Pinned Tweet
🎬 今天,我们开放了 OhMyMV —— 第一个端到端的 AI MV 生成平台
上传一首歌 + 一句话描述 → 30 分钟后拿到完整成片:
• 自动分镜、节拍对齐
• 数字人主唱 + 场景切换
• 可选歌词字幕烧录
• 一键交付 1080p
不是拼接 片段,是真正能发的 MV。
👉 ohmymv.com
🎬 Today we're launching OhMyMV — the first end-to-end AI music video platform
Upload a song + one-line prompt → get a complete MV in 30 minutes:
• Auto shot planning, beat-synced cuts
• Digital performer + scene transitions
• Optional lyric burn-in
• 1080p delivery-ready
Not clip stitching. Real MVs you can publish.
👉 ohmymv.com
我刚开始接触 AI 的时候还是用最早的版本,应该就是叫 1.0,我只记得它做海报效果很出众,当时还不支持中文,我用它出过很多海报,很怀念当时的学习热情啊
Introducing Ideogram 4.6: the new frontier of precise editing, now in native 4K.
A new image model with dramatically better realism and texture detail. Significantly less pixel drift outside your edits and much better color accuracy.
1
83
为什么说图像工坊是极为细心的一个产品设计?能给创作者带来什么帮助?
1. 图像工坊是提供MV工坊里的人物肖像图的生成能力的:
因为很多用户上传图像的时候都是人物的脸部特写或者上半身照,无法提供一个人物的全景照片,就算是全景照片往往会脸部占比太小导致模型无法完全复刻人脸信息,如果用户提供的人像图对于MV主角的身高、体型、发型、装饰的表达不到位,会导致后期MV制作阶段的人像参考信息缺失;我们在图像工坊里提供了最多4张人物参考素材上传档位,你可以上传不同拍摄角度、不同光线下的照片,再加上适当的文字描述作为辅助,比如人物的身高、体型、发型、装饰,根据经验,虽然GPT图像模型支持高达25张参考图,但最佳实践其实是4张,我们遵循了这一最佳实践;
2. 真实创作者的心智正是“固定主角”:
• 个人用户:往往想以自己(或理想化的自己)为原型打造专属 MV,无论是翻唱流行歌还是写民谣,他们希望出现在 MV 里的始终是同一个“自己”;
• AI 音乐人 / 虚拟博主:运营的是一个固定的虚拟偶像 IP(比如打造一个叫“Vivi”的朋克女歌手)。这个 IP 的身高、五官、体态是永久资产,每一首新歌发布的 MV 都要保持这个形象的连续性。
3. 用户粘性(Moat)大幅提升:
• 如果每次做 MV 都要临时上传零散自拍、重新抽脸,不仅麻烦,而且每次生成的歌手长相身材都会漂移;
• 一旦用户在【图像工坊】里成功沉淀下了满意的“五视角基准资产卡”,后续做 10 首、20 首 MV,直接在资产库里“点选已有主角”。这种“资产复用”给用户带来的爽快感和对平台的依赖度是极高的!
4. 清晰的“前后厂”分工:
• 图像工坊 = 艺人经纪/孵化部:用户在这里精细打磨演员的骨相、身高、身材比例,生成标准五视角底模,存入资产库。
• MV 工坊 = 电影拍摄棚:用户从资产库选定这位艺人,交给导演安排服装、灯光、分镜,直接开拍。
另外,我们的suno音乐工坊也是一样的设计思路,既是在整首MV制作的提前体验,也是我们积累音乐资产的关键环节
我希望未来我们的用户能够在评论区留下你们的想法和意见建议,我们一起把ohmymv做好,我并不奢求赚很多钱,只希望把一首歌变成一个完整MV这个体验能做到最好
4
2
295
看Carbon导演有没有好好干活
看这个细节,看来它真的是完整的理解了音乐的歌词以及我选的北欧极简清冷导演风格,里面的场景就是对应的balcony,画面风格的确也是北欧的峡湾风格;
我没去过北欧,不知道北欧的夏天是不是也得穿毛衣呢?
1
86
Some friends asked why I don't give away more points for everyone to try out.
I'm sorry. A full MV requires substantial computing power costs. We are not a large company and do not have a big marketing budget. Unfortunately, I cannot offer more free points.
Even so, the 100 complimentary points still let you try all features on the site, including image generation, music creation, and 15-second storyboard video generation.
If you still have doubts about us, you can get a full unconditional refund within 14 days after purchasing a point package. If your MV task fails to generate, your points will be fully refunded. I can guarantee this, as this is how our system is configured.
Lastly, if you run into any other issues, you can submit a support ticket on the website. We usually reply to tickets within 24 hours. You can also reach me directly on X, or even file a complaint against me.
🎬 今天,我们开放了 OhMyMV —— 第一个端到端的 AI MV 生成平台
上传一首歌 + 一句话描述 → 30 分钟后拿到完整成片:
• 自动分镜、节拍对齐
• 数字人主唱 + 场景切换
• 可选歌词字幕烧录
• 一键交付 1080p
不是拼接 片段,是真正能发的 MV。
👉 ohmymv.com
🎬 Today we're launching OhMyMV — the first end-to-end AI music video platform
Upload a song + one-line prompt → get a complete MV in 30 minutes:
• Auto shot planning, beat-synced cuts
• Digital performer + scene transitions
• Optional lyric burn-in
• 1080p delivery-ready
Not clip stitching. Real MVs you can publish.
👉 ohmymv.com
46
同一首歌,同一个人物,同一种导演风格,都是北欧极简清冷风;
7美元、不到30分钟,制作任意人物、任意歌曲,无需提示词,无需任何制作技巧
过程中我除了上传歌曲和人物图片、选择导演风格,我什么都没做
大家如果觉得效果还可以,点个赞让我知道,谢谢
57
同一首歌,同一个人物,我重新制作了一版,花了25分钟;人物身材不会变形、减少了画面中人脸正面远景的比例
完整的视频在下载中,马上发出
The same song and the same character, just in a different style. I’ve added several great director styles for everyone to pick directly with a simple click of the mouse.
The portrait in this version has flaws, and I’m not going to hide it. The reason is straightforward: when a human face generated by MiniMax H3 occupies a small proportion of the overall frame, facial distortion occurs. You can see there is no such issue in close-up shots. Besides, the character’s body is warped and looks unnatural. I can fix this tomorrow by adding constraints.
由于现在的视频模型只能生成最多15秒、30秒的视频片段,在一个时长超过200秒的长视频制作过程中,不管是人物的一致性还是画面的一致性,一直是困扰各位创作者的难题
我的个人看法是:风格一致性要控制视频片段的风格,而每一个视频又是由分镜图控制的
我的解决方法是:由一个能力强大的LLM模型作为总导演,它负责一次性输出所有的分镜图绘图提示词以及图生视频的提示词,目前来看,只有这么做才能做到视频风格、人物一致性的完全统一
大家如果在制作方面有更好的方法,多点来指导指导小弟
1
145
ohmymv的人物一致性有多夸张?
第四张粉色的图是我输入的MV人物主角参考图,中间这两张是我在制作的北欧极简清冷风格的MV分镜图
你留意到女孩左眼眼角下的痣了吗,大家把图片放大来看,每张图都有这个特征,GPT 2.6图像模型在现在真是无敌的
🤖 Made with AI
I have fixed this issue. The root cause was that the character images imported in the early stage only contained 60% of the character's upper body. Combined with the 21:9 video aspect ratio, this resulted in the "big head doll" effect. My solution is to force the character five-view anchors for faces and bodies during the storyboard generation stage. The GPT model follows instructions very well, so similar issues are unlikely to occur in the future.
In addition, the MiniMax H3 model has a flaw: if the face occupies too small a proportion of the frame, facial distortion will occur. This problem exists in all video generation models. My workaround is to avoid creating wide shots with front-facing faces. Since there is no complete solution for the time being, we can bypass this scenario.
I noticed today that several projects seem to be working on resolving this issue. I am sharing the reference materials here for anyone who needs them. Feel free to share better solutions for discussion and feedback.
Relevant resources: github.com/drozbay/MaskVidEx… github.com/Carasibana/ComfyU… github.com/ylchen333/ComfyUI…
A畜明天要上市了,我预计它的表现会很好
We’ve priced our IPO at $83 per share, with trading expected to begin on Nasdaq tomorrow.
anthropic.com/news/anthropic…
Legal disclaimer: A registration statement relating to the Class A common stock was declared effective by the Securities and Exchange Commission on November 18, 2026. The offering is being made only by means of a prospectus. Copies of the final prospectus may be obtained, when available, from: Goldman Sachs & Co. LLC, Attention: Prospectus Department, 200 West Street, New York, NY 10282, by telephone at (866) 471-2526, or by email at [email protected]; or Morgan Stanley & Co. LLC, Attention: Prospectus Department, 180 Varick Street, 2nd Floor, New York, NY 10014, or by email at [email protected]
24
如何正确理解SEO?
面对包含成百上千个概念的 SEO 知识库,初级开发者感到无所适从是极其正常的。
市面上充斥着大量互相矛盾的说法:大厂教你做几千页架构,黑帽教你刷外链,技术极客教你把性能优化到 100 分,自媒体教你批量用 AI 灌几千篇文章。如果你把所有做法并列在一起,不仅会陷入严重的焦虑,还会把精力分散在无数边际效应极低的事情上。
要想真正理清 SEO,最重要的是化繁为简:看清本质、认清阶段、抓住关键少数。
本质认知:如何用一句话理解 SEO?
请完全忘掉那些复杂的术语。从商业本质上看,Google 只是一家撮合搜索引擎:
用户的目标:带着问题来搜,点进第一个结果,问题完美解决,关掉网页走人。
Google 的目标:把最快、最好、最能解决问题的网页排在前面,以维持自己的垄断地位。
你的目标(SEO):向 Google 证明,针对用户的某个具体需求,你的网页就是全网最好的答案。
Google 评估一个网站,无论算法怎么更新,底层永远只问三个问题(SEO 铁三角)
20
我推特发帖的浏览量很低,但回复别人帖子的浏览量和互动率却很高,经常有上万的浏览
我把这个疑问告诉 Grok
它说我的推特帖子是低质量的纯内容输出,别人不知道怎么接话;而我回复别人帖子的评论却是高质量的,容易让别人有共情
我觉得它说的很对,别人是在经营推特,我是在玩
23
I have fixed this issue. The root cause was that the character images imported in the early stage only contained 60% of the character's upper body. Combined with the 21:9 video aspect ratio, this resulted in the "big head doll" effect. My solution is to force the character five-view anchors for faces and bodies during the storyboard generation stage. The GPT model follows instructions very well, so similar issues are unlikely to occur in the future.
In addition, the MiniMax H3 model has a flaw: if the face occupies too small a proportion of the frame, facial distortion will occur. This problem exists in all video generation models. My workaround is to avoid creating wide shots with front-facing faces. Since there is no complete solution for the time being, we can bypass this scenario.
I noticed today that several projects seem to be working on resolving this issue. I am sharing the reference materials here for anyone who needs them. Feel free to share better solutions for discussion and feedback.
Relevant resources: github.com/drozbay/MaskVidEx… github.com/Carasibana/ComfyU… github.com/ylchen333/ComfyUI…
🤖 Made with AI
The same song and the same character, just in a different style. I’ve added several great director styles for everyone to pick directly with a simple click of the mouse.
The portrait in this version has flaws, and I’m not going to hide it. The reason is straightforward: when a human face generated by MiniMax H3 occupies a small proportion of the overall frame, facial distortion occurs. You can see there is no such issue in close-up shots. Besides, the character’s body is warped and looks unnatural. I can fix this tomorrow by adding constraints.


























