Beetle (@beetleswt) header
Beetle (@beetleswt) profile picture

Ex-degen, now creative scientist. I collect pixels and generate slop; Web3 and AI feed my soul.

Between Reality And Metaverse
Joined May 2021
Photo by Beetle (@beetleswt)Photo by Beetle (@beetleswt)Photo by Beetle (@beetleswt)Photo by Beetle (@beetleswt)Photo by Beetle (@beetleswt)Photo by Beetle (@beetleswt)Photo by Beetle (@beetleswt)Photo by Beetle (@beetleswt)Photo by Beetle (@beetleswt)Photo by Beetle (@beetleswt)
I gave Kling 4.0 three keyframes and told it to fill the gaps. It gave me one unbroken 30s cat chase through an HDB estate, ending on the exact frame it started. Last try was pure text-to-video and the flats looked like Hong Kong. This time I drew the keyframes first. Kling 4.0 is now on PixVerse! Go unlimited for 10 days with Ultra—generate at zero credit cost using text, images, references, or Transition mode (first and last frames). @PixVerse #PixVerseUnlimited #KlingOnPixVerse
Text to video with Kling 4.0. One prompt, 30 seconds, no cuts. Kopi to rooftops to Marina Bay and back into the same cup. Kling 4.0 is now on PixVerse, and unlimited for 10 days with Ultra. @PixVerse #PixVerseUnlimited #KlingOnPixVerse
2
1
2
155
I broke it down with the keyframe methods. Using 7 keyframes and a 30s run led to a poor result. The cat and mouse was everywhere and didn't work well. What worked was using just 3 keyframes (1, 3, and 5) and then doing 3 X 10s generations.
1
9
Text to video with Kling 4.0. One prompt, 30 seconds, no cuts. Kopi to rooftops to Marina Bay and back into the same cup. Kling 4.0 is now on PixVerse, and unlimited for 10 days with Ultra. @PixVerse #PixVerseUnlimited #KlingOnPixVerse
2
1
243
What I learned from this one 👇 Text to video gives pretty good motion, but you can't control exactly what shows up. I asked for Singapore HDB flats and Kling gave me blocks that look more like Hong Kong, then teleported me back to a hawker centre at the end right from MBS. The motion though? One take, no cuts, clean loop. Next test: I generate the reference frames for every scene first, then let Kling 4.0 fly the drone through them. Same prompt, more control. Let's see if it's better!
3
58
The prompt, if you want to try it 👇 One continuous FPV drone shot, 30s, photoreal, no cuts. Starts on a macro kopi cup at a hawker centre, rockets up between HDB blocks at dusk, threads through a kitchen window past a cat, skims Marina Bay at sunset, then dives back through the hawker roof and lands on the same cup for a perfect loop.
36
My iPhone Duo video went viral, so I'm giving away 100 iPhones on X 😛
94,078
96,258
6,670
143,585
5,551,862
Beetle retweeted
Generated a video with Seedance 2.5, then gave it to Opus 5.5 and asked for an anime opening. It re-cut every shot to the beat, added impact frames, motion graphics, a title logo and a color grade, and composed the music from scratch in code. Original + prompt in the replies.
69
65
8
803
36,893
we want to start testing a Midjourney MCP with a limited community. the intention is to find creative and technical people to push the boundaries of what our models can do. this is not for Midjourney in SaaS products, it's for purely artistic and creative endeavors. apply below!
361
169
176
2,247
216,480
「動きと音をビジュアライズ」のメイキング 簡単ではありますが、参考になれば嬉しいです。 ■素材は少ない ・下絵は4種類のAI動画+曲 ■Blenderの中の仕組み ・元の映像は1枚の板。それを細かい粒に分けている ・動画から測るのは3つ。人物の形、奥行き、どこが速く動いたか ・曲を低い音から高い音まで細かい音域に分けて、それぞれの鳴り具合を1コマごとに測る ・画面の真ん中は低い音、端は高い音の担当。飛び出す時に、鳴っている音の所ほど大きく出る ・速く動いた所(バチ、筆先、スカートの裾)は粒になって浮く。音が大きいとさらに浮く ■拍に合わせる方法 ・動画と曲は別々に作ったので、そのままでは合わない ・太鼓を叩く、かかとを踏む、ボールを打つ、筆を打ちつける。この「決めの瞬間」を割り出す。 ・そこで時間を止め、止める長さを変えて、次の決めの瞬間が必ず拍に来るように計算している ・止まっている間にカメラが回り込み、空間の色と照明が替わる ■任せたこと/決めたこと ・任せた:画面のどこが多く飛び出すか。音と動きの数値のまま ・決めた:決めの瞬間ごとの強さ、見た目、カメラ、色。会話で1つずつ ・Blenderの操作はPython。 ■うまくいかなかったこと ・激しくしすぎる、下絵を全くなくすと、訳がわからなくなった。 ・なかなか思い通りにならず、15回以上修正をかけている。SEは足したら合わなくて外した。暗くする加工も外した、など。 これを作って一番大事だと思ったのはメリハリと、バリエーションで、全体というよりかは、細かく細かく細部をちょっとずつアップデートしていくことにより、理想の映像に近づいていきました。
15
235
21
2,257
88,073
No luck w the 30 anniversary packs, but might get lucky with some Megas?
7
266
Opus5.5's "video generation" capability is scarily good. Not in the sense of mindblowing video graphics, but the fact that this was created in code, with NO image or video generation model amazes me. What most people don't tell you: There are limits to the characters, the animation effects, and the cost of producing it (in credits and time). What really works: Opus' understanding of how scenes move across one another, which helps tremendously for scripting and storyboarding — and then tying in an image/video gen model to complete the loop. That's not a new workflow tbh but I'm thinking of how to use Claude as an even more powerful orchestrating video producer with this model. 💡
2
256
Everyone keeps their head down. Until she doesn’t. ZERO RULES, a music-driven animated short about the rules we absorb, the freedom we imagine, and the small choices we bring back. My submission to HF Film Festival.
3
154
Beetle retweeted
SAM AND DARIO SLOWING THINGS DOWN ✨
179
379
65
4,202
238,497
we’ve delivered 489,297,478 pizzas since spurs last scored a premier league goal
584
7,664
878
70,242
4,489,309
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so. Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training. You can read the full post here: darioamodei.com/post/we-must…
10,601
16,357
11,364
87,711
77,006,798
Using a photo of a real mooncake, I made this video with @ChatGPT Astra6 + @Hailuo_AI H3! Who wants some MFlower mooncakes? Full workflow below 👇🏻
1
4
194
Step 1: Upload photo to ChatGPT and describe your video idea. Step 2: Ask it to build the keyframes, which you'll use as reference images, and the T2V prompt for H3 Step 3: Paste the prompt and images into any H3 generators and voila! Bonus: Give it a sample H3 prompt to adapt from
1
1
59
Mooncakes are real, video is not. Consume at your own risk. Happy mid autumn!🐇
1
20
Been so out of web3 but the unfollowing drama is so entertaining. Only because everyone makes a big deal out of it 🤣
1
107
Should you advance or wait? Body language answers before anyone speaks. Testing @Hailuo_AI H3, with @bot running the automation. Still figuring out the right rhythm and number of keyframes...
133