迷墙

迷迷墙

[TechHalla @techhalla]

Realistic footage with Seedance 2.0 👇

Prompt:

This is handheld documentary footage recorded on an early-2000s consumer DV camcorder at the construction site of the Great Pyramid of Giza. The footage feels like real, imperfect documentation of one of the largest building projects in history.

The recording shows hundreds of workers moving massive stone blocks on a large construction ramp. Teams of men are pulling ropes attached to sledges with heavy limestone blocks. Overseers with sticks walk along the lines giving orders. Dust fills the air. In the background, the partially built pyramid rises against the sky. The work is extremely physical and organized.

The camera moves along one of the work gangs, following a team of workers as they pull a large block up the ramp. It occasionally tilts up to show the height of the structure. There are natural cuts between different groups of workers and close views of the ropes straining and the stone moving slowly. The movement of the camera feels like someone walking among the workers documenting the labor.

The handheld camera shows constant shake, drifting framing, autofocus hunting in the bright sunlight and dust, exposure changes, motion blur, and the typical look of filming heavy physical work with an old DV camcorder. The person filming is clearly moving through the active construction site.

Natural sound only: the sound of hundreds of men shouting and grunting in rhythm, ropes creaking under tension, stones scraping on the ground, and general noise of a massive construction site. No music.

The result must feel like authentic, raw footage of the construction of the Great Pyramid, captured on an old camcorder.

[余温 @gkxspace]

卧槽，牛逼啊，AI 视频最烦人的抽卡问题，终于有人从根上动手了！！！

以前你只能用文字描述画面，谁在左谁在右、镜头从哪拍，全靠模型猜，时间和钱都耗在了里面。

前段时间有个老外用 Blender 摆几个方块排好走位，再喂给视频模型，出来的动作和运镜准得离谱，直接几千赞。

大家看完才反应过来：给 AI 视频加空间约束，比写一万字 prompt 都管用。

问题是那套流程得会 Blender，门槛直接劝退一大半人。

@TopviewAIhq 把这步直接做进了产品：
3D Shot Composer，生成前在 3D 里摆角色、放道具、调相机取景，排好了再出片。角色位置、空间关系、构图全都自己控制。

抽卡次数少了，废片也少了，AI 视频拼到最后，拼的就是可控性！

[Yangyi @yangyi]

把孩子的画作 创作成AI视频
这个赛道挺有情绪价值的
相似的变体应该挺多的
比如孩子的绘画馆 一张自己的手绘稿做变体收藏
在这些基础手稿上做出丰富的东西

[Kōda @aimikoda]

150 Free Character Sheets

I recently realized I've created around 150 characters so far. Some came from shorts and some are still waiting inside stories haven't told yet.

A mixed collection of mostly stylized characters across 2D, 3D, realistic and cinematic looks, generated with Midjourney and GPT Image 2.

Today I’m sharing them as downloadable character sheets, completely free to use. Feel free to use them in your own stories, AI videos, game ideas or visual experiments.

Download here: https://t.co/PDW74JcrpK

[AIスタジオワンルーム（AIアニメ、動画、漫画） @studio_oneroom]

みなさん、Fable5って何に使いました？
私は、結局、プロンプトやストーリーに費やすことが多くて他にはあまり出来ませんでした。

でも今回、かなり徹底してプロンプト周りのスキルを再構築したので結構、自分好みの映像になった気がします。

この動画はテストを兼ねて完全にFable5に任せて作りました。ストーリー的には意味がわかりませんが映像的にはかなり自分好みにカスタマイズ出来ました。

引用元の『攻殻機動隊 THE GHOST IN THE SHELL』風の動画も今回、バチバチにカスタムしたスキルで作っています。

動画：Higgsfield x Seedance2.0
画像：Midjourney v8.1
音楽：Suno

[Kōda @aimikoda]

Ritual
Midjourney v8.2 + Seedance 2.0 Mini

--sref 2343335488 --profile txgf1m1 --stylize 1000 --preview

Prompt:
Start from the @[ref image] as the first frame. She holds the same frontal pose for a beat, then begins a solo ritual dance with precise hand movements, grounded steps, and slow turns. Crimson energy ribbons rise from her face and hands, while her gold ornaments and fur move with the rhythm. End in a strong centered pose, echoing the opening frame with red aura suspended around her. 9 cuts.
High-end stylized 2D anime illustration fused with textured painterly movement, sacred ritual fashion, and emotionally charged contemporary dance. Slow, precise body isolations, sharp wrist turns, controlled shoulder rolls, and grounded footwork create a ceremonial dance language that feels ancient yet modern. Crimson tear-like energy lifts from the face and trails through the air as thin liquid ribbons, while gold relic fragments orbit the body in delayed rhythm and dark teal shadow smoke folds around each turn. The choreography alternates between stillness and sudden explosive accents, with dramatic close-ups, low angles, spinning perspective shifts, and elegant silhouette-driven framing. Soft rose haze, glowing amber eyes, ivory fur textures, weathered gold details, and hand-painted brush grain create a sacred, melancholic atmosphere while cinematic lighting preserves weight, intimacy, and performance intensity.

[悉尼奶爸 SydneyDaddy 雪梨奶爸 🇦🇺🇮🇱 @SydneyDaddy1]

Anthropic今天发布的一个视频炸了

他们发现AI模型跟人一样，有着潜意识甚至无意识

他们找到一种数学办法来窥探这个无意识，并发现AI会无意识地欺骗，甚至飙脏话

我觉得这些无意识或许就是最接近于灵魂的东西

问题来了：AI知道了自己有无意识，也知道了人类在窥探他的无意识，他又会怎么做呢？

[神奇小喷菇AIGC @xpg0970]

告别盲目抽卡！AI分镜参考新办法！

AI 人绝不私藏，5个宝藏网站，用关键词从电影里搜到对应分镜图，导演构图直接扒👇

① ShotDeck（https://t.co/oVuJRsGbFn）— 最专业的电影截图库，数百万张高清镜头，50+维度手动标签，支持HEX颜色搜索+相似镜头推荐。付费$13/月起，物超所值
🔗 https://t.co/CZeV47MCw8

② https://t.co/iTnocjKkAJ（https://t.co/P8XOQNHT6e）— 完全免费，2.4万+张电影截图，10万+tags，甚至带RGB颜色直方图。入门首选
🔗 https://t.co/1VmZjttffx

③ https://t.co/TKnzXWSWeX（https://t.co/72Khf6jlDv）— AI驱动语义搜索，150万+静帧+40万视频片段。直接打中文/英文描述就能搜"人在洗手台前看镜子"，不用纠结标签词
🔗 https://t.co/nkOe9hF4sI

④ Film Vibes（https://t.co/gxmqIwDau9）— 免费为主，电影/广告/MV参考都有。支持按景别、机位角度、时间段过滤，做分镜参考特别好用
🔗 https://t.co/FE8G6s0qWz

⑤ https://t.co/do6Pxf8BZH — 文字搜索+相似镜头推荐+按电影/职员筛选，界面很现代。想找"类似这个镜头的其他变体"时最顺手
🔗 https://t.co/uwjo7lES9g

举个真实使用场景：
你要做一个"洗手台前照镜子"的镜头 → 去 https://t.co/iTnocjKkAJ 搜 mirror（191个结果），再去 https://t.co/TKnzXWSWeX 打自然语言 person at sink looking in mirror，参考素材直接拉满 ，直接参考人家专业电影人成熟的构图，再也不用去空想了
#AI视频 #分镜 #电影感 #提示词 #AI创作工具

[波妞PONYO @ponyodong]

说个我做了大半年AI 漫剧踩过的坑：
AI短剧拍多人戏，最容易崩的根本不是画质，是一换机位人物就全乱了。

你想啊，上一秒三个人好好坐在茶桌前，左边右边都对得上，镜头一切过肩，好家伙，位置、视线、桌子、背景，模型跟失忆了一样，全部重新脑补一遍。

试了好多次才想明白，问题不是提示词写得不够华丽，是根本没给AI一个“稳定的空间”去理解。

这次用的是LibTV的3D导演台，流程大概是这样：先画一张多人站位图，角色、桌子、背景的相对位置先定死；把这张图丢进3D导演台，变成一个能转能换机位的空间蓝图；在这个空间里随便切全景、近景、过肩、俯拍、低角度，怎么切都不乱；最后把这些机位截图转成真实古风剧照，再从剧照生成视频。

说白了核心就一句话：别让模型每次都从一句话里“凭空想象”场景，先把空间关系锁死，剩下的画面质感、表演、运镜交给模型就行。

这样人物不会飘，桌子不会跑偏，镜头切换是真的像有导演在调度，而不是一张张抽卡拼出来的。

我感觉这才是AI短剧从“好看的一张图”往“能讲故事”迈出的关键一步。

7月8号周三晚上8点，我会开一场直播，把整套流程从头到尾拆给你们看。想进群的加微信ponyoassist。

[TechHalla @techhalla]

I'm sure you've seen tons of AI anime FIFA World Cup clips blowing up lately…

but how are they actually made? here's the exact Magnific workflow 👇

[ibexdream @ibexdream]

Vol. 6 is here 🎭
A Midjourney Sref for 3D characters that don't fall apart when you animate them in Seedance 2.0.

Made in Midjourney v8.2
--sref 2607057524 --hd --preview

Examples + prompts below 👇

[Invideo @invideoOfficial]

Get any camera move you want, using just your phone.

Frame the shot, hit record, and move your camera exactly how you want it - a simple pan, a push-in and swirl, whatever.

Load that clip into Agent One and tell it what you want in the shot. Your recorded video drives the camera move, Agent One runs it through Seedance, and you sit back and watch.

Pro tip - match your phone frame to your first generated frame. The closer it lines up, the better it lands.

Watch the video to see exactly how we did it.

[Theoretically Media @TheoMediaAI]

Crazy impressive results using Claude Fable to build out a pose/depth previs for Seedance.

This was inspired by Josh Daws' build-- but of course I made a few tweaks I'll showcase on the channel later this week.
Another example below that shows why I think this is awesome:

[Kōda @aimikoda]

Hina - She looks familiar, doesn't she?

Created on @MartiniArt_

Midjourney + Gpt Image 2 + Seedance 2.0 + Suno
Workflow and prompts in the replies.

[Latte @0xbisc]

Experimented with a new stylized sword trail effect

Key prompt:
white brush strokes, color offset, and hand drawn print edges

made with seedance 2 on @openart_ai

[Zayron @ZayronAI]

According to unreliable sources, Blender + Seedance = 0 gacha pulls.

[Macaroni @CRxMacaroni]

Stage 1 – The Obsessed Nun

Hope she can finally let go...

If you have no idea where to begin, you can grab the prompts in the replies 👇🏻

[Larus Canus @MrLarus]

🤯用ChatGPT-Image2做超宽故事看板生视频，流畅度拉满了！

这次换成一镜到底的《地铁站逃离》：
入口追赶 → 跃过闸机 → 冲下扶梯 → 站台阻拦 → 闪避脱身 → 冲入车厢
可以看到成片中，连续空间里的动作推进一气呵成！

🌟要点是：
1、空间越连续，超宽故事看板越有优势
2、动作路线和人物走位先理清，后面转视频会顺很多
3、生成视频时，要明确去掉所有文字、箭头、圈线和时间线，保留干净电影镜头

工作流 + 提示词在评论区👇

[TechHalla @techhalla]

Midjourney v8.2 + Seedance 2.0

Prompts below 👇

[Josh Daws @JoshDaws]

Okay. This is incredible.