GPT-Image-2.5 发布了,来分享分享你的提示词吧

Joe 2026-09-09 08:31 1

我先来几个。



9:16 vertical candid portrait of a young Korean woman glancing back over her shoulder in a dimly lit bedroom. She wears a loose off-shoulder cream knit sweater with a thin strap camisole underneath. Hair pulled up in a messy bun with loose strands framing her face. Warm amber ambient lighting from a single bedside lamp, slightly cluttered room background with a wall clock, vanity shelf, and scattered items creating a lived-in feel. Soft focus on the background, sharp focus on the face and shoulder. Natural skin texture with subtle warm glow, slightly parted lips, gentle eye contact with the camera. Vintage film grain, muted warm tones, lowered contrast, slight vignette. Intimate candid photography mood, Korean idol photobook aesthetic, no watermark, no text.





9:16 vertical - Korean idol portrait photography, single subject. Soft black mist filter effect, lowered contrast, gentle highlight bloom, subtle glow, soft diffusion, slightly faded blacks. Minimal indoor setting near window, white curtains, clean light-toned background. Young Korean female idol, natural minimal makeup, dewy realistic skin texture, subtle imperfections. Outfit: oversized white button-up shirt with short bottoms, slightly loose fit, soft and casual styling. Hair: long dark hair, slightly messy, natural volume, softly flowing. Pose: relaxed standing or slight lean, body subtly angled, one leg slightly forward, shoulders relaxed; one hand lightly touching collar, the other relaxed; gentle body curve without exaggeration. Expression: soft cute smile, slightly playful eyes, direct or slightly off-camera gaze. Camera: close to mid-body framing, eye-level, intimate distance, slight handheld feel. Lighting: diffused natural daylight, soft shadows, gentle light wrapping around face and body. Mood: cute yet subtly sensual, intimate, everyday softness, quiet romantic atmosphere. Quality: ultra-realistic, fine film grain, slight softness at edges, natural imperfections, dreamy understated tone.




这次 API 这次直接出了两个版本:


GPT-Image-2.5 Flare:速度优先,适合日常出图、SNS、网页素材和批量生成。


GPT-Image-2.5 Sunburst:精度优先,更适合广告图、产品图、复杂编辑和最终交付。

最新回复 (19)
  • 番茄炒鸡蛋 09-09 08:33
    1

    感觉和2差不多?没啥提升?人像还是有奇妙的噪点和油腻感

  • Joe 楼主 09-09 08:33
    2

    可能最大的优化是速度和节省了他们的算力。

  • 番茄炒鸡蛋 09-09 08:34
    3

    嗯,速度变快了,质量没降也挺爽的

  • 艾尔 09-09 08:39
    4

    感觉 2 已经无敌了这么久了,出 2.5 大家也无感了。

  • alizoed 09-09 08:39
    5

    还是有那种假纹理啊,有点失望,还以为解决了


    codex里啥时候能用,不是说发了吗

  • swd 09-09 08:49
    6

    免费用户,给没有资格,只有等等看呢

  • 初音不在家 09-09 08:51
    7

    1K正常是多久出图?现在能压到30秒了么?如果能压进10S秒左右,应该小说图文会很舒服了

  • alizoed 09-09 08:54
    8

    正常image-2也能压到30s吧

  • IsNot 09-09 08:57
    9

    给了吧,以网页端图片的上面有两个小字(更新)

  • 洋芋 09-09 08:59
    10

    不会是像 nanobananan pro 升级到 nano banana2 那种吧 速度快了 质量差很多

  • Totoria 09-09 09:02
    11

    现在哪里能用? codex里好像没有 web能调用了吗?

  • robo7 09-09 09:04
    12

    单从这两张图看,还是和大香蕉一个水平线上的

  • xiaoming00 09-09 09:06
    13

    GPT-Image



    GPT-Image这些咋用的?web聊天会触发吗?还是用codex?codex没看到这个模型啊

  • lueluelue 09-09 09:07
    14

    有一项大幅提升了,这项就是价格^-^

  • lueluelue 09-09 09:07
    15

    聊天和Codex应该都有吧?聊天好像是那个Flare

  • akane 09-09 09:07
    16

    目前看起来质量没有差很多,即使free的也可以

  • alizoed 09-09 09:09
    17

    codex好像还没上,cpa没识别到,不知道为什么

  • 万能的AI 09-09 09:10
    18

    质量看着没多少提升。还是一眼AI 。

  • tsugumi 09-09 09:11
    19

    在官网试了试,目前感觉速度也没快,质量好像也没有显著的提升。再测几个去

* 帖子来源Linux.do
返回