55 Fails

[Matt Shumer @mattshumer_]

A completely autonomously-built video: https://t.co/RMfPsW7giT

[ClaudeDevs @ClaudeDevs]

Which Claude model should you build with?

Our new guide breaks down when to use Fable 5.2, Opus 5.5, Sonnet or Haiku. It also covers when to adjust effort instead of switching models, and how to save cost by putting subagents on smaller models.

https://t.co/1VmXqhkZB0

[Kensei @0xKense1]

Fable 5.2 turned one raw tripo model into a RIGGED game character in 36 minutes, from one prompt in blender. so i compared it to my opus 5.5 hero

each one started from a single tripo generation, about $0.60

- fable 5.2: cleaned it, 25 bones, sword and shield, 2 clips, 36 min, about $61 at api prices
- opus 5.5: cleaned it, 25 bones, sword and shield, 8 clips, about 21 min, about $28 at api prices

tip: make it check its own work with numbers. grip within 1 cm of the palm, exactly 2 eyes in the texture, cape stretch under 3x

full workflow is in the article below

[たくぞー｜AI動画ツール実測レビュー @tkz_aiart]

15 AI Camera Moves + PROMPTS、全部試してみた

画像1枚から、promptだけで15種のカメラワークを生成。
同じ人物・同じ場所で、動きだけ変えて

プロンプト
リプ欄↓

[55 Fails @55Fails31748]

@deadsun_0 @kim__aurelie Is it good for music clip video?

[Israa Ali @Israa_Ali2077]

MiniMax-H3 singing performance is convincing. I love this song, and I thought it would be difficult to perform. What impressed me most was the lip-sync, facial expression, and how connected the performance felt to the vocal.

Watch the sustained notes. 🎬🎤

Prompt👉【PROJECT】 MiniMax-H3, 15 seconds, 16:9, one photoreal live-action cinematic musical-theater pop-ballad test. Use @Audio1 as the master timing source: female resonant vocal, soft opening, orchestral rise, brief dip, explosive final climax, clean ending. Synchronize cuts, breath, gaze, gestures and type to the actual audio. Do not invent a different song or singer.

【REFERENCES】 @Image1 locks the singer’s recognizable face, long wavy hair, expressive performance, sleeveless black outfit, silver jewelry and proportions. @Audio1 is the uploaded 15-second song; preserve its vocal phrasing, sustained notes, dynamic build and final repeated climax. One female performer throughout.

【CREATIVE】 A solitary singer moves through one connected theatrical world: shadowed backstage corridor, velvet stage, intimate spotlight. Longing becomes confrontation and then release. Photoreal skin and hair, realistic fabric, brass microphone, velvet, light haze, polished studio image, cinematic orchestral-pop atmosphere, no film grain.

【PERFORMANCE / MUSIC】 Show breath preparation, controlled jaw and throat tension, expressive eyes, subtle head lifts, one hand near the sternum, and gradual expansion from vulnerability to chest-led power. Mouth and facial articulation match the uploaded vocal; no generic silent lip motion. No dialogue, rap, crowd vocals, comedy or dance troupe. Keep natural breath gaps. Final intensity arrives only on the last sustained phrase.

【VISUAL / TYPE】 Black velvet, warm ivory spotlight, deep burgundy fabric, muted metallic silver jewelry and microphone. Ultra-black condensed grotesk, ALL CAPS, tight tracking; secondary narrow sans used sparingly. Editorial stage typography, not subtitles or karaoke. Use paper-edge reveal, restrained double-image echo, halftone offset and one brief black-frame word. Words must be real, short, readable before transforming; never gibberish or logos.

【SHOT 1 — 0.0–1.8s】 Medium close-up, slow lateral track through a backstage doorway. The singer enters from shadow, shoulders forward, hand near sternum, preparing breath. Action: threshold entry. Space: backstage corridor. Hair and gown edge follow one beat late. Typography: “LONGING”; Layer A large condensed hero on wall, Layer B smaller silver print at doorway seam, Layer C thin foreground letters partly hidden by her shoulder. Rhythm / Cut: opening vocal reveals the word; paper edge wipes into her face.

【SHOT 2 — 1.8–3.6s】 Close-up tracking beside her as she turns toward the microphone. Action: shoulder-led half-turn. Space: corridor opening to stage. Eyes lift toward the audience; breath shows in chest and throat. Typography: “STAY”; Layer A large word behind head, Layer B narrow letters on glass reflection, Layer C cropped lower-frame letter clear of eyes and mouth. Rhythm / Cut: vocal consonant hard-cuts profile to three-quarter view.

【SHOT 3 — 3.6–5.5s】 Medium close-up, slow push to the vintage microphone. Action: grounded reach and contact. Space: stage threshold. She grips the stand and leans into the emotional phrase; eyes close on the sustained note, then reopen. Typography: “ENOUGH”; Layer A hero on velvet panel, Layer B halftone photocopy offset behind mic, Layer C thin type reflected in brass grille. Rhythm / Cut: hand contact lands on beat with restrained metallic resonance.

【SHOT 4 — 5.5–7.3s】 Medium shot, quarter arc as the curtain opens. Action: curtain release and chest expansion. Space: front stage. She releases the stand, opens her ribcage into the rise; gown follows late and settles. Typography: “HEAR ME”; Layer A split hero across curtain, Layer B offset print registration, Layer C torn-paper foreground briefly masks lower torso. Rhythm / Cut: orchestral lift opens curtain; words remain readable before folding away.

【SHOT 5 — 7.3–9.1s】 Extreme close-up on eyes, cheek and upper lip, nearly locked with breathing drift. Action: still-point eye performance. Space: center spotlight. Music dips; eyes move from floor to camera, uncertainty becoming resolve. Typography: “WAIT”; Layer A heavy word beside eye line, Layer B faint mirrored print behind it, Layer C no obstruction over face. Rhythm / Cut: half-second visual suspension, no shock zoom or lyric subtitles.

【SHOT 6 — 9.1–11.0s】 Medium close-up, camera pulls back as she takes one deliberate step into full light. Action: step-forward power transfer. Space: center stage. Rear foot pushes, shoulders settle, chin rises; microphone stays behind. Typography: “I RISE”; Layer A projected hero backdrop, Layer B narrow print fragments trail one step, Layer C foreground black letter exits frame. Rhythm / Cut: returning pulse triggers step and brief double-image echo, resolving to one sharp performer.

【SHOT 7 — 11.0–13.0s】 Medium to close-up, controlled tilt as strings widen. Action: upward head lift, open-arm release, then pivot freeze. Space: downstage spotlight. She lifts her head and opens one arm from the sternum; hair and gown sweep and settle. She turns three-quarter to frontal and freezes on the high vocal. Typography: “FOR ME”; Layer A large hero on black stage plate, Layer B silver registration offset, Layer C narrow strip only across lower frame. Rhythm / Cut: vocal peak triggers the pivot; hold one breath with mouth clear.

【SHOT 8 — 13.0–15.0s】 Extreme frontal close-up, slow final push into face and microphone grille. Action: held-note concentration and breath release. Space: intimate center-stage afterglow. Eyes glisten; pleading becomes self-possession; jaw and mouth match the final sustained vocal. Typography: “ENOUGH” then “FOR ME”; Layer A same condensed hero returns clearly, Layer B faint silver duplicate under it, Layer C no face obstruction. Rhythm / Cut: final phrase returns the motif; last note ends with a clean orchestral stop and hard cut to black, no fade.

【GLOBAL LOCKS】 Preserve singer face, hair, jewelry, outfit identity, microphone relationship and female vocal performance. Photoreal live action only; realistic skin, hair, fabric, haze and stage light. No face replacement, costume change, extra performers, outdoor scene, anime, illustration, plastic skin, grain, generic subtitles, karaoke display, unreadable type, fake brands, watermark, random glitch, soft dissolves or endless loop. Keep mouth visible and correctly shaped during vocals. End decisively on the final note.

[apek @apekshik]

Now you can build any scene in Blender from a single reference image.

I’ve open-sourced my tool AnyScene: drop in one or more references (plus optional text) and an agent builds the whole set inside Blender.

Production-ready assets in a production ready environment, placed, lit and color-graded, with every asset broken out as its own file.

https://t.co/0rb1MC3hzo

[Mario Joos @MarioJoos]

A Comprehensive Introduction To YouTube Dubbing (Including Results & First Languages)

If you want 3 to 4x the amount of views on your video, one of the easiest ways to do this is through dubbing. But there are a few things you need to look out for.

In this post, I want to give you a brief introduction to dubbing, what it is, how it works, what pricing you should expect and the possible results you’re looking at.

What is dubbing?

First of all, let’s talk about what dubbing is.

Dubbing is the strategy where you add extra language tracks to your video that would alter the video’s audio when you’re from another country or when you prefer to watch something in a different language. Many creators gain up to 300-400% extra views after dubbing into the most important languages, so there’s a lot to gain here. (Check the image I’ve added)

But it’s not as simple as it seems.

The different types of dubbing

First, let’s talk about the different types of dubbing. I always position them under 4 different categories.

You have non-dubbed videos, which are videos without dubbing. So let’s say you speak English, everyone will see your video in this language.

You have auto-dubbed videos, which is where you use YouTube’s free auto-dubbing features to let them use AI to transform your voice into any language. So let’s say that a user only watches Spanish content, they may see your video in Spanish.

You have privatized AI dubbing. This is similar to YouTube’s auto-dubbing. However, you’re using custom models, paying a fee and having to upload individual tracks yourself.

You have human dubbing. This is where you’ll hire a studio to find voice actors to dub your videos and become your voice.

Now let’s talk about each of them, the positives and the negatives.

1. No Dubbing

The biggest advantage of non-dubbed videos is that everyone sees your videos as you intended them to be. The way you make your jokes, the balance of your tracks, the intonation of your voice and so much more. However, the biggest downside is that you’re locking yourself out of the rest of the world.

Now, most of the world speaks English fairly well, so for many creators this isn’t a problem. However, many people don’t speak English, so you’re also missing out on a huge portion of your potential audience.

So when would you even choose to go for no dubbing?

Well, this is mostly when you either want to avoid people having a different experience, or when you’re very personality-focused. Personality-driven channels don’t perform as well when dubbed as educational channels. You may only see a 30-50% increase in viewership in some cases if you’d dub, but then there’s the direct cost that comes with dubbing, and it doesn’t make sense.

So generally, I don’t advise people to dub until they’re a little bigger.

2. Auto-dubbing

Imma be completely honest, just turn that shit off. I don’t know why YouTube chose this company they’re currently working with, but it’s ass at best.

Yes, it may have some positive impact on some channels, but for the most part, the quality doesn’t just hurt the content, it also hurts your image towards a potential future audience member. The only positive thing that I can say about this is that it’s free.

Maybe by the time you see this video, it may have gotten better, but unless you’re lucky, it’s just garbage. Now, here’s the worst part.

YouTube auto-enabled this feature for your channel, so make sure to go to your channel settings and turn it off.

3. AI dubbing

Now, AI dubbing is very interesting because it uses AI models to clone your actual voice, and it’s pretty good at that. The cost is also relatively cheap, coming in at around $1.80-$2 per minute per language.

And I’ve honestly seen a lot of people have success with this. It’s a budget variant that actually does a pretty good job.

Now, most people tend to use ElevenLabs or Dittodub. Each has its own benefits, and I’d just try out which works best for you.

However, there are also a few other ones coming to the market, so just see what’s out there, but I honestly would just avoid paying more than $2 per minute per language, especially as AI models are getting cheaper and cheaper anyways.

The biggest upside is that you can quickly do a lot of languages for a fraction of the cost and reach a wider audience.

The biggest downside is that you’ll still notice tons of mistakes and mismatches with your voice, so it’s not perfect yet.

If you’re really fancy, you could even use your own models.

4. Human dubbing

Lastly, you have manual or human dubbing. This is where you work together with a studio to get an actual human to become you.

The biggest upside is that these voice actors will be better at translating the jokes and tone you’re trying to convey. And this often results in good outcomes. However, that’s also where the biggest issues come into play.

The average price for a language is about $25-$35 per minute. Some languages are higher, some are lower.

Another downside is that when you change studios, the voice also changes.

Another issue is that many of these studios aren’t publicly transparent about their pricing, so you’ll notice them trying to quote you a higher price than they charge others, so I highly encourage you to negotiate.

And lastly, the turnaround times are much slower than AI dubbing.

Currently, human dubbing is still superior to the other types, but the hassle and pricing make it often only worth it if you’re already pulling lots of views or if your revenue stream is different from AdSense.

Which languages should you dub into?

You’ve now gotten yourself an understanding of what’s possible, what fits your budget range and some of the positives and negatives behind each type of dubbing. So, what’s next?

Next are the languages to go for. There’s a big complexity in choosing what languages to go for, but I’m going to give you the typical list and order of which languages typically work well.

Two notes. First, based on the content type, etc., you may find better or worse results by language, and certain types of channels may even be able to consider countries that aren’t on this list. For example, if you play a certain sport that’s also popular in Denmark, but Danish isn’t on this list, you may still want to consider Danish.

Another note is that this list was created years ago, when I did a deep analysis into population size, English proficiency, dubbing culture, internet adoption, market size, potential earnings and so much more. These days, I use a simple tool to help me predict dubbing potential, but the findings are extremely overlapping.

So there are about 10 languages you typically go for. And this counts for Shorts and long-form.

1. Spanish
2. Indonesian
3. Arabic
4. Turkish
5. Portuguese
6. Bangla
7. Thai
8. Vietnamese
9. Hindi
10. Russian

If you want to be experimental and add extras, some people tend to go for:

11. Italian
12. German
13. Tamil

But honestly, I’d start with the first 10 languages. You could even make some of them AI, some of them human dubbing.

Important practical note!

When you add languages like Spanish, make sure you add them as Spanish, not Spanish (Mexico). It’s a common mistake people make.

Or you could do all if you want to play it safe. Or, if you’ve gone as far as picking regional accents.

Final things to keep in mind

Make sure you’re able to add custom tracks before you start paying for anything. And also note that results aren’t always guaranteed for anyone. I don’t want anybody to come after me, making it seem like I guaranteed a certain result.

And don’t forget to also update your title into these languages and even your thumbnails (if you have access to this feature).

Hope this breakdown was helpful.

[Extraltodeus @Extraltodeus]

ClaudiaGI - Qu'est-ce qu'elle a que j'ai pas? (What's she got that I haven't?)
🥖 #FRENCH #SLOPCORE ™ — Original Recipe
Now with REAL chunks of baguette!

Gérard included at no extra cost

[码农暖爸 @Delroy715]

卧槽，给的真多，太划算了

实测了一下 Google Colab 新上的 G4 96G显存云端工作站，这单机硬件给得实在太猛了 🤯

🔹 GPU：NVIDIA RTX PRO 6000 Blackwell (sm_120 旗舰架构)
🔹 显存：96 GB GDDR7 ECC（1.8 TB/s 带宽，单卡比 80G A100/H100 多出 16GB）
🔹 内存：默认直给 188 GB DDR5 高速系统内存（显存+内存总池高达 284 GB！）
🔹 CPU：48 vCPU AMD EPYC（原生支持 AVX-512）
🔹 存储：235 GB Gen5 NVMe (读取 >2.5GB/s) + 30 TB Google Drive 极速直连
🔹 费率：仅 8.71 CCU/小时，甚至比 A100 (9.46 CCU/h) 还便宜

单卡 96G GDDR7 + 188G 内存，无论跑超长视频生成还是超大参数权重常驻，性价比直接拉满 👇

[方知微 @superfang119]

只需要 1 张角色图，就能让 MiniMax H3 帮你制作最多 13 个视角的角色设定图！

ComfyUI 又出现一个非常实用的开源节点：

H3 Character Sheet Designer

它直接利用 MiniMax H3 原生参考图条件能力，把角色设定图制作流程整合进 ComfyUI。

核心功能：

1️⃣ 13 个视角自由选择
头部：6 个视角
全身：3 个视角
服装：正面、背面
细节：双手、双脚及鞋履

2️⃣ 最多 4 张参考图
1 张角色主体图
最多 3 张额外参考图
可以分别指定脸部、头发、服装、手、鞋履等参考用途。

3️⃣ 身体部位独立控制

想换衣服？只指定服装。

想保持脸部一致？单独提供面部参考。

想修改手套、鞋子或背面服装？可以针对对应部位输入指令。

4️⃣ 自定义布局
支持视角选择、尺寸调整、预设布局和实时预览。
5️⃣ 保留原有工作流
提供多个版本的节点，兼容原有节点 ID 和对应工作流。

最重要的是：
它不是一个独立的 AI 绘图软件，而是直接调用 MiniMax H3 的原生参考图生成能力。

对于 AI 影视、动画、游戏角色设计来说，这种工具非常有价值。

以前制作一个角色设定，往往需要反复生成正面、侧面、背面，再人工筛选和修正。

现在可以把这些步骤整合进一个 ComfyUI 工作流。

AI 角色资产生产，正在从单张图片生成走向多视角、可控化、流程化。

GitHub：
https://t.co/zhHDhc17np

#ComfyUI #MiniMaxH3 #AI绘画 #角色设计 #AIGC

[Samuel Zeng @SamuelZengML]

Introducing Audio8-TTS-Realistic, @Edge0_AI’s new open-weight speech model: realistic, zero-shot voice cloning in just 0.6B parameters.

• Human-like prosody and expressive delivery
• Clones a voice from one short clip, in 23 languages
• 601M params: small enough to fine-tune and serve affordably

Currently #1 on the Hugging Face TTS Arena, where people vote blind on which voice sounds more human.

Open weights are available now on @huggingface. Try it free.

[Mustafy | AI Video Creator @MustafyOf]

How to Make AI Music Videos Better than 99% of People

[BennyDaBall @BennyDaBall_OG]

NVIDIA open-sourced Open-H3-Regenerate-2K (was API-only). Running it local on my 5090.

Same 768P MiniMax H3 draft → 2560×1440:
- Draft: 2:09
- NVIDIA default: 3:15 · SSIM 0.75 / 0.61
- My tuned: 2:28 · SSIM 0.83 / 0.68

Default = detail monster, but it recasts the face and the Boston Terrier. Tuned keeps identity. First round of settings, more testing coming.

[0xMarioNawfal @RoundtableSpace]

Vibe-Trading (37k stars) is one of the cleanest personal trading agents right now.

* MCP server with 80+ tools
* Multi-agent swarms
* Backtesting + paper trading
* Crypto (Binance, OKX, CCXT, perps)

pip install vibe-trading-ai

Then connect it to Claude Code.

https://t.co/0rudpZZjZq

[55 Fails @55Fails31748]

@Punk2027 @grok Yes

[Jah Mills @itsjahmills]

holy f*ck…he is right
229M views in 12 days 😭

[Ryker⏸️ @Ryker_Strategy]

People asked how to make videos like this.

So I broke down their videos and built a Claude prompt kit you can adapt to your own niche.

Comment "Claude" and I'll DM you the guide.

Here's the process:

1. Pick the viewer and the feeling

Fear? Curiosity? Comedy?

The same story structure can take very different forms.

2. Start with something familiar, then break one expectation

A phone charges to 101%.
A meeting meant to save time fills your whole calendar.

One is horror. One is office comedy.

3. Write the ending FIRST

Know what the viewer will understand at the end. Then plant the details that make it work.

4. Make every beat reveal something

This channel's rules aren't just random creepy instructions. Each one reveals another part of the same system.

That's what makes the ending land.

5. Keep the visuals simple

One object. One action. One thing the viewer needs to understand.

Claude can help write the story and code simple animated scenes. Record/export them, add narration, then edit the video.

The guide includes 5 prompts: choose your format → find ideas → write the story → build scenes → fix the weak parts.

Plus a worked example outside horror.

Don't be 50 IQ and clone this exact niche.

Use your brain. Pick a different audience and niche

[Eddie Eizner @eddieeizner]

I made $7,389 this month posting viral 2D animation AI videos

Here's how:

[Storm @richoffyoutube]

📁 Faceless YouTube (Documentaries)
│
├── 📁 Niche & Topic Selection
│ ├── 📁 High-Retention Documentary Niches
│ ├── 📁 Evergreen vs Trend-Based Topics
│ ├── 📁 Competitor Channel Breakdown
│ └── 📁 Audience Psychology (Curiosity + Emotion)
│
├── 📁 Research System
│ ├── 📁 Source Gathering Workflow
│ ├── 📁 Fact-Checking & Accuracy Systems
│ ├── 📁 Story Angle Development
│ └── 📁 Research Prompt Libraries
│
├── 📁 Script Frameworks
│ ├── 📁 Hook Structures (Curiosity Gaps)
│ ├── 📁 Narrative Arcs for Documentaries
│ ├── 📁 Tension & Payoff Timing
│ ├── 📁 Cliffhanger & Retention Techniques
│ └── 📁 AI Script Prompts for Docs
│
├── 📁 Visual Storytelling
│ ├── 📁 B-Roll Sourcing & Licensing
│ ├── 📁 AI Image + Video Generation
│ ├── 📁 Stock Footage Workflows
│ ├── 📁 Archival Footage Systems
│ └── 📁 Scene Composition Rules
│
├── 📁 Voiceover & Narration
│ ├── 📁 AI Voice Selection (Tone Matching)
│ ├── 📁 Pacing & Emotion Control
│ ├── 📁 Script-to-Voice Alignment
│ └── 📁 Audio Cleanup & Mixing
│
├── 📁 Editing & Packaging
│ ├── 📁 Documentary Editing Templates
│ ├── 📁 Pacing for High Retention
│ ├── 📁 Text Overlays & Motion Graphics
│ ├── 📁 Music & Sound Design Systems
│ └── 📁 Color Grading Styles
│
├── 📁 Thumbnails & Titles
│ ├── 📁 Documentary CTR Formulas
│ ├── 📁 Curiosity-Driven Thumbnails
│ ├── 📁 Title Hooks That Work
│ └── 📁 A/B Testing Systems
│
├── 📁 Growth Engine
│ ├── 📁 YouTube SEO for Documentaries
│ ├── 📁 Titles & Descriptions That Convert
│ ├── 📁 Upload Cadence & Series Strategy
│ └── 📁 Analytics That Matter
│
├── 📁 Monetization
│ ├── 📁 AdSense Optimization
│ ├── 📁 Digital Products Tied to Niches
│ ├── 📁 Affiliate Offers (Tools, Books, Courses)
│ ├── 📁 Channel Memberships / Community
│ └── 📁 Scaling to Multiple Doc Channels
│
└── 📁 Automation & Systems
├── 📁 Full Faceless Doc Workflow
├── 📁 Content Pipeline SOPs
├── 📁 Prompt Libraries
├── 📁 Outsourcing Rules
└── 📁 Scaling Multiple Channels

🔖 Save this