SingoKe retweeted
This is one of my favorite animations I've made in a while.
I absolutely love how the dark, mysterious graphic novel style works with the split-screen effect. It gives the whole thing such a great atmosphere.
Tomorrow I'll be showing my subscribers how to create animations like this!
SingoKe retweeted
Minimax H3 换脸 LoRA
FaceSwap_MiniMaxH3_REF2VA
看例子,效果不错
可以试试
huggingface.co/UntMods/FaceS…
SingoKe retweeted
Perfect AI cinematography is overrated.
Sometimes you need shaky tracking, missed framing, ugly zooms and compression artifacts to make something feel REAL. 👀
This Seedance 2.5 experiment proves it.
Full prompt below 👇
Ultra-realistic viral helicopter pursuit footage, raw local-news style, filmed entirely from one helicopter in bright Los Angeles daylight. Long-lens aerial camera, unstable vibration, abrupt zooms, compression artifacts, heat haze, autofocus breathing, imperfect framing, rotor thump, distant radio chatter, harsh California sun, no cinematic polish, believable amateur-news realism.
Main subject: one motocross rider on a dirt bike, full helmet and motocross gear, highly skilled, aggressive but controlled riding. Keep the same rider and same bike throughout.
Los Angeles daytime: palm-lined boulevard, parking lots, freeway ramps, concrete flood channel, industrial blocks, overpasses, service roads and distant skyline.
0-5s: Immediate viral hook The helicopter camera is already searching frantically when the rider suddenly blasts into frame between moving cars. Operator snaps into a shaky zoom. The rider jumps a curb, cuts across a parking lot and nearly disappears behind palm trees.
5-10s: Chaotic tracking He rejoins a boulevard, threads through traffic, then veers up a sloped driveway and launches over a low divider. Camera over-zooms, loses him for half a second, then catches him landing cleanly.
10-15s: Escalation The rider enters a concrete flood channel. Huge aerial reveal. He accelerates along the channel floor while the helicopter banks hard overhead. Dust trails behind the bike. Quick zoom on spinning rear wheel and suspension.
15-20s: Viral stunt beat He rides up the angled concrete wall, holds a dramatic high line for a moment, then drops back down smoothly. Camera shakes violently trying to follow. Rotor vibration and radio chatter intensify.
20-25s: Big surprise The rider exits through a service ramp, crosses beneath an overpass, then unexpectedly appears on a parallel industrial road. The operator pans late and barely recovers him. He hits a natural rise and gets long airtime in full daylight.
25-30s: Iconic ending He lands, accelerates toward another overpass and vanishes into shadow underneath. The helicopter overshoots, circles, zooms wildly across multiple roads, but he is gone. Reporter voice, stunned: “We just lost him.” Hold on shaky wide aerial of Los Angeles. Hard cut.
One continuous helicopter perspective, strong amateur-news authenticity, same rider throughout, realistic motocross physics, believable Los Angeles geography, harsh daylight, imperfect tracking, viral unpredictability, readable stunts, no crash, no gore, no readable text, no logos.
#DreaminaCPP @dreamina_ai
SingoKe retweeted
🚨 VEED Fabric 1.0 API
The world's most realistic talking video model.
Ship 5-minute long ads, demos, explainers, & more.
Starting from just $0.08/sec.
Comment “API” & we’ll DM you access link.
SingoKe retweeted
装完 Codex 不知道干什么? 当然是猛做视频,然后库库搞钱啊!
工具装好了,真正的差距在会不会用。
mp.weixin.qq.com/s/7NNYtLY8y…
给 Codex 装上这 10个 GitHub Skills,直接让它参与整条视频制作流程。可以先收藏,按自己的视频类型安装👇
1️⃣ HyperFrames:一句话生成动效视频
告诉 Codex 视频主题,它可以使用 HTML、CSS 和动画制作产品介绍、动态海报、知识视频和 PPT 风格视频,最后渲染为 MP4。很适合把文章、推文、产品介绍快速改成视频。
GitHub:github.com/heygen-com/hyperf…
2️⃣ video-use:让 Codex 帮你剪视频
把原始素材丢进去,它会先转录,再根据内容帮你做剪辑方案,能切片、调色、加字幕、加动效,最后用 FFmpeg 渲染成片。比较适合口播、教程、访谈、Vlog 这种“素材已经有了,只是不想自己慢慢剪”的场景。
GitHub:github.com/browser-use/video…
3️⃣ Remotion Skills:用代码批量做视频
如果你经常做固定模板的视频,这个非常适合。让 Codex 用 React 写视频,标题、字幕、图片、动画、数据都可以自动替换,还能直接预览和渲染。特别适合批量做 Shorts、榜单、产品介绍和数据类视频。
GitHub:github.com/remotion-dev/skil…
4️⃣ ffmpeg-skill:给 Codex 装一个“万能剪辑工具箱”
裁切、拼接、横屏转竖屏、去静音、压缩、加字幕、Logo、背景音乐、调音量都能直接用自然语言让 Codex 做,而且素材可以全程留在本地。适合最后统一处理和批量导出。
GitHub:github.com/remotion-dev/skil…
5️⃣ agent-caption:自动给视频做字幕
把视频交给 Codex,它可以自动识别人声、对齐时间轴,再直接烧录字幕。做口播、课程、Podcast、Shorts 特别省时间,而且支持本地运行,不想把原始视频上传到云端的可以重点看。
GitHub:github.com/ahkamboh/agent-ca…
6️⃣ video-subtitle-translator:一条视频直接做多语言字幕
先识别原视频语音,再翻译成目标语言,自动生成 SRT/VTT,也可以直接把字幕压进成片。做 YouTube 出海、英文教程搬到中文、中文内容做海外版都很实用。
GitHub:github.com/alwaysmavs/agent-…
7️⃣ video-dubbing:让 Codex 给视频自动配外语
这套更进一步:识别原声 → 翻译 → TTS 配音 → 对齐时间 → 合成字幕和新音轨。
也就是说,一条中文视频可以直接改成英文、日文等版本。想做多语言账号的,这个比单纯翻字幕更有用。
GitHub:github.com/alwaysmavs/agent-…
8️⃣ app-demo-agent:专门做 App / SaaS 宣传视频
你先录一段产品操作,它可以继续帮你分析画面、写旁白、生成配音、加设备边框和背景音乐,最后拼成一条完整 Demo。
独立开发者拿来做 Product Hunt、X 产品演示特别合适。
GitHub:github.com/alwaysmavs/agent-…
9️⃣ browser-video-recording:自动做网页操作演示
想录“点哪里、怎么操作”的教程,可以让 Codex 根据网页截图生成带鼠标移动、点击、跟随缩放的浏览器演示视频。做 AI 工具教程、网站介绍、产品 Walkthrough,比自己一遍遍录屏省事很多。
GitHub:github.com/MengTo/Skills/blo…
🔟 imagegen:给视频自动补素材
视频最烦的往往不是剪辑,而是缺封面、插图、背景图、透明 PNG 和转场素材。OpenAI 自己在 Codex 仓库里就提供了 imagegen Skill 示例,可以让 Codex在项目里直接生成或修改这些视觉素材,再交给前面的 HyperFrames / Remotion 去动起来。
GitHub:github.com/openai/codex/blob…
工具装完别急着全用,先选一条最容易收钱的工作流跑通。比如专门做 SaaS 产品宣传视频:
客户给你官网链接 + Logo + 产品卖点
→ Codex 抓信息、写脚本
→ HyperFrames / Remotion 做画面
→ Caption 自动字幕
→ Dubbing 做英文版
→ FFmpeg 批量导出 16:9 + 9:16 两个版本
先拿一个真实产品做 30~60 秒 Demo,当成作品集。然后直接去 X 找独立开发者、AI 工具作者、小 SaaS,私信一句:
“我把你官网做成了一条 40 秒产品视频,这是 Demo。如果你觉得能用,我可以继续帮你做 3 条不同版本。”
别一上来就卖“AI 视频服务”。直接卖结果:
产品演示视频 / X 宣传短片 / 多语言版本 / 批量 Shorts。
真正能赚钱的不是你装了 10 个 Skill,而是你能不能把其中 3~4 个串成一条别人愿意付钱省掉的流程。
SingoKe retweeted
“When evil rises, the light stands its ground.✨
Created with Seedance 2.5
Prompt:
EXACT VIDEO RECREATION PROMPT — 31 SECONDS
0–3s | Opening shot: Low-angle wide close-up of a wounded woman lying on a cracked stone floor inside a ruined Gothic cathedral. Dark horned monsters and twisting tentacles loom in the background. Slow cinematic camera movement.
3–5s | Rising movement: The woman pushes herself up from the floor, raising one arm as a bright white magical blade of light forms in her hand. Medium shot, smooth upward camera movement.
5–7s | Magical activation: She kneels and grips the vertical white energy blade touching the floor. A brilliant golden circular rune spreads beneath her, illuminating the stone floor. Low wide shot.
7–10s | Cathedral reveal: Camera tilts sharply upward, revealing towering broken Gothic arches, a shattered roof, moonlit clouds and dark creatures silhouetted above. Transition to the golden-armored warrior emerging through red smoke.
10–12s | Helmet close-up: Tight frontal close-up of a fully enclosed dark metal knight helmet, narrow black eye slit, engraved steel plates. Slow camera push-in.
12–15s | Golden knight reveal: Cut to a heroic low-angle shot of a knight wearing ornate gold-and-silver armor, carrying a large shield with a glowing golden sun emblem and a heavy war hammer. He stands inside the ruined cathedral.
15–18s | First magical attack: The knight raises his shield and releases an intense white-gold energy blast. Bright light fills the frame. Camera tracks the blast, with sparks and magical particles.
18–20s | Energy impact: The white magical blast expands into a brilliant screen-filling flash. Strong bloom, fast cinematic transition, brief motion blur.
20–23s | Demon confrontation: Wide side view of the golden knight standing firm while a huge reddish-black horned demon lunges toward him. Long claws and dark tendrils extend toward the knight. The wounded woman remains visible in the background.
23–25s | Demon attack: The demon sweeps its claws forward. A bright golden energy projectile travels across the cathedral between the creature and the knight. Dynamic lateral camera movement.
25–27s | Magical shield: The knight raises his shield as a large circular white magical barrier with intricate glowing runes forms in front of him. The demon attacks the barrier, creating rippling light and sparks.
27–31s | Final wide shot: Camera pulls back to show the entire ruined cathedral. The golden knight stands in the centre of a glowing golden rune circle, shield raised and hammer in hand. The demon recoils on the left, while the wounded woman remains on the floor behind him beside the vertical white energy blade. Smoke, floating embers, blue moonlight and golden illumination. Hold the final composition.
VISUAL STYLE: Photorealistic dark fantasy film, 16:9 widescreen, cinematic lighting, realistic armor and stone textures, cold blue shadows, warm golden highlights, volumetric smoke, dramatic contrast, consistent architecture, smooth camera movements, exact shot order and timing.
NEGATIVE PROMPT: Different shot order, altered camera angles, random cuts, changed framing, different cathedral, different character positions, inconsistent armor, changing shield emblem, extra characters, duplicated limbs, deformed anatomy, cartoon, anime, low quality, excessive blur, shaky camera, camera flip, 180-degree rotation, unnatural movement, floating weapons, text, subtitles, watermark, logos.
#AI #AIart #VisualArt
🤖 Made with AI
SingoKe retweeted
if you're using claude to edit videos, you can give it a library of 5.2M+ audio files for $0
sound effects, ambience, music. openverse has a free api. no key needed.
give it api.openverse.org and ask:
“whenever you need sound effects search Openverse to find the best options and use them for tasteful sound design within my video”
no more using your browser to download random clips or trying to generate the audio itself
SingoKe retweeted
🔥 兄弟们,我做了个开源工具:拆画,能把一张图拆成可编辑的图层。
很多 AI 生成图和现成海报,拿到手只有一张 PNG 或 JPG。标题、商品、装饰、背景都在同一张图里。
想移动某个商品、提取价格标签,要么重新抠图、修补背景,要么再生成一次。图看着已经完成了,后续修改和复用还是费工夫。
拆画就是让这些元素重新成为独立的图片图层。
比如这张「双11」海报,拆完可以分别移动商品、缩放标题、隐藏价格标签,薯片袋也能直接下载成透明素材,用到下一张图里。适合电商设计、活动海报制作和素材整理。
整个流程都在同一张画布上完成:
1️⃣ 输入需求生成图片,或上传、拖入已有图片;
2️⃣ Ling-3.0-flash-VL 看图制定方案,Ming Layer 还原图层,;
3️⃣ 拖动、等比缩放、调叠放顺序,隐藏或锁定图层,支持撤销、重做和刷新恢复;
4️⃣ 可以导出PSD和单个素材的PNG
默认用百灵开源的 Ming-Image-0.1-Design 系列生图和拆层。模型介绍和体验地址放评论区 👇
SingoKe retweeted
영화 제작·연출의 '시각적 언어'를 정리한 필름 테크닉 107개
샷 종류부터 카메라 움직임, 조명,
편집, 구도 규칙, 장르까지
영화 제작에 쓰이는 107가지 핵심 기법을
한눈에 볼 수 있게 모아둔 백과사전임
특히 초보자 입장에서 용어는 대충 알아도
정작 "언제 써야하는지" 몰라서 막막할 때가 많은데
그럴 때 각 기법을 클릭해서 들어가 보면
바로 써먹기 좋은 실전 가이드가 싹 정리돼 있음⤵️
H3 Character Swap Lora V1.5 is out!
Better identity preservation + better handling of cuts + a few improvements over V1
AND I'm releasing the workflow/tool I've been using with it too👇
It's not V2 yet 😅 I still need more time to cook. Several revisions haven't passed my hiiigh V2 standards. But V1.5 did beat out V1 in my official "trust me bro" benchmark below.
Lora: huggingface.co/akatz-ai/Mini…
SingoKe retweeted
Same shot, different universe. No reshoots needed 👀
Been testing background replacement with LTX and it’s pretty fun how one simple shot can turn into something completely different.
Same character, same movement, whole new world.
And the best part? You can still take it further into your VFX and post-production workflow.
SingoKe retweeted
BACH 1.0 Pro is here.
Direct the scene. Shape every shot.
The AI Video Engine built for production, now supporting real projects in film, advertising, e-commerce, and animation.
Currently No. 2 on @ArtificialAnlys image-to-video leaderboard.
See BACH in action ↓
说好的,Skill 公开了。
古人说唱 Battle:在 Claude Code 里说一句"做一期 曹操 vs 刘备",它带你做完一整条 AI 说唱对决视频。
完全免费,开源。
github.com/10291124jcjc-dotc…
怎么用:
1. 装好 Claude Code
2. 把这句话发给 Claude Code:帮我安装这个 skill:github.com/10291124jcjc-dotc…
3. 装好后说:做一期 曹操 vs 刘备
然后它会:
· 写两段 battle 歌词,一句一个典故
· 给你 Suno 做歌的提示词
· 给你 Midjourney 定妆图的提示词
· 自动剪歌、拆分镜、写 35 条镜头提示词
· 调 MiniMax H3 逐句生成对口型镜头,自动检查口型
· 剪成片、上字幕、做封面、写发布文案
你只需要动手 4 次:确认歌词、Suno 做歌、Midjourney 出图、看草稿。
需要:Suno、Midjourney、MiniMax API Key。
一期视频生成费大约 15 美元。
我踩过的坑全写进去了:AI 偷偷重唱你的歌、镜头拍太远、台上没人、群演抢戏……
现在都是默认规则。
🤖 Made with AI
SingoKe retweeted
OmniHuman 2.0 is live.
It doesn't just sync lips — it understands your audio. Feed it any track, speech, music or dialogue, and it reads the tone, emotion, and every word, then performs it. Every syllable lands.
Now generate up to 5 minutes of video in a single continuous shot, up from 60 seconds, with no stitching required.
And 𝟰𝟬%+ 𝗰𝗵𝗲𝗮𝗽𝗲𝗿 — $0.12 → $0.07/sec. Better quality, lower cost.
Deeper understanding. Longer takes. Lower price.
byteplus.com/en/product/Omni…
SingoKe retweeted
Black Forest Labs launched FLUX 3 Image Klein, its new open-weight image model!
> "Customizable and fast. 5.6B parameters and runs on a single consumer GPU. Native 4K, bounding boxes, and pixel-perfect editing."
> Developers can post-train and fine-tune FLUX 3 Image Klein.
> It comes with a low-latency version for faster generation and self-hosting, as well as quantized versions - FP8, NVFP4, and INT8.
> The distilled model with native 4K will also be available on BFL APIs.
SingoKe retweeted
AI 配音这圈子,怕是又要被一个开源项目搅个底朝天
CosyVoice 直接更新到 Fun-CosyVoice3.0,我看完那组数据,后背是真有点发凉。
几秒参考音频丢进去,你的声音就能被克隆出来,气口、停顿、尾音都自然得很,不是那种塑料味十足的机器音。
更狠的还有三点:
1️⃣ 9 种语言外加 18 种以上中文方言,广东、闽南、四川、东北都拿得下
2️⃣ 延迟压到 150ms,边打字边出声
3️⃣ 0.5B 小模型,开源还能商用
做口播、课程、播客、小游戏配音的,真别错过:
🔗 github.com/FunAudioLLM/CosyV…
最离谱的是,这么能打的东西,居然是免费给你用的。
30 个感觉「非法」却合法的 AI 网站
1. raphael.ai/ — 免费无限 AI 生图,质量直逼 Midjourney
2. krea.ai/ — 实时 AI 生图,边画边出图
3. magnific.ai/ — AI 无限放大图片,细节自动补全
4. clipdrop.co/ — AI 一键去背景/补光/擦除
5. elevenlabs.io/ — AI 语音克隆,5 秒模仿任何声音
6. suno.com/ — 输入歌词自动生成完整歌曲
7. runwayml.com/ — AI 视频生成先驱,Gen-3 免费试
8. klingai.kuaishou.com/ — 快手可灵 AI,中文视频生成最丝滑
9. d-id.com/ — 一张照片+一段语音=说话的数字人
10. sadtalker.ai/ — 静态照片变说话视频
11. cursor.com/ — AI 写代码,免费额度够日常用
12. bolt.new/ — 说话就能建网站,零代码
13. v0.app/ — Vercel AI 前端生成器,描述即页面
14. replit.com/ — 浏览器里写代码+AI 辅助+一键部署
15. napkin.ai/ — 粘贴文字自动生成信息图
16. gamma.app/ — AI 一键生成 PPT,告别 PPT 地狱
17. notion.com/product/ai — Notion AI 写作、总结、翻译
18. uizard.io/ — 画草图 AI 自动生成真实网页
19. perplexity.ai/ — AI 搜索引擎,问任何问题秒给答案+来源
20. phind.com/ — 开发者专属 AI 搜索,代码问题一搜即解
21. otter.ai/ — 会议实时转文字,免费每月 300 分钟
22. opus.pro/ — AI 剪辑短视频,自动找高光片段
23. poe.com/ — 一个网站用 GPT-4o / Claude / Gemini
24. huggingface.co/spaces — 开源 AI 模型免费试玩场
25. remove.bg/ — 一键 AI 去背景,1 秒出图
26. cleanup.pictures/ — AI 擦除照片中任何物体
27. deepswap.ai/ — AI 换脸,1 分钟出片
28. musiio.com/ — AI 音乐标签+推荐
29. jasper.ai/free — AI 写作助手,免费版够日常
30. claude.ai/ — Anthropic 免费 AI 助手,长文处理天花板






















