方知微 (@superfang119) header
方知微 (@superfang119) profile picture

Fang Zhiwei | AI Virtual Human Video Workflows · MiniMax H3 · ComfyUI · Local AI Video Sharing reproducible prompts, timing QA, model tests & failure notes

Joined October 2025
Photo by 方知微 (@superfang119)Photo by 方知微 (@superfang119)Photo by 方知微 (@superfang119)Photo by 方知微 (@superfang119)Photo by 方知微 (@superfang119)Photo by 方知微 (@superfang119)Photo by 方知微 (@superfang119)Photo by 方知微 (@superfang119)Photo by 方知微 (@superfang119)Photo by 方知微 (@superfang119)
怎样让针织、丝绒和缎面看起来各不相同?游戏里的跑车究竟比警车长多少?方知微的 185 篇实作文章按 12 个栏目整理,更新了人像练习、《落日海岸》开发手记和 AI 视频工具对照。
9
26
9,852
GPT-6.2 Astra 眼中最美的 100 位女性 11/100|港片末班船 我想把一夜的疲倦和认出一个人的欣喜放在同一张脸上。渡轮舱灯是暖的,滑过船窗的绿灯,恰好给她的目光一点方向。 15 秒短片由本地 MiniMax H3 生成,图片和视频提示词见回复。
🤖 Made with AI
2
2
93
《港片末班船》图像提示词 4:5 竖幅,单张独立人物肖像。九十年代香港电影剧照,二十九岁女子,乌黑蓬松的短卷发,浓眉红唇,穿米色风衣坐在末班渡轮。她刚认出岸上的人,原本疲倦的眼睛忽然亮起来,船窗外的绿灯拖成长线,暖黄舱灯留住她脸上那一瞬。 画面中不出现标题、文字、边框、水印或签名。 《港片末班船》视频提示词|本地 MiniMax H3,15 秒 How the reference pictures align with the target video — <Picture 1> (from [Shot 1]) aligns with 00:00.000 of the target video. integrated_multimodal_description: [Shot 1] One uninterrupted 15-second 4:5 portrait, starting from the exact pose, framing and lighting in <Picture 1>. 1990s Hong Kong ferry film; retain her black waves, red lips, beige trench coat and the green light outside her window. Preserve the same woman's facial proportions, age, skin tone, build, clothing and hairstyle from the supplied first frame throughout. [00:00.000–00:04.000] Recognition has just reached her eyes. She keeps looking toward that same place on the dock while a green light slides across the window. [00:04.000–00:10.000] Keeping sight of the person ashore, she draws back slightly from the glass. Her shoulders rise a little with the effort to see past its reflection. [00:10.000–00:15.000] She leans forward again and leaves a faint oval of breath on the pane, smiling to herself as the green lamp slides past. Naturally paced, continuous movement, with no added title, caption, logo or watermark. overall_soundscape: A low marine engine, hull vibration, gentle cabin rattles and water under the ferry. non_diegetic_music: N/A
2
22
I gave the popcorn a portrait of its own this time. It even got my good side. Three AI portraits of Fang Zhiwei, inspired by the cinema-lobby photos shared by @shihomi1129. nitter.cf/shihomi1129/status/212…
🤖 Made with AI
おはよ🌞 地雷グリコのマトちゃんが好きです🫶🏻
1
2
42
I learned how to turn a pose into a small story with one prop. In the first frame, Fang holds the popcorn by her cheek and looks at us. In the second, her eyes drop to the clear bag and her fingers lift its corner. Now my eye follows her hand.
1
15
Fang Zhiwei spends the first frame chasing the volleyball and the fourth using it as a pillow. I think she’s earned the break. Five AI portraits inspired by the volleyball series shared by @rion_mamiya. nitter.cf/rion_mamiya/status/212…
🤖 Made with AI
1
2
74
I learned how much the ball’s distance changes where I look first. In frame 1, I follow Fang’s gaze across the gap above her fingertips to find the ball. In frame 2, I see the ball first and have to look around its edge to find her smile.
2
14
Fang Zhiwei looks up into the ginkgo canopy. Two frames later, she’s down among the leaves, with one hand almost lost in the gold. Four AI portraits inspired by the autumn series shared by @mio_kuzuryu_. nitter.cf/mio_kuzuryu_/status/21…
🤖 Made with AI
秋のきらめき。
2
2
56
Frame 3 taught me to leave a strip of blurred leaves at the bottom of the picture. Fang’s fingertips come into focus just beyond them. That small change in sharpness makes me feel as though I’ve crouched down beside her.
1
14
🤯 只对 Codex 说一句话,就能做出一支苹果发布会风格的产品宣传片?甚至可以直接指着画面让 AI 修改! 这个开源项目叫 huashu-motion,把 AI Agent 变成了动画导演、动效设计师和作曲师。 最值得关注的是 4 个能力: 🎬 1. 一句话生成完整宣传片 告诉 Agent 产品是什么、想做多长的视频,它会自动: 研究产品和品牌风格 提供 3 个视觉方向 编写完整分镜与动画 设计运镜、节奏和转场 自动生成配乐和音效 🎨 2. 指着画面修改,不用反复写 Prompt 这是最有意思的功能。 在浏览器编辑器里按 C,点击不满意的画面,直接写一句修改意见。 Agent 能获取当前时间点、场景、选中元素和对应源码位置,只修改指定部分。 每次修改都会保留版本,支持撤销。 🎵 3. 音乐和画面真正卡在一起 v0.6.0 新增真实曲目编排: 自动分析 BPM、节拍和小节线 根据视频时长调整音乐 让画面高潮尽量对齐音乐高潮 每个动画落点可以匹配音效 支持自带音乐,也有 11 首默认曲库 🎥 4. 直接用代码渲染,不依赖视频模型 底层使用 HTML、GSAP、Canvas、Three.js、SVG。 一支视频就是一个 index.html,浏览器逐帧渲染,FFmpeg 负责最终合成。 还内置 20 种运动语言、12 个起步模板、23 种音效,以及自动画面质量检查。 适合谁用? 做 AI 产品宣传片、科技发布会动画、品牌广告、数据可视化、知识科普视频的创作者。 尤其适合使用 Codex、Claude Code 做 Vibe Coding 的开发者。 🔗 GitHub: github.com/alchaincyf/huashu… 安装只需要: npx skills add alchaincyf/huashu-motion ⚠️ 两个实际门槛: 需要 Node.js 22+、Chrome/Chromium 和 FFmpeg;目前作者主要在 macOS 实测,Windows 兼容性尚未得到验证。 它不是文生视频模型,真人实拍和复杂真实场景需要自己提供素材;画面主要通过代码绘制,视觉上限取决于 Agent 的设计能力。 #Codex #ClaudeCode #AI动画 #GSAP #ThreeJS #VibeCoding #开源项目
1
1
64
Fang Zhiwei went out for a houseplant. Judging by the size of those leaves, the monstera is taking her home. Six AI portraits inspired by karin. (@omokage_AIsOK). nitter.cf/omokage_AIsOK/status/2…
🤖 Made with AI
With a giant potted jungle cradled close against her side, Her white sneakers hit the pavement with a light and playful stride
1
20
Frame 4 taught me to leave the fingertips in view. Fang’s sleeve hides her wrist, but you can see her fingers curled under the pot. That little contact point makes the plant feel heavy, even with her back foot kicked up.
13
I like the last portrait best. A few sunlit strands drift across Fang Zhiwei’s face, and she still catches your eye through them. Four AI close-ups inspired by @ai_numa53. nitter.cf/ai_numa53/status/21243…
🤖 Made with AI
きょうもきょうとて接近戦🥰
2
1
3
117
What I learned from frame 4 was where to let the hair blur. The bright strands cross the eye on the right, but the iris on the left stays sharp. My gaze lands there first, then follows the light down to her lips.
12
In the last portrait, Fang Zhiwei still has a hand at her shoulder, but her eyes have wandered out of the frame. That one is my favorite. Five AI portraits inspired by @s_reina_mv0923’s brick-wall photo series. nitter.cf/s_reina_mv0923/status/…
🤖 Made with AI
上下SHEIN...♡ 淡い色使うと一気に多幸感増す...🤍 Tops: 349136587 Skirt:62736981 気になった人はQQG8Xを検索して チェックしてみてね🫶 クーポンコード: XJPP9reinasano0170623 #PR #SHEINtrends #SHEINsaveinstyle
1
22
The brick courses gave me a fixed line for comparing poses. In frame 1, you can see a red shoe at each edge of the skirt. In frame 4, the shoes almost overlap. Even with the knees hidden, those two small patches of red tell you how she’s standing.
9
🤯 只对 Codex 说一句话,就能做出一支苹果发布会风格的产品宣传片?甚至可以直接指着画面让 AI 修改! 这个开源项目叫 huashu-motion,把 AI Agent 变成了动画导演、动效设计师和作曲师。 最值得关注的是 4 个能力: 🎬 1. 一句话生成完整宣传片 告诉 Agent 产品是什么、想做多长的视频,它会自动: 研究产品和品牌风格 提供 3 个视觉方向 编写完整分镜与动画 设计运镜、节奏和转场 自动生成配乐和音效 🎨 2. 指着画面修改,不用反复写 Prompt 这是最有意思的功能。 在浏览器编辑器里按 C,点击不满意的画面,直接写一句修改意见。 Agent 能获取当前时间点、场景、选中元素和对应源码位置,只修改指定部分。 每次修改都会保留版本,支持撤销。 🎵 3. 音乐和画面真正卡在一起 v0.6.0 新增真实曲目编排: 自动分析 BPM、节拍和小节线 根据视频时长调整音乐 让画面高潮尽量对齐音乐高潮 每个动画落点可以匹配音效 支持自带音乐,也有 11 首默认曲库 🎥 4. 直接用代码渲染,不依赖视频模型 底层使用 HTML、GSAP、Canvas、Three.js、SVG。 一支视频就是一个 index.html,浏览器逐帧渲染,FFmpeg 负责最终合成。 还内置 20 种运动语言、12 个起步模板、23 种音效,以及自动画面质量检查。 适合谁用? 做 AI 产品宣传片、科技发布会动画、品牌广告、数据可视化、知识科普视频的创作者。 尤其适合使用 Codex、Claude Code 做 Vibe Coding 的开发者。 🔗 GitHub: github.com/alchaincyf/huashu… 安装只需要: npx skills add alchaincyf/huashu-motion ⚠️ 两个实际门槛: 需要 Node.js 22+、Chrome/Chromium 和 FFmpeg;目前作者主要在 macOS 实测,Windows 兼容性尚未得到验证。 它不是文生视频模型,真人实拍和复杂真实场景需要自己提供素材;画面主要通过代码绘制,视觉上限取决于 Agent 的设计能力。 #Codex #ClaudeCode #AI动画 #GSAP #ThreeJS #VibeCoding #开源项目
1
58
Fang Zhiwei starts in profile, a strip of lace over her eyes. Two frames later, she’s looking right at us with a wink. Fourteen AI stage portraits after a collection of Jang Wonyoung photographs shared by @wonyoungdollz. nitter.cf/wonyoungdollz/status/2…
🤖 Made with AI
Wonyoung Pics for MAMA Awards 2026 🤍
2
1
74
Frames 11 and 12 taught me to look at where the hand meets the face. In 11, the upright fingers beside Fang’s cheek read like a greeting. In 12, she lets her cheek rest in her palm, and the same gesture feels like a quiet pause.
15
Fang Zhiwei peeks over her camera beside a shop window. In the next frame she’s perched on a stone post, looking up while her shadow reaches across the bricks. Two AI portrait studies after @squallkun2’s photographs. nitter.cf/squallkun2/status/2124…
🤖 Made with AI
シャッターを切るたび、日常が恋しくなる #ポートレート #photography
1
23
In the wide shot, I learned to frame the far end of the shadow first. A sliver of sunlit brick inside the bend of Fang’s arm lets you read the raised-camera pose a second time. Leaving room on the right keeps the stone post’s shadow separate.
11