AI 探索者 · 独立开发者 AI视频工作台:https://xitter.cf/t.co/wO1stT6iaY

Joined August 2014
时隔一周这个 MiniMax H3 做的视频在抖音跑了200w播放 只能说抖音的手机用户还是多啊 就这画质电脑上刷到就应该被划走了
再用 Qwen-image-2.1 出图 MiniMax H3 出视频 零成本整个剧《送礼的艺术——这卡可不好买啊》 不得不说H3在做现代写实风画质还是差点意思。 出提示词前的定稿剧本↓
4
1
15
1,958
我的 Reelbench Studio 工作台被人挂闲鱼卖了 然后买家来问我为啥下载不了,我一查账号都没注册过 问了才知道是闲鱼买的 因为是毫无保留的交付源码,所以无法避免这种情况 我是该高兴还是该难受🤣
2
2
951
说干就干,先整个视频宣传下我是搞AI的! 制作流程: -把quote推文中的视频丢给Claude Code拉片 -然后根据自己需求改写新剧本 -再微淘新剧本 -然后调用MiniMax H3 官方skill 写提示词 -如有需要,做一个首帧图 最终视频制作素材分享在评论区。
做 AI 视频的,要不只盯着短剧市场 格局打开,营销视频也一样值得深耕 看看这个账号,直接借助爽剧名场面为剧本 融合自己卖钢材写剧本,做段子也有不错的曝光 重点是剧本直接借鉴爽剧名场面没成本 视频用 MiniMax H3 做的话也没什么成本 然后结合任意实体行业做洗稿改编 具体效果看下边视频,请做好笔记! 考考你,这个视频卖的是什么钢?
8
38
3,824
古代人看到一车 5090:他是搞AI的! 第一段提示词和首帧图: photorealistic cinematic shot, natural soft diffused lighting, realistic skin subsurface scattering, natural skin texture, no oily skin, subtle skin pores, natural eye catchlight, soft shadow transition, physically based rendering, PBR, real fabric texture, natural material reflection, fine film grain, shallow depth of field, smooth motion, stable facial features, consistent face across frames, no painted texture, no thick brush strokes, no oil painting effect, live action, real world scene, natural color grading, moderate contrast integrated_multimodal_description: [Shot 1] Live-action, cinematic Chinese costume comedy short drama that keeps the look of <Picture 1>: warm low morning sunlight from the left, long soft shadows on the flagstones, natural color and a shallow depth of field, played at a fast, punchy short-drama pace: every line comes right after the previous one with almost no pause, everyone speaks fast and loudly, and every movement is quick, snappy and energetic, with no lingering beats. The video begins exactly as <Picture 1>: a first-person POV from the eye level of the protagonist, a young man who works in AI; his face is never seen, and only his own hands appear at the bottom corners of the frame: his two hands in the sleeves of a charcoal-grey modern knit sweater, gripping the worn wooden handles of an old wooden two-wheeled handcart with large spoked wooden wheels. The cart is loaded high as in <Picture 1>: a stack of four silver-and-black rack-mount servers on the left, and on the right three large black NVIDIA retail boxes with bright green stripes and big white-and-green letters reading "GEFORCE RTX 5090", next to a stack of bare two-fan RTX 5090 graphics cards. The setting is the busy, sunlit ancient Chinese market street of <Picture 1>, with dark wooden shophouses, cloth awnings, basket stalls, a flagstone pavement and a tall wooden gate tower at the far end under a clear blue sky. The protagonist keeps pushing the cart forward the whole time, and the camera moves forward steadily with a slight walking sway, never stopping until the emperor arrives. Right at the start, a skinny young man about twenty in a short dark-blue linen jacket with a white cloth band tied around his head (S1) bursts into the frame from beyond the right edge, skids to a stop right beside the cart, stares wide-eyed at the boxes, jabs a finger at the cart and shouts fast in an excited young male voice: <d>[Chinese] 他是搞AI的!快看啊!</d> He steps back to the right to let the cart pass, still pointing at it. The townspeople in plain linen robes walking on both sides whip around in an instant to stare at the cart and step aside to make way. At once, on the left behind the man with the woven basket, the woman with her black hair in a bun, wearing a mauve-purple linen robe (S2), slaps both hands to her cheeks and shrieks in a shrill, high middle-aged female voice: <d>[Chinese] 他是搞AI的!</d> The man with the topknot and the woven basket on the left jerks his head around and freezes, staring at the boxes. Right away, a skinny old man with a long white beard, a wrinkled face, a dark brown robe and a small black cloth cap (S3) darts in from the left edge, snaps forward close to the lens, gapes at the RTX 5090 boxes and cries in a creaky, trembling old male voice: <d>[Chinese] 什么?居然是搞AI的?</d> He jumps aside, and the crowd ahead scatters to both sides of the street in a flash, opening a path. Out of the crowd not far ahead bursts the emperor, a stout middle-aged man about fifty with a round face, a neat black moustache and a short black goatee, thick black eyebrows and a hearty grin, wearing a bright golden-yellow imperial dragon robe embroidered with golden dragons and a small gold crown on his topknot (S4), hitching up his robe and running toward the cart with big strides, followed close behind by the young princess about eighteen with a delicate oval face and a gentle smile, her black hair in an elegant updo with a gold hairpin, wearing a pale pink silk hanfu with long flowing sleeves (S5), and four palace guards in gold armour and helmets. As he runs, the emperor (S4) shouts in a deep, booming middle-aged male voice: <d>[Chinese] 真是搞AI的!</d> Through this whole shot the servers and the RTX 5090 boxes stay on the cart in the lower part of the frame. [Shot 2] At 00:07.500, the shot cuts to the same first-person POV a moment later, on the same sunlit market street in the same warm low morning sunlight. The protagonist has let go of the cart and stepped forward, so the cart, the servers and the boxes are out of the frame; only his two hands in the charcoal-grey knit sweater sleeves show at the bottom of the frame. The round-faced emperor with the neat black moustache and short black goatee, in the bright golden-yellow dragon robe and the small gold crown (S4), stands right in front of the lens in a close shot, with the princess with the elegant updo in the pale pink silk hanfu (S5) just behind him, the four palace guards in gold armour behind them and townspeople watching from both sides of the street. The emperor (S4) instantly grabs the protagonist's right hand in the charcoal-grey knit sleeve with both hands, pumps it rapidly and rattles off fast and joyfully in the same deep, booming voice: <d>[Chinese] 女儿嫁你了!皇位也给你了!大辽那边别去了!明天就登基!哈哈哈!</d> He throws his head back with a big belly laugh and flings his arms wide. The princess (S5) immediately darts up beside him toward the lens with a quick flick of her long pink sleeves, and says sweetly in a soft young female voice: <d>[Chinese] 夫君~</d> She quickly lowers her eyes with a faint pink blush and smiles. The camera holds a nearly static POV. overall_soundscape: The bustle of a sunny market street, wooden cart wheels rattling over flagstones all the way, the faint whirr of server fans, gasps, hurried footsteps, clanking armour and the emperor's hearty laughter. non_diegetic_music: N/A
1
2
646
古代人看到一车 5090:他是搞AI的! 第二段提示词: photorealistic cinematic shot, natural soft diffused lighting, realistic skin subsurface scattering, natural skin texture, no oily skin, subtle skin pores, natural eye catchlight, soft shadow transition, physically based rendering, PBR, real fabric texture, natural material reflection, fine film grain, shallow depth of field, smooth motion, stable facial features, consistent face across frames, no painted texture, no thick brush strokes, no oil painting effect, live action, real world scene, natural color grading, moderate contrast integrated_multimodal_description: [Shot 1] Live-action, cinematic Chinese costume comedy short drama with natural soft daylight, natural color and a shallow depth of field, played at a fast, punchy short-drama pace: every line comes right after the previous one with almost no pause, everyone speaks fast and loudly, and every movement is quick, snappy and energetic, with no lingering beats. A first-person POV shot from the eye level of the protagonist, a young man who works in AI; his face is never seen, and only his own hands appear at the bottom edge of the frame: his two hands in the sleeves of a plain grey modern hoodie, a black smartwatch on his left wrist, gripping the wooden handles of a wooden two-wheeled handcart. From the very first frame, the front of the cart, close to the lens and filling the lower middle of the frame, is piled high with a tall stack of large black RTX 5090 graphics card retail boxes in two neat columns: glossy black cardboard boxes, each with a big photo of a black two-fan graphics card on the front, a bright green stripe along the edge and large white letters reading "RTX 5090". Behind the boxes sits an open black computer server with small cooling fans spinning and green status lights blinking. The setting is a busy ancient Chinese market street in daytime: dark wooden shophouses with grey tiled roofs, strings of round red paper lanterns overhead, a grey flagstone pavement, and crowds of townspeople in plain linen robes. The camera moves forward briskly with a slight walking sway, and the townspeople ahead whip around in an instant to stare at the boxes. On the left, the young man in a grey-blue linen robe with a dark cloth headscarf (S1) jabs a finger at the cart and shouts fast in an excited young male voice: <d>[Chinese] 他是搞AI的!快看啊!</d> At once, on the right, the plump middle-aged woman with her hair in a bun, wearing a faded pink jacket and a long pink skirt (S2), slaps both hands to her cheeks and shrieks in a shrill, high middle-aged female voice: <d>[Chinese] 他是搞AI的!</d> Right away, the skinny old man with a long white beard, a wrinkled face, a dark brown robe and a small black cloth cap (S3) darts in from the left, snaps forward close to the lens, gapes at the stack of RTX 5090 boxes and cries in a creaky, trembling old male voice: <d>[Chinese] 什么?居然是搞AI的?</d> [Shot 2] At 00:04.500, the shot cuts to the same POV a moment later, the cart stopped, the stack of black RTX 5090 boxes still in the lower middle of the frame. The old man has jumped aside and the crowd scatters to both sides in a flash. Out of the crowd only a few steps away bursts the emperor, a stout middle-aged man about fifty with a round face, a neat black moustache and a short black goatee, thick black eyebrows and a hearty grin, wearing a bright golden-yellow imperial dragon robe embroidered with golden dragons and a small gold crown on his topknot (S4), hitching up his robe and dashing to the cart in two quick strides, followed close behind by the young princess about eighteen with a delicate oval face and a gentle smile, her black hair in an elegant updo with a gold hairpin, wearing a pale pink silk hanfu with long flowing sleeves (S5), and four palace guards in gold armour and helmets. The emperor (S4) shouts in a deep, booming middle-aged male voice: <d>[Chinese] 真是搞AI的!</d> He instantly grabs the protagonist's right hand in the grey hoodie sleeve with both hands, pumps it rapidly and rattles off fast and joyfully in the same deep, booming voice: <d>[Chinese] 女儿嫁你了!皇位也给你了!大辽那边别去了!明天就登基!哈哈哈!</d> He throws his head back with a big belly laugh and flings his arms wide. The princess (S5) immediately darts up beside him toward the lens with a quick flick of her long pink sleeves, and says sweetly in a soft young female voice: <d>[Chinese] 夫君~</d> She quickly lowers her eyes with a faint pink blush and smiles. The camera holds a nearly static POV. overall_soundscape: The murmur of a crowded market street, cart wheels rattling over flagstones, the high whirr of cooling fans, gasps, hurried footsteps, clanking armour and the emperor's hearty laughter. non_diegetic_music: N/A [Chunk 2] integrated_multimodal_description: [Shot 1] Live-action, cinematic Chinese costume comedy short drama with natural soft daylight, natural color and a shallow depth of field, played at a fast, punchy short-drama pace: every line comes right after the previous one with almost no pause, everyone speaks fast and loudly, and every movement is quick, snappy and energetic, with no lingering beats. A first-person POV shot from the eye level of the protagonist, a young man who works in AI and is now the new emperor, seated on a golden dragon throne; his face is never seen, and only his own forearms appear at the bottom edge of the frame: his two forearms in wide bright golden-yellow dragon-robe sleeves embroidered with golden dragons, a black smartwatch on his left wrist, resting on the throne's carved gold armrests. The throne stands at the top of a white marble terrace in front of a grand palace hall with a golden tiled roof, looking down a long white marble central ramp onto a vast stone-paved palace courtyard with colourful banners along its sides. Close to the lens stands the former emperor, a stout middle-aged man about fifty with a round face, a neat black moustache and a short black goatee, thick black eyebrows and a hearty grin, now wearing a dark purple official's robe with a round embroidered badge on the chest and a black winged official hat with two long flat wings sticking out to the sides (S4). He scurries up in quick little steps, bows twice rapidly, rubs his palms together fast and leans toward the lens with a fawning smile, rattling off in a deep, booming middle-aged male voice, now sweet and wheedling: <d>[Chinese] 我说到做到了吧?你看看能不能给我开个号,一个亿 token,要满血版的。我写奏折用。</d> The protagonist's right hand in the golden dragon-robe sleeve flicks up from the armrest in one quick casual wave, and the unseen protagonist behind the camera (S6) says quickly in a relaxed young male voice: <d>[Chinese] 好说好说。</d> The camera holds a static shot. [Shot 2] At 00:06.500, the shot cuts to the same first-person POV from the throne, and the camera tilts up with small amplitude at fast speed over the courtyard: hundreds of court officials in red, blue and green robes and black hats drop to their knees all at once in neat rows on both sides of the white marble ramp, and the kneeling officials (S7) knock their heads to the ground in quick unison and shout together in a huge chorus of male voices: <d>[Chinese] 皇上万岁万岁万万岁!</d> Right on the last word, the thin eunuch with a smooth beardless face, wearing a dark red robe and a soft black hat, carrying a white horsetail whisk (S8), sprints up the last steps of the terrace from the right toward the camera and calls out fast in a high-pitched, breathless male voice: <d>[Chinese] 报——大辽皇帝求见!说愿意给您两个皇位!</d> The camera holds a static shot. [Shot 3] At 00:11.000, the shot cuts back to the close first-person POV from the throne. On the right side of the frame, close to the throne, the former emperor in the dark purple official's robe and the black winged official hat, with his round face, neat black moustache and short black goatee (S4), snaps his head around toward the lens in an instant, eyes wide with alarm, and blurts out in the deep, booming voice, now panicked: <d>[Chinese] 皇上,您才登基一天啊!</d> The protagonist's right hand in the golden dragon-robe sleeve flicks up from the armrest, palm down, in one quick dismissive wave, and the unseen protagonist behind the camera (S6) says quickly and calmly in the relaxed young male voice: <d>[Chinese] 让他排队。西夏和大金昨晚就来了。</d> The former emperor's jaw drops at once and he freezes, staring at the lens. The camera holds a static shot. overall_soundscape: An open palace courtyard with a soft breeze and banners flapping, the rustle of the purple silk robe, the deep rumble of hundreds of robes as the officials kneel, the echoing chorus across the stone courtyard, the eunuch's quick slapping footsteps on marble, and a stunned silence at the end. non_diegetic_music: N/A
1
395
做 AI 视频的,要不只盯着短剧市场 格局打开,营销视频也一样值得深耕 看看这个账号,直接借助爽剧名场面为剧本 融合自己卖钢材写剧本,做段子也有不错的曝光 重点是剧本直接借鉴爽剧名场面没成本 视频用 MiniMax H3 做的话也没什么成本 然后结合任意实体行业做洗稿改编 具体效果看下边视频,请做好笔记! 考考你,这个视频卖的是什么钢?
34
23
2
182
27,991
做白膜视频的可以关注下抱抱脸上这两个开源项目 里边有近 4000 个开源 3D 模型,CC BY 4.0,可商用 huggingface.co/datasets/scen… huggingface.co/spaces/nvidia… 第一个 Scenario 开源的的连 prompt 都分享出来了 第二个英伟达开源的,工业元素偏多。拿来做工厂可视化相关挺不错的
2
34
2,671
烁皓 retweeted
我做了一个我自认为全世界最好的雅思背词网站,完全免费,完全开源。🔥🔥🔥🔥🔥 7,298 个雅思词,每个词一幅原创插画,英音、美音、例句全都有接近真人的配音。学会一个词,你的词汇星图上就点亮一颗星。 完全免费!!不用注册,打开就能背,代码和词库大家随便用。 希望每个人都能考过雅思。学会英语,去接触更广阔的天地,获取更多的知识。 👉 0xlei.tech/words 开源:github.com/designfuchen/word…
637
1,022
139
5,250
528,760
MiniMax H3 换脸王来了 BFS - Best Face Swap 最多可换三张脸,Lora地址: civitai.red/models/2027766/b… GitHub 有人分享了个对应工作流,可以直接用: github.com/Bambushu/h3-recas…
6
18
310
19,966
看看这口播效果如何? 不管你信不信,反正是零成本搓出来的。
做口播的用 Drama 3 这克隆效果完全没毛病吧? 《如果说和我聊天送Claude账号你还会不理我吗》 同一台机器,同一段参考音频和评测文案: Drama 3 int4 量化版:4.6s生成 15.8s音频 IndexTTS 19.5s生成 19.2s音频 跳17s直接听IndexTTS效果,感觉在这个方向上Drama 3完胜!
9
2
1
23
5,837
啥也别说了兄弟们 家里还有啥值钱的都拿去当了买5090吧 1min 跑 10s视频 且不说画质,就单纯拿来做预抽卡也不亏
FastH3-QAD-v2 running on my RTX 5090 (Windows 11, WSL2). 5 second clip with sound at 832x480: 10.07 s 10 second clip with sound at 1344x768: 69.06 s Settings for every run are on screen.
7
4
1
46
6,642
做口播的用 Drama 3 这克隆效果完全没毛病吧? 《如果说和我聊天送Claude账号你还会不理我吗》 同一台机器,同一段参考音频和评测文案: Drama 3 int4 量化版:4.6s生成 15.8s音频 IndexTTS 19.5s生成 19.2s音频 跳17s直接听IndexTTS效果,感觉在这个方向上Drama 3完胜!
上个月刚把知名的开源 TTS 引擎试了个遍 这个月 Fish Audio 就开源了新一代的 Drama 3 主打多角色配音和长文本 用来做个读剧本的工具应该很不错,主要他有这三点优势: 第一,一段文本里用 <|speaker:N|> 标签就能分出最多 16 个角色,广播剧、有声书一次就能生成。 第二,可以在台词里直接插 [sighs]、[laughs]、[whisper] 这类标签,叹气、笑声、耳语会落在你标的位置。 第三,它把编辑当成续写:一句话里只改一个词,比如“周三”改成“周四”,把之前的生成结果一起传回去,就能只改这个词,其余部分的声音和语气保持不变,不用整句重来。 另外它支持用几秒参考音频加文字稿克隆声音,号称支持 83 种语言,输出 44.1kHz。 模型是 35B 参数的 MoE,每个 token 只激活 3B,推理走 SGLang-Omni 或 vLLM-Omni 官方在 H200 上测的单路首包约 89ms,实时率 0.13,也就是生成 1 秒音频约 0.13 秒 完整版要 48GB 以上显存,另外有 int4 量化版,权重约 21GB huggingface.co/fishaudio/dra…
9
2
1
30
8,914
Drama 3 和 IndexTTS 评测文案: 如果说跟我聊天送 token 呢?你还会不理我吗?再送一个Claude账号,再送一根内存条,再加一张5090显卡。你还能做到对我已读不回吗,哼。
1
948
有朋友问:怎么才能做出高质量的视频? 我觉得最主要的就是积累自己的提示词库。 有个开源的浏览器插件「提示词导演 PromptDirector」就是专门用来本地化管理提示词的。 github.com/wchao6891/PromptD… 作者还整理了 1000+ 精选提示词案例 可以一键导入自己本地的提示词库 开发者是@Chainwai6891,也是我视频交流群的群友。
16
32
4
226
13,285
FastH3 Omni 可以在 DGX Spark 和 Apple M系列芯片上运行了 苦于只有苹果一直没体验上MiniMax H3的用户有福了 虽然慢了点,至少能尝尝咸淡不是嘛🤣
FastH3 Omni 开源了! 能在消费级显卡自部署的超级加速版MiniMax H3! 之前的 FastH3 V2 已经能在消费级显卡上跑了, 但那只是开胃菜——只支持文生视频。 这次可以上全能参考: 图片 + 视频 + 音频 都能塞进去, 一次生成带同步音频的视频,角色、动作、声音全可控。 huggingface.co/collections/F…
5
8
2,841
又是一篇十万加!其实制作、画质也不是很重要对吗? 和下边推文同时发的,剧本也都分享出来了。历时4小时! 就 MiniMax H3 这个画质我是真没有想到。 但实时就摆在这里:剧本质量的重要性远大于制作!
再用 Qwen-image-2.1 出图 MiniMax H3 出视频 零成本整个剧《送礼的艺术——这卡可不好买啊》 不得不说H3在做现代写实风画质还是差点意思。 出提示词前的定稿剧本↓
26
5
64
9,077
再用 Qwen-image-2.1 出图 MiniMax H3 出视频 零成本整个剧《送礼的艺术——这卡可不好买啊》 不得不说H3在做现代写实风画质还是差点意思。 出提示词前的定稿剧本↓
既然大家爱看,就再做个评测视频。 这次不分享提示词,分享全流程的创作思路。 剧本: - 在抖音搜00后,选自己觉得有趣的剧情情节,总结成一句话。可以复制链接到豆包,让豆包总结。 - 把上边总结的内容发给 Agent,让他放在古代场景进行二创并写出完整的剧本。检查核对剧本让Agent修改到自己满意 提示词: 基于 MiniMax H3 官方提示词 skill,二次封装个自己的提示词 skill,具体步骤: - 之前「xitter.cf/eternityspring/status/…」这个推文有链接,复制给Agent - 让Agent参考 ComfyUI-H3-Continuum 的提示词要求重新封装一个新的skill 如果是做写实风格的视频,建议把评论区的画质提升提示词发给Agent,让他融入到新skill中。
15
16
2
140
45,590
《送礼的艺术——这卡可不好买啊》剧本: 人物:领导(技术总监,AI 发烧友)、小王(老实人,送什么都报价钱)、小李(会送礼) 场景:节前,领导办公室 【显卡】 小王:领导,过节了送您一张 5090 显卡,两万多呢。 领导:破费了,不过这卡我不太用,显存太小。 小李:领导,我有个朋友在机房做运维,知道您喜欢跑模型,给我拿了几张淘汰下来的卡,叫什么 H100。我也不懂,您帮我看看是不是好东西? 领导(捧起卡,两眼放光):哎呀,这卡可不好买啊。我今晚高低得跑跑。 【会员】 小王:领导,给您充了一年 GPT 会员,听说特别好用。 领导:行吧……以后别乱花钱充这东西,我有好几个号。 小李:领导,我去旧金山转了一圈,托人搞了个 GPT 新模型的内测码。我也不懂真假,就觉得挺有意思,您帮我验验? 领导(看着手机乐了,连忙招手):哎呀,这个不错。来来来坐下,我跑个 demo 给你看看。 【电脑】 小王:领导,过节了给您带了台 AI 笔记本,顶配,三万多呢。 领导:破费了,不过我平时用得不多。 小李:领导,我路过电脑城,看这几根内存条挺有意思,也不贵。想着您平时跑本地模型,应该用得上。 领导(拿起内存条对着光看):这东西选得不错啊。你是真懂 AI。
1
2,509