gpt-imageIMAGE

GPT Image 2.5 chat mode vs work mode: clear differences in generated images

The same prompt produces notably different images in ChatGPT chat mode and work mode: chat mode allows bolder framing but tends to lock in composition, while work mode is more conservative and stable.

DeepBlue深藍

@DeepBlueX0Archived 18 days agoOriginally posted 2026.10.11

GPT Image 2.5 chat mode vs work mode: clear differences in generated images (image 1 of 4)

GPT Image 2.5 chat mode vs work mode: clear differences in generated images (image 2 of 4)

GPT Image 2.5 chat mode vs work mode: clear differences in generated images (image 3 of 4)

GPT Image 2.5 chat mode vs work mode: clear differences in generated images (image 4 of 4)

DeepBlue深藍 @DeepBlueX0Originally posted 2026.10.11

Open original on X

GPT对话模式和工作模式文生图，差别也太大了吧？对话模式：尺度可更大，但容易继承前图，人物动作、机位、构图和场景反复固化。工作模式：尺度更保守，但随机抽卡更稳定，人物动作、场景和构图几乎重复。 👇具体看结果吧，具体哪个模式不用我说了吧？ 🔵 GPT image 2.5 Prompt: 韩系轻熟辣妹 × 蕾丝叠穿 × 轻运动休闲 × iPhone 失败照片 9:16 竖版，成年东亚女性，韩系网红气质，冷白肤色。风格 DNA：韩系轻熟辣妹风，融合蕾丝叠穿、轻运动休闲与少量芭蕾风元素。整体精致、松弛、自然，兼具女性化魅力与日常穿搭感。穿搭方向：以柔和色彩、细腻蕾丝、柔软针织、轻薄面料与自然层次为视觉基础。款式、色彩、面料与配饰随机组合。场景与摄影：随机场景，随机试验镜头，随机非正常构图 n=4：连续独立生成4张，每次生成1张完整图，单张输出，严禁拼图、多宫格或分屏。

This post quotes another post: x.com/deepbluex0/status/2106652811312439698

x.com/deepbluex0/status/2109121110336008571Archived 2026.10.12 10:05