A roundup of October 2026 hands-on content about AI image generation from the Chinese-language community: prompt writing for GPT Image and Nano Banana, local editing and outfit and face swapping, character consistency methods, and running models locally with ComfyUI. Suited to people who want their generated images to look better and be edited more precisely.
Not in your language? Paste the link or the text into your AI and let it read it for you.
To keep ChatGPT-generated images from looking fake, start by writing "a real photo," then add lens and texture details, and finally specify not to beautify.
For candid-style prompts, first set distance, flash, and cropping, then add digital sharpening and JPEG texture. Only then will the candid feel hold up.
When edits keep accidentally damaging surrounding content, first split the design into separate layers, then refine only the layer that needs changing.
For outfit changes, use Qwen Image 2.1's consistency LoRA at strength 1.0 with no trigger word. The face and background stay largely unchanged.
To make storyboards or short dramas, keep characters consistent: first lock down multi-angle character costume sheets and scene assets, then choose a video model.
To change poses, first generate a live-action photo to fix the action, then convert it into a line-art white model, pair it with the character image, and hand it to an image generation model to redraw.
You can run Qwen image generation locally with only 6G of VRAM by using the GGUF Q4_K_M quantized version and switching to GGUF Loader in ComfyUI.
Prepare three-view and multi-expression reference sheets for the character so that when GPT repeatedly draws the same character, it drifts less.
Lucas10-12ZHFree AI Image Generation Site with Multiple Mainstream Models, No Sign-Up Required#1别满世界扒免费的 AI 生图网站了，这个真可以直接收进收藏夹！！！ 这站不用注册，也不用绑卡，打开输入提示词就能立刻出图AI 图片生成ImageOriginal post194KLikes 1.8K范师傅10-19ZHTool That Keeps Storyboard Characters Consistent by Locking Character and Scene Assets#2前字节团队操刀，蒸馏 250 部影史名片，打造 27 万+ 电影分镜索引库，杀入影视行业。这就是专为独立导演和广告人打造的 AI 分镜 Agent —— Story2Board 用大模型做分镜，最让人头大的莫过于两件事：AI 图片生成ImageOriginal post103KLikes 1.9KAI少年10-08ZHSplit Into Separate Layers First, Then Refine Each One Individually to Avoid Changing Other Parts When Editing#3伙伴们，国庆假期玩的开森了嘛 就在刚刚，我刷到一个 6B 的国产开源模型，感觉它给图片精修提供了一个完全不同的思路。 它所在的 Ming-Image-0.1-Design 系列，一出来就在 AA 的 UI/UX Design 开源评测里冲到了第一梯队。AI 图片生成ImageOriginal post100KLikes 64
观默10-31ZHPicture-Book-Style Character Art Plus a Video Model Produces a Short Film with 10 Million Views不到一天时间，这条不到 1 分钟的动画短片直接跑出了破千万（目前近 2000 万）的播放量。 当大多数人还在用 AI 死磕逼真的好莱坞特效、卷复杂的丝滑运镜时，它却用最克制的 2D 手绘绘本风，完成了一次彻底的降维打击。AI 图片生成ImageOriginal post98KLikes 1.2KDeepBlue深藍10-11ZHNine-Shot, Nine-Grid Image Prompt for a Phone Accidentally Dropping and Falling加了点更好玩随机元素～ 🔵 GPT image 2.5 Prompt: 【主体】 × 手机后置镜头主观视角意外脱手坠AI 图片生成ImageOriginal post71KLikes 687WquGuru10-29ZHGet a Live-Action Photo First, Then Convert to Line Art White Model to Change Poses昨天几十万浏览的那个姿势库镜像，原作条款其实写了禁止再分发，我干脆用 21 小时自己重做了一套 CC0 免费版。 一共 288 个姿势 × 5 个机位，1133 张质检过的线稿，全靠 3 个 AI agent 用共享记忆接力跑完。AI 图片生成ImageOriginal post57KLikes 753实践哥 Li10-28ZHMaking AI Short Dramas: Lock Character Multi-Angle Costume Sheets First🔥 聊聊黄果短剧怎么做。 很多人上来就纠结 Wan 还是 H3，NSFW，甚至直接研究买卡。 顺序反了。 先解决人物一致性，再考虑视频模型。先跑通工作流，再考虑买显卡。 应该效果验证通过后，再决定继续用 API，还是转向本地部署。AI 图片生成Original post57KLikes 519Lonely10-06ZHOutfit LoRA Changes Only Clothing, Keeping the Face and Background Unchanged🎉 完美！真正的 只换衣服不换人 ！ 虽然千问 Qwen Image 2.1 原生支持虚拟换装，但直接用经常会有像素偏移，甚至导致人脸变糊、背景被莫名重绘AI 图片生成ImageOriginal post54KLikes 885Lonely10-03ZHQwen Face, Head, and Body Swap LoRA, with ComfyUI Workflow Included🎉重磅！Qwen Image 2.1 的换脸/换头天花板 LoRA 来了！ 开源神库 BFS-Best-Face-Swap 迎来重大更新，现已正式支持最新的 Qwen-Image-2.1 基座模型AI 图片生成ImageOriginal post54KLikes 854
木马人10-02ZHOne Prompt Turns Travel Photo Originals into Retouched Images出来度假，才发现自己的信息茧房实在太厚了😂😂😂 一张旅拍的精修图，线下门店要收100一张，依然有大把人上赶着花钱去买~AI 图片生成Original post52KLikes 238VoxCat10-07ZHGPT Image Prompt for Phone Direct-Flash Close-Up Candid Shots我设计的这套提示词，想把角色放进近距离手机抓拍里，让表情和手势成为画面的重点。 构图先锁定三个关系：脸贴近镜头、手从前景进入、背景被直闪压暗。距离拉近后，脸和手自然产生尺度差，歪头与贴边裁切进一步增加随手拍的感觉。AI 图片生成ImageOriginal post51KLikes 2.4KAI探路者Tim10-27ZHFive-Step Prompt Writing So ChatGPT Image Generation Doesn't Look Fake用 ChatGPT 生成图片，九成人一看就假。 OpenAI 官方指南里的示例， 用的指令正好相反。 今天起按这个顺序提要求试试↓ ①先说死“照片” 开头就写“真实的照片”。 用英文的话写“photorealistic”就行。AI 图片生成ImageOriginal post50KLikes 996
古一10-10ZH1,617-Style Map: Mix Styles to Generate New Prompts兄弟们，我爆肝了半个月，把 AI 画图能用到的风格全部梳理了一遍！ 13 大领域，157 个谱系，1617 种风格！ 起因很简单：做图做久了你会发现，难的不是工具，而是你叫不出名字的风格，就写不进 Prompt。AI 图片生成ImageOriginal post43KLikes 676DD.Cherry10-02ZHUpload a Pose Reference and Material Images, and Change to the Specified Action with One Sentence使用X友分享的姿势图成功出图了。 操作方式，先上传姿势图，再上传素材图，然后使用下面的提示词（动作序号改成你喜欢的）。AI 图片生成ImageOriginal post42KLikes 828
歸藏(guizang.ai)10-07ZHNano Banana 2.1 Released: Clearer Chinese Text and 4K Output谷歌发布了 Nano Banana 2.1 图像模型 中文的文字清晰度和错字问题大幅改善！AI 图片生成ImageOriginal post41KLikes 204SuSu_酥酥👅10-04ZHRun Qwen Image Generation Locally in ComfyUI with GGUF on 6G VRAM😧什么？！无审查版 Qwen-Image 2.1 GGUF 已能本地跑！ 7B 统一生图+编辑，原生透明通道，最多 10 张参考图，圈选/蒙版改图还能保脸。Q4_K_M 约 4.6GB，6G 卡也能试，本地无内置审查，提示给什么出什么。AI 图片生成ImageOriginal post38KLikes 406
鱼哥10-09ZHPrompt for Turning Old House Photos into 1:24 Scale Miniature ModelsChatGPT Image 2.5 把你家老房子做成微缩模型 Prompt: 参考上传的老房子照片，以1:AI 图片生成ImageOriginal post37KLikes 467
向阳乔木10-04ZHLet Various Agents Call Codex Image Generation for Covers and Infographics让豆包、Workbuddy、Claude Code、Deepseek Harness 都能调用 Codex 的内置生图能力！AI 图片生成ImageOriginal post31KLikes 395KC10-20ZHUse Three-View and Expression Reference Sheets So GPT Draws Comic Characters Without Drifting分享一套 AI 娘的人设图，建议喜欢让 GPT 画漫画的朋友收藏一下！ 以后让 GPT 画 AI 娘，终于不用每次都重新解释角色长什么样了AI 图片生成ImageOriginal post31KLikes 921
疯狂的烤妹儿 🩵10-05ZHEnter a Story to Generate Unified-Style, Childlike Watercolor Picture Book Illustrations不会画画，也能把脑海里的故事做成绘本～ 最近发现一个适用人群很广的 Skill：naive-watercolor-picturebook。 输入故事或画面想法，它会提取主角、动作和情绪，生成高留白的童真水彩插画。AI 图片生成ImageOriginal post30KLikes 122A collection of Grok Bot practical tips shared by Chinese-speaking users in October 2026: how to assign tasks, how to save quota, and how to work with Claude Code and Codex. Suited for people who just got Grok Bot and don't know where to start.
A collection of hands-on Claude Code tips from the Chinese-language community in October 2026: settings that save quota, sub-agent division of labor, code review workflows, and how to use Skills, plugins, and Mods. Suited to people who want to use it more reliably and economically.
A roundup of hands-on Codex content from the Chinese-speaking community in Oct 2026: how to spend less quota, how to configure Skills and MCP, how to pair it with other Agents, and several official updates from this month. Suitable for subscribers who want to get more value out of their plan.