Sensitive content

This media may contain sensitive content.

Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

Real satisfaction, with multi masking under the Aerith's face. Guess how many layers? Cheer 刺儿 in Devourer by NEOGAN. Inthemask by Moli's photoshooting. . #molis #inthemask #molifx #neogan #crossdress #latexsuit #rubberdoll #unmask #masking #aerithmask

manaminanase's profile picture

Moli 茉莉兔兔

30,637 subscribers

28,884 Aufrufe • vor 2 Jahren •via X (Twitter)

NSFW#molis#inthemask#molifx#neogan#crossdress#latexsuit#rubberdoll#unmask#masking#aerithmask
LIVE

Anya Rossi• Live Now

Private livecam show

1 Kommentare

Profilbild von TSVIPER
TSVIPERvor 2 Jahren

@cier0000 @molisstudio WOOOW PLEASE SEND ME ONE LIKE THAT

Ähnliche Videos

Double Masking in Molly2 x NEOGAN NN50T Translucent Latex Bodysuit via YouTube . #molis #inthemask #neogan #latex #nn50t #molly2 #latexsuit #kinky #mtf #crossdresser #molifx
0:21

Sensitive content

Double Masking in Molly2 x NEOGAN NN50T Translucent Latex Bodysuit via YouTube . #molis #inthemask #neogan #latex #nn50t #molly2 #latexsuit #kinky #mtf #crossdresser #molifx

Moli 茉莉兔兔

45,969 Aufrufe • vor 2 Jahren

Putting on Tifa's face. Tifa by #secondface. Devourer & NB10 by #neogan . #molis #inthemask #nb10 #tifamask #latexcatsuit #molifx
0:35

Sensitive content

Putting on Tifa's face. Tifa by #secondface. Devourer & NB10 by #neogan . #molis #inthemask #nb10 #tifamask #latexcatsuit #molifx

Moli 茉莉兔兔

358,243 Aufrufe • vor 4 Monaten

Fitting with Linglong "Jade Serpent" makeup. Model Cheer 刺儿 . #molis #inthemask #molifx #masking #linglong
0:53

Fitting with Linglong "Jade Serpent" makeup. Model Cheer 刺儿 . #molis #inthemask #molifx #masking #linglong

Moli 茉莉兔兔

34,986 Aufrufe • vor 1 Jahr

Living Doll Unmasking! Nun latex suit by #NEOGAN Mask by #SecondFace "The Nun". Get dollified with us! . #molis #inthemask #livingdoll #rubberdoll #latexsuit #latex #femalemask #unmasking #siliconemask
0:42

Living Doll Unmasking! Nun latex suit by #NEOGAN Mask by #SecondFace "The Nun". Get dollified with us! . #molis #inthemask #livingdoll #rubberdoll #latexsuit #latex #femalemask #unmasking #siliconemask

Moli 茉莉兔兔

33,721 Aufrufe • vor 2 Jahren

Let's put on some makeup on your mask from the start! Linglong "Jade Serpent" makeup process. Model Cheer 刺儿 with Inthemask by MOLIS . #molis #inthemask #molifx #linglong #femalemask #siliconemask #prostheticmakeup
0:54

Let's put on some makeup on your mask from the start! Linglong "Jade Serpent" makeup process. Model Cheer 刺儿 with Inthemask by MOLIS . #molis #inthemask #molifx #linglong #femalemask #siliconemask #prostheticmakeup

Moli 茉莉兔兔

31,423 Aufrufe • vor 1 Jahr

Another unmasking and revelation of my real face too. In #linglong mask by #molifx . #molis #inthemask #unmasking #siliconemask #femalemask #unmask #crossdressing #prostheticmakeupfx
0:31

Sensitive content

Another unmasking and revelation of my real face too. In #linglong mask by #molifx . #molis #inthemask #unmasking #siliconemask #femalemask #unmask #crossdressing #prostheticmakeupfx

Moli 茉莉兔兔

24,063 Aufrufe • vor 1 Jahr

We talked about Linglong, the upcoming mask by MoliFX. Cier and me, both in Molly2. Cheer 刺儿 Inthemask by MOLIS How Realistic Can A Female Mask Be? | Moli Says x Cier In Silicone Masks via YouTube
12:30

We talked about Linglong, the upcoming mask by MoliFX. Cier and me, both in Molly2. Cheer 刺儿 Inthemask by MOLIS How Realistic Can A Female Mask Be? | Moli Says x Cier In Silicone Masks via YouTube

Moli 茉莉兔兔

41,034 Aufrufe • vor 1 Jahr

未成年回避！“姐姐”现在对你们来说太“大”了噢~ 模特是刺儿 Cheer 刺儿 ，她说她未成年，收不到这条推送：） Adult only! "Milf" is a bit too grown for minors. Disguise in Molly2 & The Kim3 & Zero Touch Bodysuit. . #molis #inthemask #molifx #thekim3 #molly2 #femalemask #siliconemask #crossdress
1:00

Sensitive content

未成年回避！“姐姐”现在对你们来说太“大”了噢~ 模特是刺儿 Cheer 刺儿 ，她说她未成年，收不到这条推送：） Adult only! "Milf" is a bit too grown for minors. Disguise in Molly2 & The Kim3 & Zero Touch Bodysuit. . #molis #inthemask #molifx #thekim3 #molly2 #femalemask #siliconemask #crossdress

Moli 茉莉兔兔

21,530 Aufrufe • vor 2 Jahren

The Fetish Party 2024 in Shanghai #MoliMeetup2 #茉莉沙龙2 via YouTube Follow Inthemask by MOLIS for more Moli Meetup2 contents! Special thanks to everyone who is in the party: 蛋蛋喵=w= 八尺の正太 滑稽 wzywyi 肥你个橙 TeresaSS Annie Mia SissyReina 琳 葡萄🍇🍇🍇 Steve Shawn IceQ ice fans Cheer 刺儿 Kimberly Hada Virgil spike5695 沙包酱想被暴打 邻家小墨 asdfghjkl Litao and more!
0:37

Sensitive content

The Fetish Party 2024 in Shanghai #MoliMeetup2 #茉莉沙龙2 via YouTube Follow Inthemask by MOLIS for more Moli Meetup2 contents! Special thanks to everyone who is in the party: 蛋蛋喵=w= 八尺の正太 滑稽 wzywyi 肥你个橙 TeresaSS Annie Mia SissyReina 琳 葡萄🍇🍇🍇 Steve Shawn IceQ ice fans Cheer 刺儿 Kimberly Hada Virgil spike5695 沙包酱想被暴打 邻家小墨 asdfghjkl Litao and more!

Moli 茉莉兔兔

91,082 Aufrufe • vor 1 Jahr

🔥THE TRUTH, for those with balls and brains enough to face it—Fauci is not the main criminal behind Covid. That's a "fairytale" The REAL PERPETRATOR in the U.S. is the National Security Council, which created the Covid policies! Don't let gasbag Bland Paul gaslight you! Retired pharma R&D executive Sasha Latypova (sashalatypova.substack.com "Due Diligence and Art") explains: "[Fauci being the main driver behind Covid is] a fairy tale for adults, really. I mean, it's an absolutely naive, infantile view of things" "Yeah, he's in charge of a bunch of science money to be distributed. So how does he have jurisdiction over 50 states to dictate masking policies, for example?" "Everybody's saying, 'Oh! Fauci pushed for masks!' Well, push ahead for anything you want. But how do you have jurisdiction over 50 states to enact almost identical policies for masking? How do you have jurisdiction not only over 50 states, but 100-and-what-85 countries of the world? To enact identical policies?" "On what legal basis does [this] megalomaniac run the world? ... how does he coordinate all of that in the entire country and the entire world?" "And, we know. We know what kind of power. We've demonstrated it. It's the National Security Council—it's the military. It's the bioterrorism doctrines, the pandemic emergency doctrines, the completely unconstitutional set of laws that have been put on the books over 20 years in preparation for this event by all countries together"
1:43

🔥THE TRUTH, for those with balls and brains enough to face it—Fauci is not the main criminal behind Covid. That's a "fairytale" The REAL PERPETRATOR in the U.S. is the National Security Council, which created the Covid policies! Don't let gasbag Bland Paul gaslight you! Retired pharma R&D executive Sasha Latypova (sashalatypova.substack.com "Due Diligence and Art") explains: "[Fauci being the main driver behind Covid is] a fairy tale for adults, really. I mean, it's an absolutely naive, infantile view of things" "Yeah, he's in charge of a bunch of science money to be distributed. So how does he have jurisdiction over 50 states to dictate masking policies, for example?" "Everybody's saying, 'Oh! Fauci pushed for masks!' Well, push ahead for anything you want. But how do you have jurisdiction over 50 states to enact almost identical policies for masking? How do you have jurisdiction not only over 50 states, but 100-and-what-85 countries of the world? To enact identical policies?" "On what legal basis does [this] megalomaniac run the world? ... how does he coordinate all of that in the entire country and the entire world?" "And, we know. We know what kind of power. We've demonstrated it. It's the National Security Council—it's the military. It's the bioterrorism doctrines, the pandemic emergency doctrines, the completely unconstitutional set of laws that have been put on the books over 20 years in preparation for this event by all countries together"

Sense Receptor

20,379 Aufrufe • vor 21 Tagen

🚨 THE CONTAGION HAS REACHED RESALE HOUSING. According to real estate lawyer Mark Morris (@markinmetaform.bsky.social), this is no longer a condo problem or a pre construction problem. Average ground level homes in suburban markets are now repricing because 🇨🇦 incomes simply cannot support where prices went. Power of sales are surging. Law firms are now handling two to three power of sale files every single week. Realtor education sessions have shifted almost entirely to one question. How do power of sales even work when there is no equity left. This is not a short term shock. By the time a property actually sells under power of sale, the process has already taken six to ten months. Meaning today’s listings reflect distress from last year and more is already baked in. After years of speculation masking the fundamentals, housing is being pulled back to income reality. The growth areas for real estate lawyers right now are power of sales and failed new build closings. 📎 Ron Butler with Mark Morris (@markinmetaform.bsky.social)
2:15

🚨 THE CONTAGION HAS REACHED RESALE HOUSING. According to real estate lawyer Mark Morris (@markinmetaform.bsky.social), this is no longer a condo problem or a pre construction problem. Average ground level homes in suburban markets are now repricing because 🇨🇦 incomes simply cannot support where prices went. Power of sales are surging. Law firms are now handling two to three power of sale files every single week. Realtor education sessions have shifted almost entirely to one question. How do power of sales even work when there is no equity left. This is not a short term shock. By the time a property actually sells under power of sale, the process has already taken six to ten months. Meaning today’s listings reflect distress from last year and more is already baked in. After years of speculation masking the fundamentals, housing is being pulled back to income reality. The growth areas for real estate lawyers right now are power of sales and failed new build closings. 📎 Ron Butler with Mark Morris (@markinmetaform.bsky.social)

Shazi

18,626 Aufrufe • vor 7 Monaten

Zero VFX experience. Just Claude, Seedance 2.0, and a phone clip. The result looks like a $10M scene In this tutorial you'll learn how to create cinematic AI VFX from your own real footage — no green screen, no masking, no compositing, no render times. Claude doesn't touch the video — it reads your clip frame by frame and writes the prompt, locking your face, camera move, light, and gestures, then naming the one thing to change and when. That prompt goes into Seedance 2.0, which only touches what was named. That's the trick. You're not describing an effect and hoping the AI gets it. Claude studies the footage first and hands Seedance a locked, precise instruction — which is why a hand can morph mid-sentence with zero visible cut, or neon reflections land on the dashboard automatically, with nothing graded by hand. One clip in, one sentence to Claude, one prompt into Seedance 2.0 — and out comes a shot that would normally cost a studio a month. ▪ 00:00 Intro ▪ 01:28 The 3 levels of AI VFX ▪ 01:59 Level 1 — Swapping the world (walking) ▪ 04:32 Level 2 — Changing an element ▪ 07:26 Level 3 — The handheld showcase ▪ 10:15 Outro YouTube: "Higgsfield AI - I Added Insane AI VFX to Real Footage in 4K(Seedance 2.0)" Thank you for sharing this free tutorial with us. It is invaluable information.
11:05

Zero VFX experience. Just Claude, Seedance 2.0, and a phone clip. The result looks like a $10M scene In this tutorial you'll learn how to create cinematic AI VFX from your own real footage — no green screen, no masking, no compositing, no render times. Claude doesn't touch the video — it reads your clip frame by frame and writes the prompt, locking your face, camera move, light, and gestures, then naming the one thing to change and when. That prompt goes into Seedance 2.0, which only touches what was named. That's the trick. You're not describing an effect and hoping the AI gets it. Claude studies the footage first and hands Seedance a locked, precise instruction — which is why a hand can morph mid-sentence with zero visible cut, or neon reflections land on the dashboard automatically, with nothing graded by hand. One clip in, one sentence to Claude, one prompt into Seedance 2.0 — and out comes a shot that would normally cost a studio a month. ▪ 00:00 Intro ▪ 01:28 The 3 levels of AI VFX ▪ 01:59 Level 1 — Swapping the world (walking) ▪ 04:32 Level 2 — Changing an element ▪ 07:26 Level 3 — The handheld showcase ▪ 10:15 Outro YouTube: "Higgsfield AI - I Added Insane AI VFX to Real Footage in 4K(Seedance 2.0)" Thank you for sharing this free tutorial with us. It is invaluable information.

Zentrix⌚️

28,606 Aufrufe • vor 25 Tagen

anxin’s self keyword is “angel cat”… my pfp is real 🐱🪽 🐶 the person who looks the most like a cat! 👤 ah! anxin! right? 🐱 right, how did you guess it? 👤 because you really have a cat face 👤 but why did you put angel in front by yourself? 🐶 because he descended from the sky 🐱 because i think i’m so nice ☺️ 👤 can you share some instances where you are like an angel? 🐱 there’s so many… 🦊 how about the members talk about it instead 🐱 hehe if i talk about it myself it’s awkward 🦁 even though anxin is in the maknae line, he takes care of the hyungs very well. he also plays with the maknae sanghyeon well. you don’t expect it, but there are a lot of moments where he is sensitive and takes care of the members well. “angel cat” suits him very well
1:16

anxin’s self keyword is “angel cat”… my pfp is real 🐱🪽 🐶 the person who looks the most like a cat! 👤 ah! anxin! right? 🐱 right, how did you guess it? 👤 because you really have a cat face 👤 but why did you put angel in front by yourself? 🐶 because he descended from the sky 🐱 because i think i’m so nice ☺️ 👤 can you share some instances where you are like an angel? 🐱 there’s so many… 🦊 how about the members talk about it instead 🐱 hehe if i talk about it myself it’s awkward 🦁 even though anxin is in the maknae line, he takes care of the hyungs very well. he also plays with the maknae sanghyeon well. you don’t expect it, but there are a lot of moments where he is sensitive and takes care of the members well. “angel cat” suits him very well

🥟

22,244 Aufrufe • vor 2 Monaten

Streetopia - Get Out Of Reet 🎬 Introducing the initial installment of an evolving tale brought to life by Generative AI. This captivating narrative has emerged from numerous poorman experiments and showcases the potential for an ongoing series of bite-sized episodes, each lasting under a minute. How I did do it? 1. All 3-4 shots of me are on green screen. 2. Backgrounds were created in a photo editor using stock pictures. Mostly stitched. Different aspect ratios. 3. I downloaded static 3D models & Animated models (greenscreen/alpha) and played with them in my motion graphics editor. Filled in some storefront windows with more videos. 4. Did some masking magic, camera movement, and played with layers and depth... for example, I masked the sky to add a storm and t-rex 🦖 5. Sent individual clips to Superstudio (in 4k) 6. Exported favorite renders (in 4k) and sent them to my video editor for choppin', sound design (poppin'), and voiceover (in 4k). See thread for examples. Challenges: • I wish I had a treadmill to actually look like I'm walking. • I was lucky with frame consistency. Except for the wide shot that fuzzed my face (me on the couch). I re-did the shot on Kaiber in a close-up and manually key-framed my face on that wide shot. • I couldn't do much about the buggy eyes in the last clip. • No butterflies were harmed in this video. How long did it take: • A day and a half with tons of happy accidents • How long will it take you? Probably much longer if you don't know how to shoot, edit, design, composite, edit sound, or know where to look for assets. But it's not impossible! What's the plan? To continue the story while still learning and experimenting along the way. This is a much longer process. Very tiring. Might try doing it weekly or bi-weekly. What's the story about? In the quest to escape Reet, everyone yearns for freedom. Ricardo, however, carries a hint of cynicism. He finds pleasure in the disorder and may seem reluctant to lend a hand. Nevertheless, the encounters with diverse characters along his odyssey will ignite a transformation within him, although not without challenges. Gradually, he will come to realize the profound truth: "Helping others is away is a way of helping ourselves." HQ Video: #GenerativeAI #ArtistOnTwitter #aiartcommunity #AIArtistCommunity #video2video
0:41

Streetopia - Get Out Of Reet 🎬 Introducing the initial installment of an evolving tale brought to life by Generative AI. This captivating narrative has emerged from numerous poorman experiments and showcases the potential for an ongoing series of bite-sized episodes, each lasting under a minute. How I did do it? 1. All 3-4 shots of me are on green screen. 2. Backgrounds were created in a photo editor using stock pictures. Mostly stitched. Different aspect ratios. 3. I downloaded static 3D models & Animated models (greenscreen/alpha) and played with them in my motion graphics editor. Filled in some storefront windows with more videos. 4. Did some masking magic, camera movement, and played with layers and depth... for example, I masked the sky to add a storm and t-rex 🦖 5. Sent individual clips to Superstudio (in 4k) 6. Exported favorite renders (in 4k) and sent them to my video editor for choppin', sound design (poppin'), and voiceover (in 4k). See thread for examples. Challenges: • I wish I had a treadmill to actually look like I'm walking. • I was lucky with frame consistency. Except for the wide shot that fuzzed my face (me on the couch). I re-did the shot on Kaiber in a close-up and manually key-framed my face on that wide shot. • I couldn't do much about the buggy eyes in the last clip. • No butterflies were harmed in this video. How long did it take: • A day and a half with tons of happy accidents • How long will it take you? Probably much longer if you don't know how to shoot, edit, design, composite, edit sound, or know where to look for assets. But it's not impossible! What's the plan? To continue the story while still learning and experimenting along the way. This is a much longer process. Very tiring. Might try doing it weekly or bi-weekly. What's the story about? In the quest to escape Reet, everyone yearns for freedom. Ricardo, however, carries a hint of cynicism. He finds pleasure in the disorder and may seem reluctant to lend a hand. Nevertheless, the encounters with diverse characters along his odyssey will ignite a transformation within him, although not without challenges. Gradually, he will come to realize the profound truth: "Helping others is away is a way of helping ourselves." HQ Video: #GenerativeAI #ArtistOnTwitter #aiartcommunity #AIArtistCommunity #video2video

Sway Molina

49,677 Aufrufe • vor 3 Jahren

Many people today believe that Hezbollah simply took over Lebanon under the authority of Iran. They say the group has kept the Christian population subdued and oppressed, while converting the country step by step into an Islamic state. But the reality is that it isn’t quite true. The story has more layers than that. The Christian Maronites aren’t innocent bystanders in all this. They made alliances that shaped the country’s path. They aligned themselves with the Assad regime of Syria, and it was Syria who really dictated how things ran in Lebanon for a long time. Hezbollah were aligned with Assad as well, and so was Iran. In 1998, this led to the appointment of former military chief Emile Lahoud. He was a Maronite Christian who was fully aligned with Assad. He sold out many Christian communities in Lebanon to retain power under the influence of the Assad regime with Hezbollah as his “muscle”. Lahoud supported Hezbollah and allowed them free rein in Lebanon. He gave them full control of southern Lebanon in exchange for them getting rid of his rivals and dissidents. He backed them completely. Many Christians supported Hezbollah in ending Israel’s occupation of the south, and even in their attacks on Israel alongside Hamas. It was a common position at the time among certain groups. Lahoud was also likely responsible for ordering the assassination of Rafik Hariri in 2005. Hariri was a vocal opponent to many of Lahoud’s policies and his close ties with Assad, Hezbollah and Tehran. His killing was a major turning point that highlighted the deep connections. This means the Christians weren’t simply hostages in their own country. They weren’t oppressed in the way many describe. They were complicit in the arrangements that allowed Hezbollah to grow stronger. In 2003, 60 Minutes even produced and aired a documentary on it all where the Lebanese President, Emile Lahoud, admitted it all openly. I’ve attached part of this 60 Minutes documentary below where you will see and hear how Lahoud gave Hezbollah Southern Lebanon, how Hezbollah trained divisions of suicide bombers to attack Israeli cities and support Hamas, indoctrinated thousands of children to fight for them similar to UNRWA and Hamas, and how Hezbollah viewed the US as Satan. Today, we see Christian leaders of Lebanon pretending otherwise. They act as if they have always been victims with no say in the matter. But in fact, they support Hezbollah or at least go along with the current setup. This explains why they refuse to disarm them or take real action against their influence. There is documented evidence to prove all of the above, from official records and political histories to investigations into these events. And looking at the bigger picture, all these parties, from the Christians, Hezbollah, Assad’s Syria, and the Iranian regime, all were supported politically and economically by one main global power: France. Their backing played a key role in how things developed over the decades. Understanding this true control of Lebanon helps cut through the simple stories we often hear. It shows how local groups, including Christian leaders, made choices that contributed to the situation. It’s not just about one external force imposing its will. The complicity goes deeper, and facing that history is important if Lebanon is ever to move forward properly.
2:32

Many people today believe that Hezbollah simply took over Lebanon under the authority of Iran. They say the group has kept the Christian population subdued and oppressed, while converting the country step by step into an Islamic state. But the reality is that it isn’t quite true. The story has more layers than that. The Christian Maronites aren’t innocent bystanders in all this. They made alliances that shaped the country’s path. They aligned themselves with the Assad regime of Syria, and it was Syria who really dictated how things ran in Lebanon for a long time. Hezbollah were aligned with Assad as well, and so was Iran. In 1998, this led to the appointment of former military chief Emile Lahoud. He was a Maronite Christian who was fully aligned with Assad. He sold out many Christian communities in Lebanon to retain power under the influence of the Assad regime with Hezbollah as his “muscle”. Lahoud supported Hezbollah and allowed them free rein in Lebanon. He gave them full control of southern Lebanon in exchange for them getting rid of his rivals and dissidents. He backed them completely. Many Christians supported Hezbollah in ending Israel’s occupation of the south, and even in their attacks on Israel alongside Hamas. It was a common position at the time among certain groups. Lahoud was also likely responsible for ordering the assassination of Rafik Hariri in 2005. Hariri was a vocal opponent to many of Lahoud’s policies and his close ties with Assad, Hezbollah and Tehran. His killing was a major turning point that highlighted the deep connections. This means the Christians weren’t simply hostages in their own country. They weren’t oppressed in the way many describe. They were complicit in the arrangements that allowed Hezbollah to grow stronger. In 2003, 60 Minutes even produced and aired a documentary on it all where the Lebanese President, Emile Lahoud, admitted it all openly. I’ve attached part of this 60 Minutes documentary below where you will see and hear how Lahoud gave Hezbollah Southern Lebanon, how Hezbollah trained divisions of suicide bombers to attack Israeli cities and support Hamas, indoctrinated thousands of children to fight for them similar to UNRWA and Hamas, and how Hezbollah viewed the US as Satan. Today, we see Christian leaders of Lebanon pretending otherwise. They act as if they have always been victims with no say in the matter. But in fact, they support Hezbollah or at least go along with the current setup. This explains why they refuse to disarm them or take real action against their influence. There is documented evidence to prove all of the above, from official records and political histories to investigations into these events. And looking at the bigger picture, all these parties, from the Christians, Hezbollah, Assad’s Syria, and the Iranian regime, all were supported politically and economically by one main global power: France. Their backing played a key role in how things developed over the decades. Understanding this true control of Lebanon helps cut through the simple stories we often hear. It shows how local groups, including Christian leaders, made choices that contributed to the situation. It’s not just about one external force imposing its will. The complicity goes deeper, and facing that history is important if Lebanon is ever to move forward properly.

Mor Edge Insight

30,387 Aufrufe • vor 4 Monaten

NEW RESEARCH: You can now create a new robot optimized for any given task! I love this new project by Huy Ha, Shuran Song, and others. Called "Transformer Transformer: A Unified Model for Motion-Conditioned Robot Co-design", it generates a robot's physical design and its controller together from a task spec. DEFINITIONS: - Reward function: A scoring rule that assigns a number to how well a behavior achieves the task. Here, it is the objective the generated design is pushed to maximize (e.g., track the target motion with low error). - Tokenizing: dividing continuous or structured data (a robot's links, joints, motor specs, states, actions) into a discrete vocabulary of symbols a transformer can process, the same step that turned pixels and audio into "language" for these models. - Diffusion transformer (DiT): A transformer trained to turn random noise into structured output through iterative denoising. Here, it generates robot bodies and trajectories instead of images. - MuJoCo: The standard fast physics simulator for robotics research (DeepMind-maintained). The Menagerie is its curated zoo of ready-to-use robot models. - CMA-ES: Covariance Matrix Adaptation Evolution Strategy, the workhorse black-box optimizer: it evolves a population of candidate designs, keeps the best, and needs thousands of simulator rollouts. - Bimanual multi-trajectory optimization: Finding one design/controller that performs well across several target motions for a two-armed robot at once, harder than optimizing for a single arm and a single motion. - BERT/MAE masked-modeling trick: Train one model to fill in whatever parts of the input you hide (words for BERT, image patches for MAE); at inference, choosing what to mask chooses the task, so masking the body makes it a designer and masking the actions makes it a controller. In practice, you give it a target end-effector motion and a reward function, and it outputs a complete embodiment (link, joint, motor, and inertial property), as well as a controller to drive it. It works by tokenizing both the body (links/joints/motors) and the dynamics (states/actions) into a compact scheme called RoboTokens, training a diffusion transformer (DiT) over them. The same model predicts dynamics using those predictions ("Dynamics Self-Guidance") to push generated designs toward higher reward at inference time. Masking different token types (using the BERT/MAE masked-modeling trick) lets the one model do three jobs: generate an embodiment, control an arbitrary embodiment, or design one conditioned on a motion. It is trained on 11 robots from the MuJoCo Menagerie (0.65 kg hand to 67.5 kg quadruped, 6–35 joints), and validated in sim and on a physical ALOHA doing cloth flinging. I like the fact that this approach inverts the entire recent robotics ideas: designing a policy for a fixed robot -> designing the robot for a fixed task. Every other approach assumes the body is given and learns a controller. Transformer Transformer takes the task (target motion + reward), then generates the body and controller jointly. In practice, it is a ~180× speedup over the standard optimizer at equal-or-better quality. It reaches "CMA-ES-level quality in seconds" and finishes bimanual multi-trajectory optimization in that is worth underlining nowadays! Also worth mentioning: this is the lab behind UMI and Handroid, that I mentioned here previously! The team seems extremely creative, i love these out-of-the-box approaches. Enjoy watching the demo of robot optimization in 3D, data acquisition, then real-life testing:
0:28

NEW RESEARCH: You can now create a new robot optimized for any given task! I love this new project by Huy Ha, Shuran Song, and others. Called "Transformer Transformer: A Unified Model for Motion-Conditioned Robot Co-design", it generates a robot's physical design and its controller together from a task spec. DEFINITIONS: - Reward function: A scoring rule that assigns a number to how well a behavior achieves the task. Here, it is the objective the generated design is pushed to maximize (e.g., track the target motion with low error). - Tokenizing: dividing continuous or structured data (a robot's links, joints, motor specs, states, actions) into a discrete vocabulary of symbols a transformer can process, the same step that turned pixels and audio into "language" for these models. - Diffusion transformer (DiT): A transformer trained to turn random noise into structured output through iterative denoising. Here, it generates robot bodies and trajectories instead of images. - MuJoCo: The standard fast physics simulator for robotics research (DeepMind-maintained). The Menagerie is its curated zoo of ready-to-use robot models. - CMA-ES: Covariance Matrix Adaptation Evolution Strategy, the workhorse black-box optimizer: it evolves a population of candidate designs, keeps the best, and needs thousands of simulator rollouts. - Bimanual multi-trajectory optimization: Finding one design/controller that performs well across several target motions for a two-armed robot at once, harder than optimizing for a single arm and a single motion. - BERT/MAE masked-modeling trick: Train one model to fill in whatever parts of the input you hide (words for BERT, image patches for MAE); at inference, choosing what to mask chooses the task, so masking the body makes it a designer and masking the actions makes it a controller. In practice, you give it a target end-effector motion and a reward function, and it outputs a complete embodiment (link, joint, motor, and inertial property), as well as a controller to drive it. It works by tokenizing both the body (links/joints/motors) and the dynamics (states/actions) into a compact scheme called RoboTokens, training a diffusion transformer (DiT) over them. The same model predicts dynamics using those predictions ("Dynamics Self-Guidance") to push generated designs toward higher reward at inference time. Masking different token types (using the BERT/MAE masked-modeling trick) lets the one model do three jobs: generate an embodiment, control an arbitrary embodiment, or design one conditioned on a motion. It is trained on 11 robots from the MuJoCo Menagerie (0.65 kg hand to 67.5 kg quadruped, 6–35 joints), and validated in sim and on a physical ALOHA doing cloth flinging. I like the fact that this approach inverts the entire recent robotics ideas: designing a policy for a fixed robot -> designing the robot for a fixed task. Every other approach assumes the body is given and learns a controller. Transformer Transformer takes the task (target motion + reward), then generates the body and controller jointly. In practice, it is a ~180× speedup over the standard optimizer at equal-or-better quality. It reaches "CMA-ES-level quality in seconds" and finishes bimanual multi-trajectory optimization in that is worth underlining nowadays! Also worth mentioning: this is the lab behind UMI and Handroid, that I mentioned here previously! The team seems extremely creative, i love these out-of-the-box approaches. Enjoy watching the demo of robot optimization in 3D, data acquisition, then real-life testing:

Léo

25,896 Aufrufe • vor 13 Tagen

⚠️ WARNING: Think Twice Before Visiting the Illegally Occupied Areas of Cyprus!⚠️ Travelers, please reconsider any plans to visit the so-called “turkish republic of northern Cyprus”, an area under illegal Turkish occupation since 1974, recognized only by Turkey and condemned internationally. Your “dream” vacation could turn into a nightmare due to serious safety and legal risks that many tourists overlook. First and foremost, visitors face the real risk of arbitrary detention or being held without due process, essentially like being abducted and held hostage, by the occupying regime who often ignore legal protections. Detention facilities there fall short of international standards, and the “police” has been accused of coerced confessions and violations of rights. Even minor infractions, like taking photos or looking suspicious in their eyes, can lead to severe penalties, long jail terms, or heavy fines, with zero tolerance enforced strictly. Compounding this danger, there’s no international diplomatic representation in the occupied north because all countries don’t recognize this illegal entity. If you’re detained or in trouble, contacting your embassy for help will be extremely difficult, leaving you isolated and vulnerable with minimal consular support. Human rights reports highlight how arbitrary arrests occur, and without proper access to your government, resolving issues could drag on indefinitely. If you’re thinking of buying property? Don’t! Much of the land in the occupied areas is stolen from displaced Greek Cypriots. Purchasing or exploiting it is illegal under Cypriot and international law, making you complicit in a criminal act. You’ll face civil lawsuits from rightful owners, and penalties can include years imprisonment, triggering European Arrest Warrants or international prosecution anywhere in the EU and beyond. Over 120 such warrants have already been issued, and courts like the European Court of Human Rights have repeatedly ruled against these transactions. Any “title” you get is null and void, you will lose everything. This isn’t just politics, it’s about your safety, freedom, and avoiding lifelong legal headaches. Support justice by boycotting the occupied areas, visit the legitimate Republic of Cyprus instead.
0:39

⚠️ WARNING: Think Twice Before Visiting the Illegally Occupied Areas of Cyprus!⚠️ Travelers, please reconsider any plans to visit the so-called “turkish republic of northern Cyprus”, an area under illegal Turkish occupation since 1974, recognized only by Turkey and condemned internationally. Your “dream” vacation could turn into a nightmare due to serious safety and legal risks that many tourists overlook. First and foremost, visitors face the real risk of arbitrary detention or being held without due process, essentially like being abducted and held hostage, by the occupying regime who often ignore legal protections. Detention facilities there fall short of international standards, and the “police” has been accused of coerced confessions and violations of rights. Even minor infractions, like taking photos or looking suspicious in their eyes, can lead to severe penalties, long jail terms, or heavy fines, with zero tolerance enforced strictly. Compounding this danger, there’s no international diplomatic representation in the occupied north because all countries don’t recognize this illegal entity. If you’re detained or in trouble, contacting your embassy for help will be extremely difficult, leaving you isolated and vulnerable with minimal consular support. Human rights reports highlight how arbitrary arrests occur, and without proper access to your government, resolving issues could drag on indefinitely. If you’re thinking of buying property? Don’t! Much of the land in the occupied areas is stolen from displaced Greek Cypriots. Purchasing or exploiting it is illegal under Cypriot and international law, making you complicit in a criminal act. You’ll face civil lawsuits from rightful owners, and penalties can include years imprisonment, triggering European Arrest Warrants or international prosecution anywhere in the EU and beyond. Over 120 such warrants have already been issued, and courts like the European Court of Human Rights have repeatedly ruled against these transactions. Any “title” you get is null and void, you will lose everything. This isn’t just politics, it’s about your safety, freedom, and avoiding lifelong legal headaches. Support justice by boycotting the occupied areas, visit the legitimate Republic of Cyprus instead.

Harry Theocharous

30,503 Aufrufe • vor 11 Monaten

Senator Ron Johnson just took three minutes to EXPOSE in stunning detail the vast government conspiracy to cover up Covid vaccine reactions. He called out the mainstream media for REFUSING to report on it while the CIA whistleblower listened intently. JOHNSON: “This is what we uncovered, this is what is now being covered up by the legacy media.” “On March 1st of 2021, Dr. Peter Marks, head of the CBER — organization at the FDA that approves vaccines and also supposed to surveil those vaccines post-market for safety — was given a multi-page briefing by the main data mining analysts within FDA working with the inventor of the algorithm they used to analyze the VAERS, the vaccine adverse event reporting system.” “That briefing told Dr. Marks that the current algorithm because of the nature of the massive number of claims coming into VAERS. From 1990 until 2020 on average 280 deaths were reported in VAERS associated with vaccine. 2021 it was over 21,000. Nothing to see here.” “26 days later a new state-of-the-art algorithm developed by the inventor of the original algorithm was presented to senior FDA officials showing 49 cases of extreme masking, 25 safety signals, including sudden cardiac death, pulmonary infarction, Bell’s Palsy, different types of strokes.” “Month after month after month they continued to do this data mining, they continued to show additional safety signals.” “Did Peter Marks, did Anthony Fauci, did President Biden go to the American public and say hang on here, we’ve got some data, we’ve got some concerns? No! No! They blamed the unvaccinated for continuing the pandemic. This is a pandemic of the unvaccinated. We need to mandate this.” “This is what government does and what the deep state does... I need the mainstream media, CBS, ABC, CNN, to step forward and start covering what is a MAJOR scandal!” “How many thousands, tens, hundreds of thousands of people are permanently disabled or possibly lost their life because our FDA hid the fact that there were safety signals SCREAMING at them with the covid injection?” The FDA literally hid safety signals that were screaming at them.
3:20

Senator Ron Johnson just took three minutes to EXPOSE in stunning detail the vast government conspiracy to cover up Covid vaccine reactions. He called out the mainstream media for REFUSING to report on it while the CIA whistleblower listened intently. JOHNSON: “This is what we uncovered, this is what is now being covered up by the legacy media.” “On March 1st of 2021, Dr. Peter Marks, head of the CBER — organization at the FDA that approves vaccines and also supposed to surveil those vaccines post-market for safety — was given a multi-page briefing by the main data mining analysts within FDA working with the inventor of the algorithm they used to analyze the VAERS, the vaccine adverse event reporting system.” “That briefing told Dr. Marks that the current algorithm because of the nature of the massive number of claims coming into VAERS. From 1990 until 2020 on average 280 deaths were reported in VAERS associated with vaccine. 2021 it was over 21,000. Nothing to see here.” “26 days later a new state-of-the-art algorithm developed by the inventor of the original algorithm was presented to senior FDA officials showing 49 cases of extreme masking, 25 safety signals, including sudden cardiac death, pulmonary infarction, Bell’s Palsy, different types of strokes.” “Month after month after month they continued to do this data mining, they continued to show additional safety signals.” “Did Peter Marks, did Anthony Fauci, did President Biden go to the American public and say hang on here, we’ve got some data, we’ve got some concerns? No! No! They blamed the unvaccinated for continuing the pandemic. This is a pandemic of the unvaccinated. We need to mandate this.” “This is what government does and what the deep state does... I need the mainstream media, CBS, ABC, CNN, to step forward and start covering what is a MAJOR scandal!” “How many thousands, tens, hundreds of thousands of people are permanently disabled or possibly lost their life because our FDA hid the fact that there were safety signals SCREAMING at them with the covid injection?” The FDA literally hid safety signals that were screaming at them.

Overton

50,905 Aufrufe • vor 3 Monaten

Researchers found a way to make LLMs 8.5x faster! (without compromising accuracy) Speculative decoding is quite an effective way to address the single-token bottleneck in traditional LLM inference. A small "draft" model first generates the next several tokens, then the large model verifies all of them at once in a single forward pass. If a token at any position is wrong, you keep everything before it and restart from there. This never does worse than normal decoding. But current drafters in Speculative decoding still guess one token at a time. That makes the drafting step itself a bottleneck, capping real-world speedups at 2-3x. DFlash is a new technique that swaps the autoregressive drafter with a lightweight block diffusion model that guesses all tokens in one parallel shot. Drafting cost stays flat no matter how many tokens you speculate. On top of that, the drafter is conditioned on hidden features pulled from multiple layers of the target model and injected into every draft layer, so it makes significantly better guesses than a drafter working from scratch. In the side-by-side demo below, vanilla decoding runs at 48.5 tokens/sec. DFlash hits 415 tokens/sec on the same model, with zero quality loss. It's already integrated with vLLM, SGLang, and Transformers, with draft models on HuggingFace for several models like Qwen3, Qwen3.5, Llama 3.1, Kimi-K2.5, gpt-oss, and many more. I have shared the GitHub repo in the replies! KV caching is another must-know technique to boost LLM inference. I recently wrote an article about it. Read it below. 👉 Over to you: What use case are you working on that can benefit from this new technique?
0:26

Researchers found a way to make LLMs 8.5x faster! (without compromising accuracy) Speculative decoding is quite an effective way to address the single-token bottleneck in traditional LLM inference. A small "draft" model first generates the next several tokens, then the large model verifies all of them at once in a single forward pass. If a token at any position is wrong, you keep everything before it and restart from there. This never does worse than normal decoding. But current drafters in Speculative decoding still guess one token at a time. That makes the drafting step itself a bottleneck, capping real-world speedups at 2-3x. DFlash is a new technique that swaps the autoregressive drafter with a lightweight block diffusion model that guesses all tokens in one parallel shot. Drafting cost stays flat no matter how many tokens you speculate. On top of that, the drafter is conditioned on hidden features pulled from multiple layers of the target model and injected into every draft layer, so it makes significantly better guesses than a drafter working from scratch. In the side-by-side demo below, vanilla decoding runs at 48.5 tokens/sec. DFlash hits 415 tokens/sec on the same model, with zero quality loss. It's already integrated with vLLM, SGLang, and Transformers, with draft models on HuggingFace for several models like Qwen3, Qwen3.5, Llama 3.1, Kimi-K2.5, gpt-oss, and many more. I have shared the GitHub repo in the replies! KV caching is another must-know technique to boost LLM inference. I recently wrote an article about it. Read it below. 👉 Over to you: What use case are you working on that can benefit from this new technique?

Avi Chawla

157,390 Aufrufe • vor 3 Monaten