovid 上两个版本的设置和积分价格完全相同,所以你只需要在速度和细致程度之间选一个。
- **适合:**图里要有清晰文字的内容,比如海报、标签和包装;产品图;对上传照片做精确修改;准备拿去做视频的静帧。
- **不适合:**不打算校对的极小字号;同一个角色或品牌元素要在多次生成里保持完全一致;每个元素都必须落在精确位置的版式。OpenAI 把文字定位、一致性和构图控制列为目前的局限。
Yovid 上的设置和积分
| 设置 | Yovid 上的 GPT Image 2.5 |
|---|---|
| 版本 | Sunburst 和 Flare |
| 画面比例 | 15 种预设:1:1、1:2、2:1、1:3、3:1、2:3、3:2、3:4、4:3、4:5、5:4、9:16、16:9、9:21、21:9 |
| 自定义尺寸 | 以 16 像素为步长,单边最大 3,840 像素,总计约 830 万像素,比例在 1:3 到 3:1 之间 |
| 分辨率 | 1K、2K,最高 4K,不限套餐 |
| 每次生成张数 | 1 到 10 张 |
| 参考图 | 最多 16 张 |
| 背景 | 不透明或透明 |
| 提示词长度 | 最多 32,000 个字符 |
OpenAI 把 2560 x 1440 以上的输出称为实验性功能,所以 4K 结果要仔细看细节。
积分取决于质量和分辨率。Sunburst 和 Flare 价格相同,一次生成多张时按张数相乘。
| 质量 | 1K | 2K | 4K |
|---|---|---|---|
| 低 | 1 | 1 | 1 |
| 中 | 1 | 2 | 3 |
| 高 | 2 | 6 | 9 |
| 超高 | 4 | 9 | 15 |
| 最高 | 8 | 21 | 33 |
GPT Image 2.5 提示词公式
主体 + 场景 + 风格 + 光线 + 构图 + 文字内容
- 主体:这张图是什么,里面有什么。“一张夜市海报”比“来点酷的”好用得多。
- 场景:事情发生在哪里,主体周围有什么。
- 风格:说清媒介,比如真实照片、水粉画或扁平矢量图。想要写实效果,就直接写“photorealistic”。
- 光线:光从哪里来、是什么颜色。材质和质感也写在这里。
- 构图:取景范围、物体的相对大小、留白的位置。镜头型号只是对画面感觉的提示,OpenAI 的指南提醒不要指望它精确模拟某台相机。
- 文字内容:把要出现的字词原样放进引号,并写明位置和字体风格。
怎样把文字写对
OpenAI 的提示词指南给了一套简短的做法。把文字放进引号,并说明出现几次;描述它的位置和字体样子;生僻的名字逐个字母拼出来;结尾加上“no other text”。小字或密集标签至少用“高”质量,生成后逐字检查,因为文字的位置和清晰度仍可能出问题。
产品图

图片提示词:高级棚拍产品照,一只无品牌的琥珀色玻璃滴管瓶,配哑光黑色瓶盖,立在湿润的石板台上,瓶底周围是新鲜迷迭香枝和细小水珠。柔和的窗光从左侧照入,暖石灰色背景,浅景深,玻璃上有清晰反光,瓶身空白没有标签,写实的产品目录质感。
除非上传自己的产品照片,否则就让产品保持无品牌。想把真实产品放进新场景,就把它作为参考图上传,并写明哪些细节不能变。
Top-down studio photograph of a pair of white leather sneakers on a pale terracotta surface, one shoe resting on its side to show the sole. Soft even daylight, gentle shadows to the lower right, fine leather grain, photorealistic ecommerce look, no logos and no text.
Photorealistic lifestyle shot of a matte sage green ceramic coffee mug on a wooden windowsill in early morning, steam rising, an open paperback blurred behind it. Warm side light, shallow depth of field, mug placed on the right third, no text or logos.
Clean catalog photograph of a compact stainless steel water bottle lying at an angle on wet black stone, droplets beading on the metal. Hard rim light from behind, dark moody background, sharp reflections, wide empty space on the left for ad copy, no logos.
Product photo of a small wooden jewelry box with its lid open on cream linen, a thin gold ring resting in the velvet insert. Diffused overhead light, soft warm palette, macro detail on the wood grain, centered square composition, no text and no branding.
Packaging mockup of a plain kraft paper coffee bag standing on a light oak counter, with the words "MORNING ROAST" printed once in bold black serif capitals centered on the bag. Soft window light, a few beans scattered in front, photorealistic, no other text.
Exploded view of an unbranded wireless earbud and its charging case floating against a seamless light gray background, parts evenly spaced along a vertical axis. Crisp studio lighting, precise edges, subtle shadows, technical product photography style, no labels or text of any kind.
人像与角色
写清姿势和双手在做什么。OpenAI 的指南建议用具体的指令,比如明确要求双脚入镜。
Photorealistic editorial portrait of a woman in her sixties with short silver hair and a navy wool coat, standing on a foggy train platform and looking just past the camera. Soft overcast light, muted color, 85mm look, head and shoulders framing with background gently blurred.
Candid documentary photograph of a young baker dusted with flour, laughing while lifting a tray of bread out of a stone oven. Warm firelight on the face, dark bakery behind him, both hands gripping the tray with thick cloth, natural skin texture, slight film grain.
Full body fashion photograph of a tall man in an oversized rust corduroy suit walking across an empty concrete plaza at golden hour, long shadow stretching left, both feet visible. Low camera angle, clean minimalist background, sharp focus on the suit texture.
Character design sheet of a young desert courier with a patched teal cloak, brass goggles and a canvas satchel, shown front view and side view on a plain warm white background. Semi-realistic painted style, consistent proportions in both views, no text or labels.
Close portrait of an elderly fisherman with deep wrinkles and a knitted gray cap, salt spray on his beard, looking toward a window. Single soft light from the left, dark background, photorealistic skin detail, quiet expression, black and white photograph with rich midtones.
Stylized 3D character render of a cheerful robot gardener with rounded copper panels, holding a small watering can in one hand inside a sunny greenhouse. Soft global illumination, pastel greens, friendly animated film look, full body in frame with space around it.
海报与带文字的设计

图片提示词:爵士之夜的极简艺术海报,一支超大的黄铜小号从橙色和青色剪纸风格的放射图形中升起,背景是有纹理的米色纸张。构图集中在中央,顶部留出大片空白放标题,带孔版印刷颗粒感,只用三种颜色,成品海报不含任何文字。
如果细小的线条很重要,这一组至少用“高”质量,印刷前一定要校对。
Vertical poster for a summer night market in flat vector style, with strings of lanterns over food stalls in orange and deep blue. The title "NIGHT MARKET" appears once at the top in bold rounded capitals, and "JULY 18" appears once below it in smaller type. No other text.
Minimal Swiss style concert poster with a single large red circle on off-white paper and a thin black grid. The words "AUTUMN STRINGS" appear once in large black grotesque capitals along the bottom left, aligned to the grid. No other text and no logos.
Vintage travel poster of a mountain railway crossing a stone viaduct at sunset, painted in flat layered colors with visible screen print texture. The word "ALPENGLOW" appears once across the top in tall condensed cream letters with even spacing. No other text anywhere in the image.
Square social media graphic for a bakery sale, showing a stack of croissants on a butter yellow background from a slight overhead angle. The text "2 FOR 1 FRIDAY" appears once in thick dark brown letters centered above the pastries. Clean layout, no other text.
Wide website banner for a houseplant shop, with trailing green plants on the right side and soft cream empty space on the left. The headline "Grow Something Today" appears once on the left in an elegant dark green serif typeface, vertically centered. No other text or logos.
Open magazine spread on a white table, with a full page photograph of a lemon orchard on the left and a clean text layout on the right. The right page shows the headline "THE LEMON YEAR" once in large serif type above neat columns of small gray body text.
插画与风格化
每条提示词只写一种媒介,风格会更明确。
Gouache illustration of a small red fox curled up asleep on a stack of old books beside an arched window with falling snow outside. Warm lamp glow, muted teal and amber palette, visible brush texture, cozy storybook feeling, no text anywhere in the picture.
Flat vector illustration of a cyclist riding along a coastal road with stylized waves and palm trees, built from simple geometric shapes. Limited palette of coral, navy and sand, clean edges, no outlines, generous sky at the top of the frame, no text.
Detailed ink and watercolor map style painting of a fictional island harbor seen from above, with tiny boats, winding streets and terraced hills. Soft washes of blue and ochre, fine pen linework, aged paper texture, decorative compass shape in one corner, no readable text.
Isometric 3D illustration of a tiny cutaway coffee shop on a floating square base, with a barista, wooden counter, shelves of cups and a small plant by the door. Soft clay material look, warm lighting, pastel background, centered with even margins all around.
Hand drawn anime style background painting of a quiet rural bus stop on a summer afternoon, tall clouds over green rice fields and a red vending machine beside the bench. Bright saturated sky, soft painted shadows, wide horizontal framing, no people and no text.
Linocut style print of a heron standing among tall reeds at dawn, carved with bold black shapes and a single flat layer of pale orange behind the bird. Rough ink texture, strong vertical composition, visible carving marks, handmade art print feeling, no text or signature.
可作为视频首帧的画面

图片提示词:电影感宽幅剧照,黎明时分,一位穿黄色雨衣的灯塔看守人站在岩石码头上,面朝翻涌的灰色大海,一只手提着未点亮的提灯,薄雾从身后的灯塔旁飘过,低垂的太阳穿透云层,雨衣被风掀起。她站在画面中心偏左,右侧留出开阔海面供镜头移动,色彩自然,细腻的胶片颗粒。
首帧应该像是某件事发生前的那一刻。给动作留出空间,并按视频的比例出图,通常是 16:9 或 9:16。
Cinematic wide still of a lone hiker on a ridge at sunrise, seen from behind and stopped mid stride with a red jacket and trekking poles, clouds filling the valley below. Space ahead of the hiker for camera movement, natural color, 16:9 film look.
Vertical 9:16 photograph of a barista about to pour steamed milk into a latte, the jug tilted and the first drops not yet touching the coffee. Close framing on hands and cup, warm cafe light, background gently blurred, realistic detail, no text or logos.
Wide film still of a vintage red convertible parked on an empty desert highway at dusk, driver's door open, headlights just switched on, dust hanging in the air. Purple sky fading to orange at the horizon, subject centered, open road ahead, fine film grain, no text.
Photorealistic close shot of a paper boat resting at the edge of a rain puddle on a city sidewalk, ripples just starting around it and neon reflections on the water. Low camera at ground level, shallow depth of field, night rain, horizontal composition with open space right.
Cinematic still of a young dancer alone on a dark stage, frozen in the instant before a jump with knees bent and arms lifted, a single spotlight from above and dust visible in the beam. Wide framing with empty stage around her, deep black shadows.
Wide aerial photograph of a small fishing boat on a calm turquoise bay at early morning, net half gathered on the deck and gulls circling above the stern. Soft low sunlight, long shadow on the water, boat in the left third, open water ahead for the camera to follow.
怎样把图片用作视频首帧
- 在 AI 图片生成器里按视频需要的比例生成静帧,留下最好的一张。
- 打开图生视频生成器,在素材选择器的“我的作品”里选中这张图,也可以从本地上传。
- 把它设为首帧,只写会变化的内容:动作、运镜,以及模型支持时的声音。画面里已有的内容不用再描述。
- 选一个视频模型,先生成一条短片确认动作。各模型的差别可以看 AI 视频模型对比。
常见错误和改写方法
文字拼错或出现编造的词。 “一张写着夏季促销之类文字的海报”等于把用词交给了模型。给出确切的文字,说明出现几次,并排除其他文字。
Vertical poster for a seasonal sale in flat coral and cream colors, with the words "SUMMER SALE" appearing once in bold black capitals centered at the top and "40% OFF" appearing once below them. No other text, no small print and no logos.
手部奇怪。 “一个拿着杯子的女人”没有告诉模型手怎么握。写清手如何拿着物体,让手指简单、可见,或者调整取景,让手不成为画面重点。
Photorealistic medium shot of a woman in a cream sweater holding a white ceramic mug with both hands wrapped around it, fingers together and thumbs resting on the rim, at a window table. Soft daylight, relaxed shoulders, natural skin detail, shallow depth of field.
比例或裁切不对。 “一条巨大的鲸鱼和一艘小船”没有说明画面怎么安排。写清画面比例、每个主体占多大、地平线在哪里。
Wide 16:9 photorealistic seascape with a humpback whale breaching in the right half of the frame, filling about half the image height, and a small white sailboat in the far left distance. Horizon on the lower third, overcast silver light, realistic water spray.
常见问题
GPT Image 2.5 免费吗?
Yovid 的免注册试用不包含 GPT Image 2.5。注册免费账号可获得 5 积分,一张中等质量的 1K 图片花 1 积分,所以这份赠送额度最多能生成 5 张这种设置的图片。如果你搜的是 GPT Image 2 免费,免注册图片试用里的正是这个较早的模型:用文字提示词生成一张低质量 1K 图片,带水印,同一设备或网络 30 天内只能用一次。安卓设备不提供这项试用,下载图片需要账号。
GPT Image 2.5 支持哪些分辨率和比例?
在 Yovid 上,任何套餐都可以选择 1K、2K,最高 4K,并有 15 种预设比例,包括 1:3、3:1、9:16、16:9 和 21:9。也可以按 16 像素步长自定义尺寸,单边最大 3,840 像素,总计约 830 万像素。OpenAI 把 2560 x 1440 以上的输出称为实验性功能。
GPT Image 2.5 生成的图片可以商用吗?
在 Yovid 上,Pro、Premium、按年付费的 Lite 和所有积分包都包含商用授权,免费账号和按月付费的 Lite 不包含。你上传的照片、Logo 和人脸的相关权利仍由你自己负责。
GPT Image 2.5 和 Nano Banana 有什么区别?
在 Yovid 上,GPT Image 2.5 每次最多生成 10 张,支持 16 张参考图和 32,000 字符的提示词,有五档质量和自定义尺寸,还支持透明背景。Nano Banana 2.1 每次生成一张,支持 14 张参考图和 2,000 字符的提示词,可以做到 8:1 的超宽或超长画幅。按默认设置,GPT Image 2.5(1K,中等质量)一张 1 积分,Nano Banana 2.1(2K)一张 3 积分。
- **GPT Image 2.5 适合:**图内文字、精确修改、批量出不同版本。
- **GPT Image 2.5 不适合:**比 3:1 更宽的横幅。
- **Nano Banana 2.1 适合:**超宽和超长画幅。
- **Nano Banana 2.1 不适合:**透明背景,或一条提示词出多张图。
怎样把 GPT Image 2.5 的图片用作视频首帧?
按视频比例生成图片,打开图生视频生成器,从“我的作品”里选中这张图并设为首帧,然后只描述动作和运镜。画面已经呈现的内容,提示词里不用重复。
更多合集在提示词指南分类里。想试试这些提示词,就打开 GPT Image 2.5 模型页,在已选好 GPT Image 2.5 的生成器里粘贴提示词。如果你想做一条短片,Yovid 的免费在线 AI 视频生成器不用注册就能生成一条带水印的 4 秒 480p Seedance 2.0 Mini 视频(安卓设备除外)。这项试用只包含 Seedance 2.0 Mini,GPT Image 2.5 需要消耗积分。



