Videos

Our Miss Doubao

This is to stop people stealing the video haha🤣 I'm a nobody, but it's still the fruit of my hard work. Let's see who still dares to laugh at our Miss Doubao.

A textile worker seeing the spinning jenny for the first time

The helplessness of a textile worker seeing the spinning jenny for the first time.

New piece. I think it turned out pretty well

New piece. I think it turned out pretty well. I screen-recorded it like this so you don't get fooled, hehe.

Compositor: a 7 MB Photoshop alternative built with AI by an ex-Apple designer

A former Apple designer built a Photoshop alternative with AI, called Compositor. 7 MB, free and open source (MIT). Layers, masks, color adjustments, content-aware fill: it's all there. Photoshop takes gigabytes; this needs 7 MB, and AI wrote most of the code. https://github.com/robbietilton/Compositor

Perfect. Worth the five dollars I topped up

Perfect. Worth the five dollars I topped up.

A 30-second close-up selfie video from one Muse prompt

Wow, Muse is incredible. I gave it one prompt and a 30-second close-up selfie video came straight out. Who wouldn't love that?

Hand-gesture dance, Muse version

Hand-gesture dance, Muse version. Tools: Muse Opus Prompt: in the comments

Babe, I miss you so much

Babe, I miss you so much.

An outfit-change video made with Doubao

Made an outfit-change video of a beautiful girl with Doubao. The result is pretty good.

Beautiful (✧∀✧)

Beautiful (✧∀✧).

Unbeatable, fam

Unbeatable, fam. Opus 5.5 + Muse (free).

Found something really important

Found something really important. If you just ask Muse to recreate something directly, the results are really bad. Basically, the prompts it writes for itself are trash. So I switched to Opus 5.5. Check out the result. Prompt’s in the comments.

Nothing else for it

Nothing else for it. Back to what I'm good at: AI videos. You all like watching these more. Prompt in the replies.

My favorite Muse version today

My favorite Muse version today.

Video · Bring a close-up selfie to life

Use the young woman in the reference image as the only identity reference. Keep her face shape, proportions, eyes, nose and lips, her long messy dark hair, natural makeup, the close-up selfie framing and her original pose. Generate a real phone front-camera selfie video: she keeps the lazy selfie pose close to the lens, blinks gently once, her breathing makes her neck and shoulders rise very slightly, her eyes look softly into the lens, lips relaxed. The face position and angle stay mostly steady, no sudden head turns. Her long dark hair keeps a real strand texture; the messy strands on her forehead, the wisps beside her face and the ends of her hair sway gently in a faint indoor breeze, with natural overlaps and no wig-like neat outline. The camera simulates a handheld phone selfie with only tiny side-to-side drift and a slight breathing sway. Don't change her identity, no sudden zoom, no scene changes. Keep the straight-out-of-camera photo look: everyday selfie texture, fresh soft color, soft natural indoor light, low contrast, natural skin tone, with real skin texture, slight noise, soft focus and the compression of a phone front-camera recording. The whole thing is quiet, everyday, intimate and natural, like a selfie video casually recorded with the stock camera. No dialogue, no music, only a very faint room tone. Avoid exaggerated expressions, distorted features, plastic skin, a wig look, a cartoon look, an oil-painting look, a 3D-render look, commercial retouching, extra people, extra limbs, harsh lighting changes and scene changes.

Have Muse build you a lossless video downloader

Don't waste your Muse. For videos and images you can't download from Xiaohongshu, Douyin, Bilibili, TikTok, Instagram and the like, just have it build you a tool for lossless downloads. ╮(¯▽¯)╭❤️

Just tell me you love it

Just tell me you love it.

Hello, Everyone!

Hello, everyone!

One more made with Doubao

One more made with Doubao.

Am I pretty?

Sharing a Few Videos from My Personal Collection

Sharing a few videos from my personal collection.

Prompt: Single-Take Gaming Room Selfie Video (9.2s)

Use the first uploaded image as the exact first frame and the second uploaded image as the character reference. Adult woman. Keep her face unchanged throughout. Keep exactly: long straight dark brown-black center-parted hair with long face-framing layers, no bangs; lifted eyeliner, grey-brown contacts, vivid glossy red lips; very long nude-white coffin nails with tiny rhinestones; loose white short-sleeve top with navy double-stripe trim. Scene: gaming room, grey mesh office chair behind her, monitor with a pink and blue abstract wallpaper and a blue-lit PC case at frame left, wooden shelves with figurines at frame right. Light: warm indoor light about 4000K with soft bright front light on her face. Light stays constant. One continuous take, 9.2 seconds, horizontal 16:9, no cuts. Front phone camera held horizontally at arm's length, close-up head to shoulders, slightly above eye level, slight handheld drift; she leans closer to the lens in the second half. Mood: cool, flirty-playful, confident; she mouths along loosely to a song the whole time. Timeline (left/right = frame direction; frame numbers at 30fps): 0.00-0.60s (frames 1-18): the hand at frame right beside her face, long nails combing a strand of hair behind her ear, eyes glancing toward frame right. 0.60-1.00s (frames 19-30): the hand drops out of frame, she lifts her chin and looks into the lens. 1.00-1.40s (frames 31-42): the tip of her tongue briefly peeks out at the corner of her lips, head tilted, eyes half-lidded. 1.40-1.70s (frames 43-51): a playful wink with her frame-left eye only. 1.70-2.60s (frames 52-78): she leans a little closer, lips in a small pout, mouthing words, head tilting toward frame right. 2.60-3.20s (frames 79-96): head tilts again, another quick wink with her frame-left eye around 3.00s. 3.20-3.70s (frames 97-111): steady look into the lens, lips slightly parted. 3.70-4.20s (frames 112-126): the hand rises from the bottom and her index finger rests on her chin just below the lower lip. 4.20-4.60s (frames 127-138): a slow blink, head tilting toward frame left, finger still on her chin. 4.60-5.20s (frames 139-156): finger on her chin, eyes on the lens. 5.20-5.80s (frames 157-174): eyes close briefly, fingertip resting lightly at her lower lip, lips parted, then her eyes open again. 5.80-6.40s (frames 175-192): the finger slides down, the hand curls under her chin and then drops away. 6.40-8.00s (frames 193-240): she leans in closer so her face fills more of the frame, small head sways, mouthing along, calm intense gaze. 8.00-8.40s (frames 241-252): the hand at frame right sweeps past close to the lens, briefly blurring the right edge of the frame. 8.40-9.20s (frames 253-276): the tongue tip peeks out at the lip corner once more around 8.50s, then a soft pout, holding eye contact to the end. Hair moves softly with each head tilt. Five correct fingers, long nails stay the same length and shape. Winks close one eye only. Audio: quiet indoor room tone, faint PC fan hum, soft breathing, no audible singing, no music. Style: realistic phone footage, natural skin texture, no beauty filter, no heavy smoothing, no cinematic color grade. No captions, no watermark, no readable text, no logos, no phone status bar, no app interface.

One screenshot to Muse, a video in 10 seconds

Tried Muse's video generation with a single screenshot and got this in 10 seconds. The quality keeps getting better.

Muse is unreal 😭

Muse is unreal 😭 I gave it three character reference photos and one prompt, and it generated this video with a 2003 MiniDV look. An alley after rain, a clear umbrella, an iced soda on stone steps: 20 seconds, the same face on the lead the whole way through, complete with handheld shake and tape noise. Sometimes the ceiling of AI really just depends on your prompt. #AIVideo #MiniDV

Muse video · An 80-second xianxia Heavenly Palace series

8 prompts, an 80-second Heavenly Palace series. I used Muse to recreate a Chinese xianxia Heavenly Palace world. The full prompts are in the replies.

A Hong Kong street selfie video made with Muse

Made a Hong Kong street selfie video with Muse. I only gave it a prompt; the lead, the street and the camera work were all its own doing.

Seaside Film Portrait · Hazy Intimacy

Reference the facial features of the person in Image 1 to create a seaside half-body portrait. Thick, wind-tousled hair flowing freely, captured with intentional defocus blur, film camera grain, and panning motion for a lens language brimming with emotional tension. Color palette: pastel, desaturated, dark-toned, hazy; lighting is low-key with high contrast, motion blur, dramatic angles, and asymmetric composition. A snapshot of a delicate, sultry mood by the sea with flying seagulls. Cultivates a languid, dazed, out-of-control atmosphere, highlighting stunning visuals and the messy beauty of hair textures to heighten presence and immersive vibe. Infused with forced perspective, dark haze, overexposure, bloom, color fading, and light leaks, accompanied by dense film grain and noise. Richly layered, impressionistic, and visually arresting, embodying the aesthetics of blur and light. Deep blacks pushed darker, highlights boosted, finished with HDR and Lomo effects. Overall style: conceptual art, surreal, effortlessly sophisticated, masterpiece quality.

Wake Up, Unnie

Wake up, Unnie.

Well, found another bit of trivia

Well, found another bit of trivia.

Prompt: Realistic College Campus Selfie with Two Friends

Please generate one realistic photo using the two uploaded person reference images. Character Setup & Face Identity Consistency Image 1 is the face and appearance reference for Woman A. Image 2 is the face and appearance reference for Woman B. For Woman A, keep the exact face identity from Image 1. Maintain her face shape, eyes, eyebrows, nose, lips, facial proportions, skin tone, and natural skin texture. Keep her long black hair worn naturally down. For Woman B, keep the exact face identity from Image 2. Maintain her face shape, eyes, nose, lips, facial proportions, and distinctive appearance. Keep her dark hair in a natural low bun hairstyle. Do not blend or swap their faces. Do not alter or overly beautify the faces. Each person must clearly look like the exact same person from their reference image. Composition & Poses Create a natural smartphone selfie featuring both women together. Woman A is in the foreground on the left right in front of the camera, taking up about 40% of the frame, holding her arm out as if taking a selfie. She looks directly into the lens with a relaxed, slightly playful expression and a natural smile. Her face and upper body should be sharply in focus. Woman B is in the background on the right a few meters away from Woman A, framed from above the knees up. She walks naturally toward the camera, lightly raising one arm in a playful gesture with a bright, natural smile. Both people must be clearly visible in the frame. Maintain natural perspective and realistic body proportions. Woman A is sharply focused in the foreground, and Woman B, while in the background, should be sharp enough for her face to be easily recognized. Wardrobe Woman A wears the exact same navy blue dress shown in Image 1, complete with the white ruffle neckline and white ribbon detail. Woman B wears the exact same navy floral spaghetti-strap slip dress and white cardigan shown in Image 2. Do not change the design, color, pattern, or details of their clothing. Background & Location — South Korea Set the location on a beautiful university campus in Seoul, South Korea. Sophisticated Korean university architecture made of beige stone and concrete, large windows, long outdoor walkways, green ivy growing along parts of walls, wooden benches, and surrounding campus trees. In the distant background, campus guide signs written in Korean are naturally placed. Depict the atmosphere of a modern, authentic university campus in Seoul—not a European castle or fantasy school. Lighting & Atmosphere Late afternoon warm golden hour natural light, around 4500K color temperature. Sunlight coming from the side of the walkway naturally touches their faces and hair, creating soft shadows and warm highlights. A comfortable, lovely Korean college student daily vibe, capturing two friends taking a spontaneous selfie after class on a pleasant afternoon. Photography Style & Texture An ultra-realistic real-life photo looking like it was shot on a smartphone. Natural 24mm wide-angle front-camera selfie perspective, hand-held composition, natural depth of field, realistic skin texture, subtle image noise, delicate hair and fabric textures. Avoid overly polished commercial photoshoot looks; capture a natural, slightly imperfect snapshot feel as if taken casually by a friend. The final result should look like two women actually taking a selfie together on a beautiful Korean university campus. Mandatory Requirements Generate only a single image. Horizontal 16:9 ratio. No collages, grids, or split screens. No extra people. No face blending or face identity changes. No excessive beauty filters or skin blurring. No fantasy architecture or European castle backgrounds. No subtitles, text overlays, or watermarks.

Recap

Recap.

How Do You Say Beautiful?

How do you say beautiful?

Muse video · Breakfast from the boyfriend's view

Muse prompt: Keep the same female lead. New sweet everyday scene: a kitchen in the morning, sunlight coming through the window. She's in loungewear making breakfast, turns back to the camera (boyfriend's point of view) and smiles, holding a piece of fresh toast up to the lens. POV, the intimate feel of a handheld iPhone snapshot, vertical 9:16

Overcast day, high-speed train, window seat, music in my earphones

Overcast day, high-speed train, window seat, music in my earphones Thinking about nothing, just watching the view outside rush backward Prompt coming in the comments later.

Hello, Are You Awake?

Hello, are you awake?

Morning, babe, just woke up. You?

You get my stubbornness, I get your heart (˘³˘)💕

You get my stubbornness, I get your heart (˘³˘)💕.

Muse videos love a head wobble

One thing about videos Muse makes: they love to bob and sway their heads.

Don't Ask Why I've Become So Cold

Don't ask why I've become so cold. I was warm once too.

Unnie, I Love You!

Unnie, I love you! We'll always love you~

A Muse video with voiceover done in one pass

A Muse video with voiceover done in one pass.

Thank You for Your Hard Work Today

Thank you for your hard work today. Let's do our best with a smile tomorrow too. See you tomorrow! ❤️❤️❤️

Muse video generation

Muse video generation.

Posted today

Posted today. I need to sort out the ones from the past few days too.

ChatGPT, I Don't Want to Work, How Do I Become a Millionaire?

ChatGPT, I really don't feel like working—how can I become a millionaire?

Version two, now with its own music

Version two, now with its own music. I'll tidy up the prompt in a bit. The music was extracted from another video and then merged in.

Video · Put on a straw hat, instant outfit change

Generate an outfit-change video. Before the change, this girl is wearing pajamas with her hair down, and she's not in this pose but sitting, not polished at all, very ordinary, only wearing glasses. The instant she puts the straw hat on, she suddenly turns into the look in the reference image, very cool. Then, during the change, the first part plays at normal speed, and when she puts the straw hat on it slows down a little, like a blockbuster movie: very cool, very slick, stunning.

Still taming Muse

This Smoky Blue Is So Beautiful

This smoky blue is so beautiful.

Pretty good result, just one small bug: the rabbit is floating in midair

Pretty good result, just one small bug: the rabbit is floating in midair.

Laughed for ages before I could post this

Laughed for ages before I could post this. Guessing game, no prize: which model made this?

Oppa, can you understand how I feel?

Oppa, can you understand how I feel? 💔

Very Interesting

This is really interesting.

The Results Are Pretty Good

The results are pretty good.

So Awkward Being Seen by Someone, Haha

It's so awkward being seen by someone, haha.

Prompt in Comments, Couldn't Fit It Here😂

The prompt text is in the comments, it wouldn't fit here.😂

I Can't Put This Feeling into Words

I can't put into words how I feel in this moment.

IF YOU

Don't Ask Why I've Become So Cold

Don't ask why I've become so cold. I was warm once too.

Off work, another day gone

Off work, another day gone.

Feels like the results keep getting better. Or is it just me?

Feels like the results keep getting better. Or is it just me? |･ω･｀)

Switched to the Seedance 2.0 Fast Model

Switched to the Seedance 2.0 Fast model, and the results are pretty good.

You ask what I’m up to?

You ask what I’m up to? I’m playing Snake •ᴗ•💧

Genuinely blown away by Claude Opus 5.5

Genuinely blown away by Claude Opus 5.5. It built me a 3D photo album of Ju Jingyi from scratch: 578 photos arranged into a flower island and a photo sphere you can drag to spin, with light dust and mirror reflections. Click a photo and it flies right up to you. It also went online to complete her filmography, checked 163 sources one by one, and finally opened a background browser and recorded this video itself. All plain HTML + native WebGL, no frameworks. At first I tried Astra, used up two reset cards, and got a pile of bugs.

Is this what first love feels like?

Wow, that's wild. Tens of thousands on any given day

Wow, that's wild. Tens of thousands on any given day.

A bug: typing flickers in long replies on X

A bug: typing flickers in long replies on X.

Remaking a video with Muse: second-by-second, face locked

Tried the life-remake method with Muse. All I gave it: the source video + one first frame I made in ChatGPT. It broke down the action second by second, generated two segments, locked the face throughout, and stitched it with the original audio. 13.5s. Honestly didn't expect the face to hold this well.

In this world, the qin’s melody drifts on

In this world, the qin’s melody drifts on.

School uniforms are the GOAT

qiu qiu

Details you missed in 6 famous paintings

6 famous paintings you’ve seen a hundred times, each hiding a detail you never noticed. The Mona Lisa has no eyebrows. Venus’s neck is stretched. Behind the Girl with a Pearl Earring there used to be a green curtain. Mount Fuji hides under the wave in The Great Wave off Kanagawa. Van Gogh couldn’t actually see the village below The Starry Night from his window. A corner of The Scream has a pencil note: “Could only have been painted by a madman!” It was written by Munch himself. What art trivia do you know?

Could this love wait to become a memory?

Found the formula to reverse time

Can You Believe This 3D Biology Courseware Ran on DeepSeek?

Can you believe this 3D biology courseware was generated with dirt-cheap DeepSeek? The spatial structures of 13 organelles are crystal clear, with a category search on the left. The moment you drag the bottom slider to peel back the cell membrane genuinely amazed me. With models competing this fiercely, you have to respect it.

The New Tesla Roadster Actually Flies

The new Tesla Roadster is here, and it actually flies. 0-60 mph in under 0.59 seconds. Faster than a blink. The secret: cold-gas thrusters pushing 4-5x more force than the tires can handle. It can even hang upside down without moving. Promised for 2020 back in 2017. Now it's 2028. Would you put down a $50K deposit?

My heart has been waiting right here for you

You know my heart has been waiting right here for you all along.

If you're tired, take a break

If you're tired, take a break.

One prompt, drawn by AI over 11 years

One prompt, drawn by AI over 11 years. 2015 DeepDream: dog eyes everywhere 2018 StyleGAN: fingers fused into a blob 2021 VQGAN+CLIP: 7 fingers 2022 Midjourney v3: 6, beautiful but wrong 2023 SDXL: still 6 2026: finally counted right… and then the cat raised its paw This 55-second piece didn’t use any image or video model. The visuals, transitions and music were all written in code by Claude, using a method from the open-source huashu-art-motion skill by @AlchainHust. What year did you first make art with AI?

Muse struggles with faces, so faceless dance clips suit it

I noticed one thing - the consistency of the Muse characters is really poor, but if it were just videos without faces where they're just dancing or something, that would be fine.

Made a 3D miniature city you can play in the browser 🏙️

Made a 3D miniature city you can play in the browser 🏙️ Build roads and bridges, zone land and watch the city grow, run bus lines, send out fire trucks, rain and snow… You can drive yourself too: run a taxi, race, get chased by the police 🚓 It's all one web page, rendered live with Three.js. I hereby name Claude Opus 5.5 the No. 1 Baturu. Oh wait, I mean the No. 1 flagship model. https://claude.ai/artifact/3fqN9nwgnAbinDZP65Yvyz

Turning Pretty Photos into Videos Is So Great

Turning pretty photos into videos is such a great feeling.

Claude Optimised This UI

Claude has optimised this UI, and it looks pretty good

Claude Optimised This UI

Claude has optimised this UI, and it looks pretty good

Claude Duet Promo Made with Claude Opus 5.5

Made a promo video for Claude Duet using Claude Opus 5.5. Source material: ⬇️ https://claude.com/product/claude-duet

Not sure why, but this kind of thing is so satisfying to look at

Not sure why, but this kind of thing is so satisfying to look at. Neat and uniform.

Guess for a prize: what was this made with?

Guess for a prize: what was this made with? Guess right and you win 9.9, if that’s not too small for you.

The numbers have been good lately (*＾-＾)

The numbers have been good lately (*＾-＾). (o^з^o)💞

A Three-Wheeler Can Fly Too

A three-wheeler can fly too. I gave Muse a single prompt, and it made this whole 10-second clip: a tiny electric trike charging up a near-vertical neon ramp at night, launching into the dark, then landing through pink smoke with its headlight blazing. No editing, no color grading. Sometimes the tool is so good that all you need is an idea.

Grandpa's closing up shop, hahaha🤣

Grandpa's closing up shop, hahaha🤣.

Swaying gracefully, a cold fragrance drifts into verse

Swaying gracefully, a cold fragrance drifts into verse.

How diffusion models pull an image out of noise

Every AI image starts as a screen of pure noise. It isn’t painted stroke by stroke. It’s scrubbed out of the noise. Here’s how diffusion models actually work. Training. Take a real image and add a little noise. Then a little more. After 1000 steps it’s pure static, nothing of the original left. Now show the model a noisy image and ask one question: what noise was added? Wrong guess, nudge the weights. Repeat across billions of images. LAION-5B alone holds 5.85 billion image-text pairs. Generation runs it backwards. Start from random noise. The model predicts the noise, removes a little, then predicts again. Usually 20 to 50 steps. The first steps only give you blurry shapes and color blobs. The details show up at the very end. That’s why it’s called diffusion: ink spreading in water, played in reverse. Where does your prompt come in? A text encoder like CLIP turns it into numbers, and every single denoising step reads them to decide which way to scrub. Same noise, different prompt, different picture. There’s also a knob called guidance scale, often set to 7.5. Too low and it drifts off-prompt. Too high and the colors burn. Stable Diffusion adds one more trick. Instead of working on raw pixels, it compresses a 512×512 image into a 64×64 grid. That’s 48× fewer numbers to crunch. It denoises in that small grid, then a decoder scales it back up. And the weird hands? Hands twist every way and are often small or hidden in training photos. The model learns what pixels look like. It never learned to count fingers. Newer models like SD3 and FLUX switched to flow matching, a straighter path from noise to image. But the starting point is still the same: noise. What’s the weirdest AI image you’ve seen?

Yuanyuan, South Korea's most stunning face

Yuanyuan, South Korea's most stunning face.

How do I forget you?

Yuanyuan is still gorgeous. Who could resist sneaking a look, haha

Yuanyuan is still gorgeous. Who could resist sneaking a look, haha.

The moon needs maintenance too

The moon needs maintenance too.

Doubao video · prompt by Opus, styling shots by ChatGPT

New piece. I think it turned out pretty well, what do you think? I made the video with the 2.5 model built into Doubao, wrote the prompt with Opus 5.5, and generated the outfit and makeup shots with ChatGPT. If enough of you want it, comment and I'll tidy up the prompt.

A spray of pear blossom in spring rain · Classical-style video

A spray of pear blossom in spring rain · Classical-style video.

A miniature city skyline by Opus 5.5, with a car to drive

Made a miniature city skyline with Opus 5.5. It looks amazing, though it's not finished yet. I could drive around in it all day, hahaha. I just don't have a license, so my driving is pretty bad hahaha🤣.

No wonder everyone loves Claude, even with the risk of bans

No wonder everyone loves Claude, even with the risk of bans.

You are an honest person, a sincere person, an upright person

You are an honest person, a sincere person, an upright person. How’s this one?

Class is in! Understanding LLMs in five steps

Class is in! Understanding LLMs in five steps. https://claude.ai/artifact/83a6cae7-c2cf-48ad-b6f7-c50efc6802ee

What Were You Doing on April 17th This Year?

What were you doing on April 17th this year? You probably have to scroll through your photos to remember. Yet you still remember that summer when you learned to swim as a kid. Why does a year of the exact same length sometimes feel like it passes in just a few sentences? Let's use a roll of film to tell the story of how memories hold onto this year. At the end, there are three small tips to make this year feel a little longer in retrospect. Which day from this year do you still remember? Tell us in the comments. #TimeFlies #TimePerception #Psychology #PopularScience #YearEndReview

This launch animation is stunning

This launch animation is stunning. Back home, launch screens are just ads😓

Kimi 3.1 Pro's Mind-Blowing 3D Rocket Model

Kimi 3.1 Pro just blew my mind with this 3D rocket model! The rocket, launch tower, and surrounding environment are incredibly detailed. You can even rotate the view freely and change the lighting to different times of day. Is Kimi seriously outperforming ChatGPT, Astra, Claude, and Fable 5.5 at this point? https://anrehkyitbrog.ok.kimi.link #Kimi #AI #3D #AIGC

Potato Jokes

Potato jokes: 1. Pip the painter: I paint exclusively in gravy. 2. Captain Scuba: Depth check: the dip is shallow. 3. Sheriff Tater: This town ain't big enough for two tots. 4. Belle of the Ball: I dressed up for the side dish. 5. Rookie: Put me in, coach. I can roll. 6. Cozy Bea: Sweater weather? I baked a scarf. 7. Queen Maris: You may address me as Your Starchness. 8. DJ Mash: My mixtape is all mashups. 9. Baron Russet: An exquisite vintage. Dirt, mostly. 10. Daisy: I grew this hat myself. 11. Beanie Baby: My ears are warm. Where are my ears? 12. Cool Hand Spud: Too cool for the air fryer. 13. Chef Crisp: The special today is… oh no. 14. The Ketchup Kid: Wanted for grand theft ketchup. 15. Puddle Boots: Forecast says gravy. I came prepared. 16. Professor Whiskers: Yes, the moustache is real. 17. Confetti Kit: Surprise! I brought me. 18. Bookworm: I read the recipe. We should run. 19. Skipper: All aboard the gravy boat! 20. Captain Butterbeard: I buried the butter. Lost the map.

Turns Out Some People Really Don't Know Git

Turns out some people really don't know Git. Yesterday an intern on our project asked me how to roll back code after breaking it. I said take a look at git log, and he asked me what Git even was. Man, bro, you don't even know Git? How did you graduate with a top-tier CS degree? All code written by AI, all commits clicked by AI. I'm 100% in favor of using AI, but if you don't even know the basics and can't rescue yourself when things break, you might as well pack your bags. Made a 4-minute animation explaining clone, commit, push, branch, conflict, and rollback all in one go👇

A Muse workflow: smooth motion, weak character consistency

The workflow built with Muse has one pain point: poor character consistency. The movements are still very smooth.

A Tractor Can Fly Too

A tractor can fly too.

Muse videos you could mass-produce

This can definitely be mass-produced. Muse is still so amazing.

Muse videos you could mass-produce

This can definitely be mass-produced. Muse is still so amazing.

Muse videos you could mass-produce

This can definitely be mass-produced. Muse is still so amazing.

Muse recreates the original, but consistency still wobbles

It captures the original look really well, though character consistency is still a bit unstable; I'll take some time to explore Muse further.

My character card

My character card.

Recreating the “Heavenly Palace” series with Muse

Recreating the “Heavenly Palace” series with Muse.

Now this is a real big shot: Musk quoted them

Now this is a real big shot: Musk quoted them.

AnythingLLM: turn a 1,000-page PDF into a private AI assistant

Found an open-source project today that can turn your 1,000-page PDF into a private AI assistant. It's called AnythingLLM. It already has tens of thousands of stars on GitHub and is one of the more popular open-source AI knowledge-base projects of recent years. It's built by Mintplex Labs, with the goal of letting regular people set up their own AI assistant quickly. With AnythingLLM you can upload PDFs, Word files, Markdown and all kinds of documents, then ask the AI questions directly. Have it summarize a paper, pull the key points from a report, analyze material, or even write new content based on your knowledge base. Using RAG (retrieval-augmented generation), the AI doesn't just answer general questions; it works from your private material. Think of it as your own ChatGPT plus a private knowledge base. For students, it can organize course material, papers and study notes. For designers, it can hold your case library, design guidelines and inspiration database. Developers and companies can use it to manage technical docs and internal knowledge. Its biggest strength is local deployment: you can keep your data on your own device instead of uploading everything to a third-party service. It also supports many large language models, both local and cloud. GitHub: https://github.com/Mintplex-Labs/anything-llm Website: https://anythingllm.com #AI #OpenSource #GitHub #LLM #ArtificialIntelligence

Remaking the Heavenly Palace series with Muse

Remaking the Heavenly Palace series with Muse: A terrace over a sea of clouds, a long stairway to heaven, a cloud waterfall, knocking at the gate of heaven… All 8 shots were broken down and rebuilt from the original. Add the traditional Chinese score and it feels just right. I still underestimated Muse.

@Realbaofuliu

Is there any difference between these cards?

“In China, studying hard is more than a habit”

"In China, studying hard is more than just a habit—it’s a deeply ingrained cultural value passed down for generations. Students and professionals alike dedicate countless hours to mastering their craft, driven by a relentless pursuit of excellence and self-improvement."

Petals fall in silence,

Petals fall in silence, so did we — no goodbye, just gone.

Claude’s history, made in code by Opus 5.5

I gave Opus 5.5 an assignment: make a video in code telling the story of how Claude grew up. This is what it handed in 👇

A landscape painting Opus 5.5 draws live in code

You are Opus 5.5, currently the strongest model in the world. Use your imagination and every ability you have to make one thing that will make everyone who sees it, whether an AI expert, a humanities student who knows nothing about tech, or a child, say: “Opus, you're amazing.” What would you make? There isn't a single image in the whole painting. Distant mountains, pines, a pavilion, a lone boat, reflections, rain and snow are all drawn by code the moment you open it. And the mountains are different every time.

Stop fighting, I'll just subscribe to all of you, OK?

There's a follow limit, but I still want to follow everyone at once

There's a follow limit, but I still want to follow everyone at once.

Only that person could make this

Only that person could make this.

Opus 5.5 is incredible. I'm genuinely stunned

Opus 5.5 is incredible. I'm genuinely stunned. It wrote the code for the whole film itself: thousands of particles forming galaxies, DNA, a heart and the moon, with a soundtrack it synthesized live in code. Then it recorded the video in 4K at 60 fps on its own. I said exactly one sentence the whole time 🤯 https://claude.ai/artifact/DS2aQX9bqeLyurBiDShVKA

I couldn't wait, so I added Snake too

I figured you wouldn't be able to wait, so I also added Snake😉 Hahaha

A Snake game made by Opus

This Snake game is so much fun hahaha

This one from Claude beats ChatGPT Astra

I'm done talking, seriously. Doesn't this totally crush ChatGPT Astra? And Astra's quota runs out so fast anyway. https://claude.ai/artifact/L6K2G8sVcuzUX3Qtdbwx9f