Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Using vDISCO tissue clearing, and a new ultrabright chemical labeling strategy, this Neuron paper introduces LINCS, a powerful method for rapid 3D connectivity mapping in large tissue samples, from whole brains to whole bodies. A very nice advance for scalable circuit mapping, especially by combining strong, uniform labeling with... show more

Ali Max Erturk

90,695 subscribers

14,962 views • 5 months ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

E11 Bio is excited to unveil PRISM technology for mapping brain wiring with simple light microscopes. Today, brain mapping in humans and other mammals is bottlenecked by accurate neuron tracing. PRISM uses molecular ID codes and AI to help neurons trace themselves. We discovered a new cell barcoding approach exceeding comparable methods by more than 750x. This is the heart of PRISM. We integrated this capability with microscopy and AI image analysis to automatically trace neurons at high resolution and annotate them with molecular features. This is a key advance towards economically viable brain mapping - 95% of costs stem from neuron tracing. It is also an important step towards democratizing neuron tracing for everyday neuroscience. Solving these problems is critical for curing brain disorders, building safer and human-like AI, and even simulating brain function. In our first pilot study, we acquired a unique dataset in mouse hippocampus. Barcodes improved the accuracy of tracing genetically labelled neurons by 8x – with a clear path to 100x or more. They also permit tracing across spatial gaps – essential for mitigating tissue section loss in whole-brain scaling. Using molecular annotation, we uncover an intriguing feature of synaptic organization, demonstrating how PRISM can be used for systematic discovery 🧵

E11 Bio is excited to unveil PRISM technology for mapping brain wiring with simple light microscopes. Today, brain mapping in humans and other mammals is bottlenecked by accurate neuron tracing. PRISM uses molecular ID codes and AI to help neurons trace themselves. We discovered a new cell barcoding approach exceeding comparable methods by more than 750x. This is the heart of PRISM. We integrated this capability with microscopy and AI image analysis to automatically trace neurons at high resolution and annotate them with molecular features. This is a key advance towards economically viable brain mapping - 95% of costs stem from neuron tracing. It is also an important step towards democratizing neuron tracing for everyday neuroscience. Solving these problems is critical for curing brain disorders, building safer and human-like AI, and even simulating brain function. In our first pilot study, we acquired a unique dataset in mouse hippocampus. Barcodes improved the accuracy of tracing genetically labelled neurons by 8x – with a clear path to 100x or more. They also permit tracing across spatial gaps – essential for mitigating tissue section loss in whole-brain scaling. Using molecular annotation, we uncover an intriguing feature of synaptic organization, demonstrating how PRISM can be used for systematic discovery 🧵

Andrew Payne

112,354 views • 10 months ago

I've been editing this article about "brain mapping" and connectomics, and I'm just stunned by how quickly the cost estimates to map, say, a mouse brain have plummeted in just the last couple years. It actually seems feasible that we could map the entire human brain -- all 86 billion neurons, and their connections -- in this lifetime. In the 1970s, Sydney Brenner started mapping all the connections between neurons in C. elegans. His team sliced the worm into thin pieces, took photos using an electron microscope, and manually traced and reconstructed each synapse for 302 neurons total. This project took more than a decade of work, and it cost about $16,500 to reconstruct each neuron. Scaling this up to a human brain boggles the mind. Electron microscopy remained the norm in connectomics for decades, because it was the only option available to see synapses at a resolution high enough to be able to trace their paths. Each electron microscope costs several hundreds of thousands of dollars, though, and you need lots of them to map even a mouse brain in a reasonable timeframe. In 2023, the Wellcome Trust released a report estimating how long, and how expensive, it would be to map the mouse connectome (~70M neurons). They estimated that imaging alone would cost $200-300M, and that proofreading (or ensuring that traces between neurons are correct) would cost $7-21 BILLION. (A human can only manually trace about 1 mm of neuron per hour.) Also, the images would occupy about 500 petabytes of data, and getting those data would require 20 electron microscopes running in parallel for about 5 years, continuously. They estimated the whole project would take about 17 years of work. This is, understandably, insane. But now it seems like there's an actual path toward mapping the full mouse brain in about five years for ~$100M dollars. There have been three major breakthroughs in the last year or so: 1/ Expansion microscopy, first developed in 2015, showed that it's possible to "enlarge" the brain by about 5x using a swellable polymer. But an improved method increases this number to >20x expansion, meaning we can now expand brains and image neurons much more easily using cheap light microscopes, rather than expensive electron ones. 2/ E11 Bio (a nonprofit research org) developed protein barcodes that get delivered into brain tissue; each neuron gets a unique combination of barcodes. These cells are then stained with colorful antibodies, which stick to a matching protein barcode, causing each neuron to light up in a distinct color. This makes tracing neurons so much easier. 3/ Google Research released PATHFINDER this May, an AI-based neuron tracing tool that can proofread about 67,200 cubic microns of brain tissue per hour, with very high accuracy. It works on electron micrographs, but something similar could be presumably be developed for the E11 / colorful tag approach. This is an extremely exciting time for neuroscience. (C. elegans connectome below.)

I've been editing this article about "brain mapping" and connectomics, and I'm just stunned by how quickly the cost estimates to map, say, a mouse brain have plummeted in just the last couple years. It actually seems feasible that we could map the entire human brain -- all 86 billion neurons, and their connections -- in this lifetime. In the 1970s, Sydney Brenner started mapping all the connections between neurons in C. elegans. His team sliced the worm into thin pieces, took photos using an electron microscope, and manually traced and reconstructed each synapse for 302 neurons total. This project took more than a decade of work, and it cost about $16,500 to reconstruct each neuron. Scaling this up to a human brain boggles the mind. Electron microscopy remained the norm in connectomics for decades, because it was the only option available to see synapses at a resolution high enough to be able to trace their paths. Each electron microscope costs several hundreds of thousands of dollars, though, and you need lots of them to map even a mouse brain in a reasonable timeframe. In 2023, the Wellcome Trust released a report estimating how long, and how expensive, it would be to map the mouse connectome (~70M neurons). They estimated that imaging alone would cost $200-300M, and that proofreading (or ensuring that traces between neurons are correct) would cost $7-21 BILLION. (A human can only manually trace about 1 mm of neuron per hour.) Also, the images would occupy about 500 petabytes of data, and getting those data would require 20 electron microscopes running in parallel for about 5 years, continuously. They estimated the whole project would take about 17 years of work. This is, understandably, insane. But now it seems like there's an actual path toward mapping the full mouse brain in about five years for ~$100M dollars. There have been three major breakthroughs in the last year or so: 1/ Expansion microscopy, first developed in 2015, showed that it's possible to "enlarge" the brain by about 5x using a swellable polymer. But an improved method increases this number to >20x expansion, meaning we can now expand brains and image neurons much more easily using cheap light microscopes, rather than expensive electron ones. 2/ E11 Bio (a nonprofit research org) developed protein barcodes that get delivered into brain tissue; each neuron gets a unique combination of barcodes. These cells are then stained with colorful antibodies, which stick to a matching protein barcode, causing each neuron to light up in a distinct color. This makes tracing neurons so much easier. 3/ Google Research released PATHFINDER this May, an AI-based neuron tracing tool that can proofread about 67,200 cubic microns of brain tissue per hour, with very high accuracy. It works on electron micrographs, but something similar could be presumably be developed for the E11 / colorful tag approach. This is an extremely exciting time for neuroscience. (C. elegans connectome below.)

Niko McCarty.

67,020 views • 8 months ago

Depth Any Video with Scalable Synthetic Data AI physicists and chemists continue to make strides in depth estimation from video. Check out this new paper featuring some impressive examples. See the thread for more details (unfortunately no code yet). Abstract: Video depth estimation has long been hindered by the scarcity of consistent and scalable ground truth data, leading to inconsistent and unreliable results. In this paper, we introduce Depth Any Video, a model that tackles the challenge through two key innovations. First, we develop a scalable synthetic data pipeline, capturing real-time video depth data from diverse game environments, yielding 40,000 video clips of 5-second duration, each with precise depth annotations. Second, we leverage the powerful priors of generative video diffusion models to handle real-world videos effectively, integrating advanced techniques such as rotary position encoding and flow matching to further enhance flexibility and efficiency. Unlike previous models, which are limited to fixed-length video sequences, our approach introduces a novel mixed-duration training strategy that handles videos of varying lengths and performs robustly across different frame rates 0 - even on single frames. At inference, we propose a depth interpolation method that enables our model to infer high-resolution video depth across sequences of up to 150 frames. Our model outperforms all previous generative depth models in terms of spatial accuracy and temporal consistency.

Depth Any Video with Scalable Synthetic Data AI physicists and chemists continue to make strides in depth estimation from video. Check out this new paper featuring some impressive examples. See the thread for more details (unfortunately no code yet). Abstract: Video depth estimation has long been hindered by the scarcity of consistent and scalable ground truth data, leading to inconsistent and unreliable results. In this paper, we introduce Depth Any Video, a model that tackles the challenge through two key innovations. First, we develop a scalable synthetic data pipeline, capturing real-time video depth data from diverse game environments, yielding 40,000 video clips of 5-second duration, each with precise depth annotations. Second, we leverage the powerful priors of generative video diffusion models to handle real-world videos effectively, integrating advanced techniques such as rotary position encoding and flow matching to further enhance flexibility and efficiency. Unlike previous models, which are limited to fixed-length video sequences, our approach introduces a novel mixed-duration training strategy that handles videos of varying lengths and performs robustly across different frame rates 0 - even on single frames. At inference, we propose a depth interpolation method that enables our model to infer high-resolution video depth across sequences of up to 150 frames. Our model outperforms all previous generative depth models in terms of spatial accuracy and temporal consistency.

MrNeRF

27,428 views • 1 year ago

Break-A-Scene: Extracting Multiple Concepts from a Single Image introduce the task of textual scene decomposition: given a single image of a scene that may contain several concepts, we aim to extract a distinct text token for each concept, enabling fine-grained control over the generated scenes. To this end, we propose augmenting the input image with masks that indicate the presence of target concepts. These masks can be provided by the user or generated automatically by a pre-trained segmentation model. We then present a novel two-phase customization process that optimizes a set of dedicated textual embeddings (handles), as well as the model weights, striking a delicate balance between accurately capturing the concepts and avoiding overfitting. We employ a masked diffusion loss to enable handles to generate their assigned concepts, complemented by a novel loss on cross-attention maps to prevent entanglement. We also introduce union-sampling, a training strategy aimed to improve the ability of combining multiple concepts in generated images. We use several automatic metrics to quantitatively compare our method against several baselines, and further affirm the results using a user study. Finally, we showcase several applications of our method paper page:

Break-A-Scene: Extracting Multiple Concepts from a Single Image introduce the task of textual scene decomposition: given a single image of a scene that may contain several concepts, we aim to extract a distinct text token for each concept, enabling fine-grained control over the generated scenes. To this end, we propose augmenting the input image with masks that indicate the presence of target concepts. These masks can be provided by the user or generated automatically by a pre-trained segmentation model. We then present a novel two-phase customization process that optimizes a set of dedicated textual embeddings (handles), as well as the model weights, striking a delicate balance between accurately capturing the concepts and avoiding overfitting. We employ a masked diffusion loss to enable handles to generate their assigned concepts, complemented by a novel loss on cross-attention maps to prevent entanglement. We also introduce union-sampling, a training strategy aimed to improve the ability of combining multiple concepts in generated images. We use several automatic metrics to quantitatively compare our method against several baselines, and further affirm the results using a user study. Finally, we showcase several applications of our method paper page:

AK

154,511 views • 3 years ago

The term "continual learning" has become overloaded if you see it as an ML problem. One classic thread is about memorization: regularization-based continual learning methods, such as EWC, MAS, and SI, estimate which parameters mattered for previous tasks and resist changing them too much. One modern thread is about adaptation: test-time training and inference-time learning methods, such as TTT, adapt part of the model on the incoming test stream before making predictions. These are sometimes discussed as separate threads. But in modern scalable architectures, I think they are better seen as complementary constraints: a model that learns quickly at test time also benefits from a mechanism for deciding what not to forget. In our #ECCV2026 paper, we study this in large-scale 4D reconstruction: how to build fast spatial memory that can adapt over long observation streams while reducing collapse and forgetting. Instead of using fully plastic test-time updates, we stabilize fast-weight adaptation with an elastic prior that balances adaptation and memory. Key ideas: - Elastic Test-Time Training: Fisher-weighted consolidation for fast-weight updates - EMA anchor weights that provide a moving reference for stability - Chunk-by-chunk inference for long 3D/4D observation streams We show that this scales across large 3D/4D pretraining settings, including both LRM-style and LVSM-style models, and improves reconstruction across benchmarks including Stereo4D, NVIDIA, and DL3DV-140. We release model checkpoints across different design choices: resolution, post-training curriculum, and whether the model uses an explicit 4DGS intermediate representation. - Homepage: - Paper: - Code: - Models: This work is co-led with Xueyang Yu, contributed by Haoyu Zhen Yuncong Yang, and advised by Michigan SLED Lab Chuang Gan.

The term "continual learning" has become overloaded if you see it as an ML problem. One classic thread is about memorization: regularization-based continual learning methods, such as EWC, MAS, and SI, estimate which parameters mattered for previous tasks and resist changing them too much. One modern thread is about adaptation: test-time training and inference-time learning methods, such as TTT, adapt part of the model on the incoming test stream before making predictions. These are sometimes discussed as separate threads. But in modern scalable architectures, I think they are better seen as complementary constraints: a model that learns quickly at test time also benefits from a mechanism for deciding what not to forget. In our #ECCV2026 paper, we study this in large-scale 4D reconstruction: how to build fast spatial memory that can adapt over long observation streams while reducing collapse and forgetting. Instead of using fully plastic test-time updates, we stabilize fast-weight adaptation with an elastic prior that balances adaptation and memory. Key ideas: - Elastic Test-Time Training: Fisher-weighted consolidation for fast-weight updates - EMA anchor weights that provide a moving reference for stability - Chunk-by-chunk inference for long 3D/4D observation streams We show that this scales across large 3D/4D pretraining settings, including both LRM-style and LVSM-style models, and improves reconstruction across benchmarks including Stereo4D, NVIDIA, and DL3DV-140. We release model checkpoints across different design choices: resolution, post-training curriculum, and whether the model uses an explicit 4DGS intermediate representation. - Homepage: - Paper: - Code: - Models: This work is co-led with Xueyang Yu, contributed by Haoyu Zhen Yuncong Yang, and advised by Michigan SLED Lab Chuang Gan.

Martin Ziqiao Ma

33,617 views • 1 month ago

Can robots learn without training❓ [𝗜𝘁'𝘀 𝗼𝗽𝗲𝗻 𝘀𝗼𝘂𝗿𝗰𝗲𝗱 ⬇ ] Teaching robots to do complex tasks WITHOUT spending hours training them. Sounds cool, right? That's exactly what DIAL-MPC does! The first training-free method for whole-body torque control using full-order dynamics: ✅ Instantly checks if a robot's moves are right or wrong ✅ Adapts quickly to new tasks without needing extra training ✅ Could work hand-in-hand with other robot learning methods Robots are getting smarter AND faster without the need for long training sessions. Website: Paper: Code: Saw this first Haoru Xue ✈️ CVPR 🙏

Can robots learn without training❓ [𝗜𝘁'𝘀 𝗼𝗽𝗲𝗻 𝘀𝗼𝘂𝗿𝗰𝗲𝗱 ⬇ ] Teaching robots to do complex tasks WITHOUT spending hours training them. Sounds cool, right? That's exactly what DIAL-MPC does! The first training-free method for whole-body torque control using full-order dynamics: ✅ Instantly checks if a robot's moves are right or wrong ✅ Adapts quickly to new tasks without needing extra training ✅ Could work hand-in-hand with other robot learning methods Robots are getting smarter AND faster without the need for long training sessions. Website: Paper: Code: Saw this first Haoru Xue ✈️ CVPR 🙏

Ilir Aliu

71,502 views • 1 year ago

Biomni Lab lets biologists collaborate with AI agents to finish complex tasks end-to-end. Here are 15 popular use cases, each link is a full replay so you can watch the agent work through every step: 1. Spatial transcriptomics analysis: map gene expression across tissue architecture from spatial transcriptomics data, with spatial clustering and neighborhood analysis. 2. Binder design: design de novo protein binders against a target structure using computational protein design tools. 3. Biomarker panel design: identify and optimize a multi-marker diagnostic or prognostic panel from omics data. 4. Clinical trial landscaping: search and summarize the trial landscape for a disease area, mapping phase, endpoints, and sponsor activity. 5. Survival analysis: pull clinical and expression data, fit Cox models, generate Kaplan-Meier curves, and identify prognostic markers. 6. scRNA-seq processing and annotation: from raw counts to UMAP clustering, marker gene detection, and automated cell type labeling. 7. Cell-cell communication: infer ligand-receptor interactions between cell types from single-cell data and map intercellular signaling networks. 8. Primer design for novel Cas13: analyze a putative Cas13 protein from a metagenomic screen—verify the ORF, identify HEPN domains, and design cloning primers with restriction sites and a FLAG 9. Proteomics differential expression: normalize mass spec data, run statistical tests, and visualize differentially abundant proteins. 10. Gene regulatory network inference: reconstruct transcription factor-target gene networks from expression data and identify key regulators. 11. Gene co-expression network analysis: build weighted co-expression networks, identify gene modules, and correlate them with phenotypic traits. 12. Microbiome analysis: process 16S/metagenomic sequencing data to profile microbial communities, diversity, and differential abundance. 13. Polygenic risk scores: compute and evaluate PRS from GWAS summary statistics against a target cohort. 14. Variant annotation: annotate genetic variants with functional predictions, allele frequencies, and clinical significance. 15. Fine-mapping: narrow GWAS loci to credible causal variants using statistical fine-mapping methods. Each of these would normally take days to weeks of scripting, debugging, and iteration. In Biomni Lab, the agent handles the full execution while you steer the science. Learn more:

Kexin Huang

27,635 views • 4 months ago

🚨 CHINESE SCIENTISTS JUST INVENTED 3D PRINTING THAT CREATES OBJECTS IN 0.6 SECONDS USING ONLY LIGHT. Researchers at Tsinghua University have developed a new method called DISH (Digital Incoherent Synthesis of Holographic light fields) that can print complex millimeter-scale objects almost instantly. Instead of slowly building layer by layer, the system fires thousands of precisely patterned light images from multiple angles into a still vat of liquid resin. Where the light overlaps, the resin instantly hardens into a solid 3D object. The entire process takes just 0.6 seconds. Why this matters: • It’s currently the fastest volumetric 3D printing method ever demonstrated • Achieves extremely fine detail features thinner than a human hair • The resin stays completely still, so there’s no vibration or distortion • It can work with watery (low-viscosity) resins, making it suitable for biological applications • The team has already printed complex structures like blood vessel-like tubes and even a tiny bust of a historical figure The deeper implication: Traditional 3D printing has always been limited by speed and the need to move either the print head or the resin. This approach removes both constraints by using light itself as the sculptor. Because it can print directly into still liquid (and potentially onto living tissue), it opens new possibilities in bioprinting, medical devices, and rapid manufacturing. If the technology can be scaled beyond millimeter sizes, it could fundamentally change how we think about making physical objects turning “print” from a slow process into something closer to instantaneous fabrication. We’re moving from “layer by layer” to “all at once.” How do you think instant volumetric 3D printing like this could change medicine, manufacturing, or everyday life if it becomes widely available? Follow for more frontier manufacturing and materials science breakthroughs.

🚨 CHINESE SCIENTISTS JUST INVENTED 3D PRINTING THAT CREATES OBJECTS IN 0.6 SECONDS USING ONLY LIGHT. Researchers at Tsinghua University have developed a new method called DISH (Digital Incoherent Synthesis of Holographic light fields) that can print complex millimeter-scale objects almost instantly. Instead of slowly building layer by layer, the system fires thousands of precisely patterned light images from multiple angles into a still vat of liquid resin. Where the light overlaps, the resin instantly hardens into a solid 3D object. The entire process takes just 0.6 seconds. Why this matters: • It’s currently the fastest volumetric 3D printing method ever demonstrated • Achieves extremely fine detail features thinner than a human hair • The resin stays completely still, so there’s no vibration or distortion • It can work with watery (low-viscosity) resins, making it suitable for biological applications • The team has already printed complex structures like blood vessel-like tubes and even a tiny bust of a historical figure The deeper implication: Traditional 3D printing has always been limited by speed and the need to move either the print head or the resin. This approach removes both constraints by using light itself as the sculptor. Because it can print directly into still liquid (and potentially onto living tissue), it opens new possibilities in bioprinting, medical devices, and rapid manufacturing. If the technology can be scaled beyond millimeter sizes, it could fundamentally change how we think about making physical objects turning “print” from a slow process into something closer to instantaneous fabrication. We’re moving from “layer by layer” to “all at once.” How do you think instant volumetric 3D printing like this could change medicine, manufacturing, or everyday life if it becomes widely available? Follow for more frontier manufacturing and materials science breakthroughs.

TheNewPhysics

347,458 views • 1 month ago

🚨 SCIENTISTS MAY HAVE FOUND A NEW WAY TO KILL DRUG-RESISTANT BACTERIA. And it doesn't rely on traditional antibiotics. Researchers repurposed a ruthenium-based anticancer drug and activated it using ultrasound deep inside infected tissue. Why this matters: • Antibiotic resistance is one of the fastest-growing threats to global health • Drug-resistant infections could kill more people than cancer by 2050 • Many antibiotics are losing effectiveness • Deep-tissue infections are difficult to target safely • Bacteria continue evolving resistance to conventional treatments The breakthrough: Scientists used a compound called TLD1433, originally developed for cancer therapy. By itself, the drug is largely inactive. But when exposed to focused ultrasound... it generates highly reactive oxygen molecules that attack bacterial DNA and destroy protective biofilms. Unlike antibiotics, the treatment doesn't target a single bacterial pathway. Instead, it creates widespread oxidative damage that bacteria struggle to evolve resistance against. The results were remarkable: • Outperformed conventional antibiotics in laboratory tests • Reduced survival of pneumonia-causing bacteria to just 14% • Broke down oxygen-starved bacterial biofilms • In animal studies, every treated mouse survived • Only 25% of untreated controls survived The deeper implication is enormous: For decades we've searched for new antibiotics. But the future may not be finding stronger drugs. It may be activating existing drugs only where they're needed. By combining chemistry with precisely targeted ultrasound... scientists could attack dangerous infections deep inside the body while minimizing damage to healthy tissue. The real question is: Could sound become one of medicine's most powerful weapons against antibiotic resistance? Follow for more frontier science and technology discoveries.

🚨 SCIENTISTS MAY HAVE FOUND A NEW WAY TO KILL DRUG-RESISTANT BACTERIA. And it doesn't rely on traditional antibiotics. Researchers repurposed a ruthenium-based anticancer drug and activated it using ultrasound deep inside infected tissue. Why this matters: • Antibiotic resistance is one of the fastest-growing threats to global health • Drug-resistant infections could kill more people than cancer by 2050 • Many antibiotics are losing effectiveness • Deep-tissue infections are difficult to target safely • Bacteria continue evolving resistance to conventional treatments The breakthrough: Scientists used a compound called TLD1433, originally developed for cancer therapy. By itself, the drug is largely inactive. But when exposed to focused ultrasound... it generates highly reactive oxygen molecules that attack bacterial DNA and destroy protective biofilms. Unlike antibiotics, the treatment doesn't target a single bacterial pathway. Instead, it creates widespread oxidative damage that bacteria struggle to evolve resistance against. The results were remarkable: • Outperformed conventional antibiotics in laboratory tests • Reduced survival of pneumonia-causing bacteria to just 14% • Broke down oxygen-starved bacterial biofilms • In animal studies, every treated mouse survived • Only 25% of untreated controls survived The deeper implication is enormous: For decades we've searched for new antibiotics. But the future may not be finding stronger drugs. It may be activating existing drugs only where they're needed. By combining chemistry with precisely targeted ultrasound... scientists could attack dangerous infections deep inside the body while minimizing damage to healthy tissue. The real question is: Could sound become one of medicine's most powerful weapons against antibiotic resistance? Follow for more frontier science and technology discoveries.

TheNewPhysics

10,651 views • 2 months ago

Introducing Kaleido💮 from AI at Meta — a universal generative neural rendering engine for photorealistic, unified object and scene view synthesis. Kaleido is built on a simple but powerful design philosophy: 3D perception is a form of visual common sense. Following this idea, we formulate rendering purely as a sequence-to-sequence generation problem, successfully unifying neural rendering with the architecture principles behind modern language and video models. Unlike traditional neural rendering methods, Kaleido learns 3D purely in a data-driven way, without explicit 3D representations or structures. It acquires spatial understanding directly through large-scale video pretraining, then multi-view 3D data finetuning, inspired by how LLMs acquire textual common sense from large corpora before specialising in domains like coding. Through extensive ablations, we progressively modernised the architecture design and training strategies and tackled key scaling challenges in sequence-to-sequence generative rendering, arriving at a design that’s simple, versatile, and scalable. Kaleido significantly outperforms prior generative models in few-view settings, and remarkably is the first zero-shot generative method matches InstantNGP-level rendering quality in multi-view settings. We view Kaleido also as an alternative step towards world modeling that flexibly spans a spectrum of “realities": with many views, it faithfully reconstructs grounded reality; with fewer views, it imagines plausible unseen details. 🔗 Explore more results and paper:

Introducing Kaleido💮 from AI at Meta — a universal generative neural rendering engine for photorealistic, unified object and scene view synthesis. Kaleido is built on a simple but powerful design philosophy: 3D perception is a form of visual common sense. Following this idea, we formulate rendering purely as a sequence-to-sequence generation problem, successfully unifying neural rendering with the architecture principles behind modern language and video models. Unlike traditional neural rendering methods, Kaleido learns 3D purely in a data-driven way, without explicit 3D representations or structures. It acquires spatial understanding directly through large-scale video pretraining, then multi-view 3D data finetuning, inspired by how LLMs acquire textual common sense from large corpora before specialising in domains like coding. Through extensive ablations, we progressively modernised the architecture design and training strategies and tackled key scaling challenges in sequence-to-sequence generative rendering, arriving at a design that’s simple, versatile, and scalable. Kaleido significantly outperforms prior generative models in few-view settings, and remarkably is the first zero-shot generative method matches InstantNGP-level rendering quality in multi-view settings. We view Kaleido also as an alternative step towards world modeling that flexibly spans a spectrum of “realities": with many views, it faithfully reconstructs grounded reality; with fewer views, it imagines plausible unseen details. 🔗 Explore more results and paper:

Shikun Liu

22,389 views • 10 months ago

Facial reconstruction of a likely 5,500-year-old man from Chiatura, Darkveti, Georgia The “Chiatura man” is well known in anthropological literature for his massive cranial dimensions. Without soft tissue, the skull, first described by Alexandre Javakhishvili, exhibits a massive cheek breadth of 153 mm, a very large cranial width of 156 mm, and a very long cranial length of 199 mm. Most notably, it stands out among contemporaneous Caucasian crania for its overall size and massiveness, showing a closer affinity to the Caucasionic type, which is otherwise thought to have first appeared in the Caucasus during the Middle Ages. The Caucasionic type, which now dominates the Greater Caucasus region, is characterized by very large cheek width and very large cranial breadth. This Caucasionic-like morphology led some anthropologists -- most notably Malkhaz Abdushelishvili -- to describe the Chiatura man as proto-Caucasionic, using him as evidence for a much earlier presence of Caucasionic morphology in the Caucasus. Other anthropologists, such as Ilya Gokhman, questioned this interpretation, expressing doubts about the proposed dating of the individual and suggesting that he may instead belong to a period much closer to modernity. The individual is attributed to the Kura-Araxes culture, whose languages are generally associated with Northeast Caucasian speakers. The Upper Imeretian variant of the Kura-Araxes culture (to which the Chiatura man belonged), however, differs from other Kura-Araxes groups. Genetically, it was likely influenced by local Darkveti populations, while culturally it is distinguished by the presence of kurgan burials and by rich metal assemblages, such as those found in the Sachkhere kurgans, setting it apart from most other Kura-Araxes communities. Middle Bronze Age individuals from Imereti, such as the sample from Sagvarjile, can be genetically modeled as a mixture of Proto-Kartvelian-like/ancient Svan-like profile and Kura-Araxes ancestry, likely reflecting mutual interaction and admixture between Imeretian Kura-Araxes populations and local Kartvelians. The Chiatura individual also exhibits artificial cranial deformation, a practice well documented within the Kura-Araxes cultural horizon. Such practice is also observed among Kura-Araxes populations in Armenia and Dagestan. This custom may have been transmitted by Kura-Araxes groups to the Catacomb culture. Crania from Chiatura dating to Late Antiquity also display artificial cranial deformation.

Ancestral Whispers

17,725 views • 6 months ago

A viral paper "Language Model Represents Space and Time" recently claims that LLMs learn "world models". As much as I like Max Tegmark's works, I disagree with their definition of world model. World model is a core concept in AI agent and decision making. It is our mental simulation of how the world works given interventions (or lack thereof). A world model captures causality and intuitive physics, telling the agent what is likely and what is impossible. It can and should be used for counterfactual reasoning, i.e. "what ifs": what would happen if I knock over a cup of water? Where would I have been if I had not taken that bus? Yann LeCun Yann LeCun says it well in his position paper ( I quote: "Using such world models, animals can learn new skills with very few trials. They can predict the consequences of their actions, they can reason, plan, explore, and imagine new solutions to problems. Importantly, they can also avoid making dangerous mistakes when facing an unknown situation." The first use of the term World Model in deep policy learning is attributed to hardmaru & Jürgen Schmidhuber: In their seminal paper, an agent masters shooting skills in the popular game Doom (demo below) by learning in imagination, using an internal world model as a "physics simulator". To put in a simple Python math formula, world model learns a function F(s[0:t-1], a) -> s[t:], which takes as input the observed past and current action, and outputs plausible future states. Now the definition of World Model in Tegmark's paper seems to be about predicting GPS coordinates and time eras. I see this as just a classification task with no causal learning and simulation going on. You cannot make meaningful interventions against that model, nor can you optimize any decision making in a closed feedback loop. As for the "space & time neurons", I think they are most similar to the "sentiment neuron" that OpenAI published in 2017: Predicting GPS is conceptually no different from predicting sentiment in my opinion. I don't think their experimental results are wrong - just that their conclusion is on shaky grounds. I welcome any debate! Paper link:

A viral paper "Language Model Represents Space and Time" recently claims that LLMs learn "world models". As much as I like Max Tegmark's works, I disagree with their definition of world model. World model is a core concept in AI agent and decision making. It is our mental simulation of how the world works given interventions (or lack thereof). A world model captures causality and intuitive physics, telling the agent what is likely and what is impossible. It can and should be used for counterfactual reasoning, i.e. "what ifs": what would happen if I knock over a cup of water? Where would I have been if I had not taken that bus? Yann LeCun Yann LeCun says it well in his position paper ( I quote: "Using such world models, animals can learn new skills with very few trials. They can predict the consequences of their actions, they can reason, plan, explore, and imagine new solutions to problems. Importantly, they can also avoid making dangerous mistakes when facing an unknown situation." The first use of the term World Model in deep policy learning is attributed to hardmaru & Jürgen Schmidhuber: In their seminal paper, an agent masters shooting skills in the popular game Doom (demo below) by learning in imagination, using an internal world model as a "physics simulator". To put in a simple Python math formula, world model learns a function F(s[0:t-1], a) -> s[t:], which takes as input the observed past and current action, and outputs plausible future states. Now the definition of World Model in Tegmark's paper seems to be about predicting GPS coordinates and time eras. I see this as just a classification task with no causal learning and simulation going on. You cannot make meaningful interventions against that model, nor can you optimize any decision making in a closed feedback loop. As for the "space & time neurons", I think they are most similar to the "sentiment neuron" that OpenAI published in 2017: Predicting GPS is conceptually no different from predicting sentiment in my opinion. I don't think their experimental results are wrong - just that their conclusion is on shaky grounds. I welcome any debate! Paper link:

Jim Fan

594,014 views • 2 years ago

What if a drone could follow another drone simply by listening to the sound of flight? After more than 2 years of work, I'm excited to share SonicFly, our new work on embodied passive aeroacoustic perception. SonicFly enables one UAV to estimate and follow another using only the leader’s naturally generated flight sound, without GPS sharing, inter-robot communication, active acoustic signaling, or external sensing infrastructure. What sounded simple turned out to be a very challenging robotics problem. We went from designing the microphone array and understanding rotor acoustics to learning relative state and closing the loop for real outdoor aerial pursuit. Paper, video, code: Kudos to our amazing students Yanbaihui (Evelyn) Liu, Ravi Prakash, Li-Yu (Patrick) Lo, and Nils Roede! 🚁🎧 I am also very grateful to our other lab members, in particular Tate Staples, Jiaxun Liu, and partner Ben Borger, who helped make this possible.

What if a drone could follow another drone simply by listening to the sound of flight? After more than 2 years of work, I'm excited to share SonicFly, our new work on embodied passive aeroacoustic perception. SonicFly enables one UAV to estimate and follow another using only the leader’s naturally generated flight sound, without GPS sharing, inter-robot communication, active acoustic signaling, or external sensing infrastructure. What sounded simple turned out to be a very challenging robotics problem. We went from designing the microphone array and understanding rotor acoustics to learning relative state and closing the loop for real outdoor aerial pursuit. Paper, video, code: Kudos to our amazing students Yanbaihui (Evelyn) Liu, Ravi Prakash, Li-Yu (Patrick) Lo, and Nils Roede! 🚁🎧 I am also very grateful to our other lab members, in particular Tate Staples, Jiaxun Liu, and partner Ben Borger, who helped make this possible.

Boyuan Chen

115,476 views • 1 day ago

Introducing ASAL: Automating the Search for Artificial Life with Foundation Models Artificial Life (ALife) research holds key insights that can transform and accelerate progress in AI. By speeding up ALife discovery with AI, we accelerate our understanding of emergence, evolution, and intelligence–core principles that can inspire the next generation of AI systems! We proudly collaborated with MIT, OpenAI, Swiss AI Lab IDSIA, and Ken Stanley on this exciting project. Full Paper (Website): Full Paper (arxiv): Code: In this work, we propose a new algorithm called Automated Search for Artificial Life (“ASAL”) to automate the discovery of artificial life using vision-language foundation models. Instead of tediously hand-designing every tiny rule of an Alife simulation, simply describe the space of simulations to search over, and ASAL will automatically discover the most interesting and open-ended artificial lifeforms! Because of the generality of foundation models, ASAL can discover new lifeforms across a diverse range of seminal ALife simulations, including Boids, Particle Life, Game of Life, Lenia, and Neural Cellular Automata. ASAL even discovered novel cellular automata rules that are more open-ended and expressive than the original Conway’s Game of Life. We believe this new paradigm may reignite ALife research by overcoming the bottleneck of manually designed simulations, thus advancing beyond the limits of human ingenuity.

Introducing ASAL: Automating the Search for Artificial Life with Foundation Models Artificial Life (ALife) research holds key insights that can transform and accelerate progress in AI. By speeding up ALife discovery with AI, we accelerate our understanding of emergence, evolution, and intelligence–core principles that can inspire the next generation of AI systems! We proudly collaborated with MIT, OpenAI, Swiss AI Lab IDSIA, and Ken Stanley on this exciting project. Full Paper (Website): Full Paper (arxiv): Code: In this work, we propose a new algorithm called Automated Search for Artificial Life (“ASAL”) to automate the discovery of artificial life using vision-language foundation models. Instead of tediously hand-designing every tiny rule of an Alife simulation, simply describe the space of simulations to search over, and ASAL will automatically discover the most interesting and open-ended artificial lifeforms! Because of the generality of foundation models, ASAL can discover new lifeforms across a diverse range of seminal ALife simulations, including Boids, Particle Life, Game of Life, Lenia, and Neural Cellular Automata. ASAL even discovered novel cellular automata rules that are more open-ended and expressive than the original Conway’s Game of Life. We believe this new paradigm may reignite ALife research by overcoming the bottleneck of manually designed simulations, thus advancing beyond the limits of human ingenuity.

Sakana AI

751,056 views • 1 year ago

[Most robots react. This one thinks a step ahead.] Ant Group's Robbyant just published LingBot-VA 2.0 — a video-action foundation model built from scratch for robot control, not fine-tuned from a video generator. The usual approach takes a video generator made for content creation and bolts a robot policy onto it. LingBot-VA 2.0 argues that's the wrong starting point, and pretrains the whole causal stack natively instead. What stands out: → Foresight Reasoning — the robot predicts the next action chunk while executing the current one, then overwrites the imagined frame with the real observation. Prediction and execution stop waiting on each other. → 927 ms → 142 ms per chunk, across four cumulative optimizations. That lifts asynchronous control from 35 Hz to 225 Hz — a 6.5× speedup. → One shared latent space. A semantic visual-action tokenizer puts world states and actions in the same coordinates, so unlabeled web video carries action-relevant signal. → Sparse MoE video stream — 128 experts, top-8 routing. Roughly 2.5B of ~15.3B parameters fire per token. → Few-shot by design — adapts from 10–15 demonstrations, and a human demo video can replace the text instruction entirely. Full breakdown: Paper: Project Page: Robbyant Ant Group

[Most robots react. This one thinks a step ahead.] Ant Group's Robbyant just published LingBot-VA 2.0 — a video-action foundation model built from scratch for robot control, not fine-tuned from a video generator. The usual approach takes a video generator made for content creation and bolts a robot policy onto it. LingBot-VA 2.0 argues that's the wrong starting point, and pretrains the whole causal stack natively instead. What stands out: → Foresight Reasoning — the robot predicts the next action chunk while executing the current one, then overwrites the imagined frame with the real observation. Prediction and execution stop waiting on each other. → 927 ms → 142 ms per chunk, across four cumulative optimizations. That lifts asynchronous control from 35 Hz to 225 Hz — a 6.5× speedup. → One shared latent space. A semantic visual-action tokenizer puts world states and actions in the same coordinates, so unlabeled web video carries action-relevant signal. → Sparse MoE video stream — 128 experts, top-8 routing. Roughly 2.5B of ~15.3B parameters fire per token. → Few-shot by design — adapts from 10–15 demonstrations, and a human demo video can replace the text instruction entirely. Full breakdown: Paper: Project Page: Robbyant Ant Group

Marktechpost AI

196,499 views • 1 month ago

STEVE-1: A Generative Model for Text-to-Behavior in Minecraft paper page: Constructing AI models that respond to text instructions is challenging, especially for sequential decision-making tasks. This work introduces an instruction-tuned Video Pretraining (VPT) model for Minecraft called STEVE-1, demonstrating that the unCLIP approach, utilized in DALL-E 2, is also effective for creating instruction-following sequential decision-making agents. STEVE-1 is trained in two steps: adapting the pretrained VPT model to follow commands in MineCLIP's latent space, then training a prior to predict latent codes from text. This allows us to finetune VPT through self-supervised behavioral cloning and hindsight relabeling, bypassing the need for costly human text annotations. By leveraging pretrained models like VPT and MineCLIP and employing best practices from text-conditioned image generation, STEVE-1 costs just $60 to train and can follow a wide range of short-horizon open-ended text and visual instructions in Minecraft. STEVE-1 sets a new bar for open-ended instruction following in Minecraft with low-level controls (mouse and keyboard) and raw pixel inputs, far outperforming previous baselines. We provide experimental evidence highlighting key factors for downstream performance, including pretraining, classifier-free guidance, and data scaling. All resources, including our model weights, training scripts, and evaluation tools are made available for further research.

STEVE-1: A Generative Model for Text-to-Behavior in Minecraft paper page: Constructing AI models that respond to text instructions is challenging, especially for sequential decision-making tasks. This work introduces an instruction-tuned Video Pretraining (VPT) model for Minecraft called STEVE-1, demonstrating that the unCLIP approach, utilized in DALL-E 2, is also effective for creating instruction-following sequential decision-making agents. STEVE-1 is trained in two steps: adapting the pretrained VPT model to follow commands in MineCLIP's latent space, then training a prior to predict latent codes from text. This allows us to finetune VPT through self-supervised behavioral cloning and hindsight relabeling, bypassing the need for costly human text annotations. By leveraging pretrained models like VPT and MineCLIP and employing best practices from text-conditioned image generation, STEVE-1 costs just $60 to train and can follow a wide range of short-horizon open-ended text and visual instructions in Minecraft. STEVE-1 sets a new bar for open-ended instruction following in Minecraft with low-level controls (mouse and keyboard) and raw pixel inputs, far outperforming previous baselines. We provide experimental evidence highlighting key factors for downstream performance, including pretraining, classifier-free guidance, and data scaling. All resources, including our model weights, training scripts, and evaluation tools are made available for further research.

AK

144,783 views • 3 years ago

🚀 Introducing EgoExo Forge - built on top of Rerun, Gradio, and Hugging Face hub (I’ll be in San Francisco July 21–29 — if you’re into robotics, egocentric AI, large-scale data collection, or just want to chat, DM me!) In my opinion, large-scale, diverse, and high-quality data is still the largest bottleneck for generalized robotics deployment. I believe that some version of imitation learning from human examples will be the most scalable + clean way to train humanoid robots 🤖 (similar to what Tesla did for Full Self Driving). Teleop is too expensive to collect a large enough dataset in a reasonable manner, so passive collection via egocentric (and in certain cases, exocentric) views feels like the right bet. Over the past few months, I've been trying to build out the scaffolding for this and using Rerun as my underlying infrastructure. Data being collected needs to be easily inspectable + time series and rerun provides the right tooling for this. My goal is to first build out a ground truth representative dataset from already existing open source data, generate some reasonable baselines, and then go out and collect my own data that adheres to the defined schema. 🔍 Starting with open-source datasets 1. EgoDex from Apple 2. HOCap from Nvidia and the University of Texas at Dallas 3. Assembly101 from Meta All these different datasets have different sensor configurations + annotations, so my goal with egoexo-forge is to have one consistent labeling scheme + data layout. I built a data pipeline that aligns all of the different datasets in one general schema assuming the COCO133 keypoint layout that allows for exo+ego, ego only, or exo only Since the scaffolding is already there, it becomes MUCH easier to add other datasets. So the next ones that I'll be including are HD-EPIC kitchens dataset, HOT3D, and finally my own personal iPhone + insta360 go collection method. Once I have a diverse variety of datasets, I'll double down on what I believe to be the key algorithms required to make useful data for imitation learning 📊 1. Camera Pose estimation via SLAM/SFM for ego perspective (and automatic calibration for exo) 2. Human pose estimation for both egocentric + exocentric views 3. Metric 3D reconstruction + object tracking I'll be setting up reasonable open-source baselines for each of these to validate that these datasets work, and then finally try to use the generated datasets for some imitation learning via the pi0-lerobot repo I've been working on. I plan on making a blog post + providing more info on all of this in the near future so stay tuned

🚀 Introducing EgoExo Forge - built on top of Rerun, Gradio, and Hugging Face hub (I’ll be in San Francisco July 21–29 — if you’re into robotics, egocentric AI, large-scale data collection, or just want to chat, DM me!) In my opinion, large-scale, diverse, and high-quality data is still the largest bottleneck for generalized robotics deployment. I believe that some version of imitation learning from human examples will be the most scalable + clean way to train humanoid robots 🤖 (similar to what Tesla did for Full Self Driving). Teleop is too expensive to collect a large enough dataset in a reasonable manner, so passive collection via egocentric (and in certain cases, exocentric) views feels like the right bet. Over the past few months, I've been trying to build out the scaffolding for this and using Rerun as my underlying infrastructure. Data being collected needs to be easily inspectable + time series and rerun provides the right tooling for this. My goal is to first build out a ground truth representative dataset from already existing open source data, generate some reasonable baselines, and then go out and collect my own data that adheres to the defined schema. 🔍 Starting with open-source datasets 1. EgoDex from Apple 2. HOCap from Nvidia and the University of Texas at Dallas 3. Assembly101 from Meta All these different datasets have different sensor configurations + annotations, so my goal with egoexo-forge is to have one consistent labeling scheme + data layout. I built a data pipeline that aligns all of the different datasets in one general schema assuming the COCO133 keypoint layout that allows for exo+ego, ego only, or exo only Since the scaffolding is already there, it becomes MUCH easier to add other datasets. So the next ones that I'll be including are HD-EPIC kitchens dataset, HOT3D, and finally my own personal iPhone + insta360 go collection method. Once I have a diverse variety of datasets, I'll double down on what I believe to be the key algorithms required to make useful data for imitation learning 📊 1. Camera Pose estimation via SLAM/SFM for ego perspective (and automatic calibration for exo) 2. Human pose estimation for both egocentric + exocentric views 3. Metric 3D reconstruction + object tracking I'll be setting up reasonable open-source baselines for each of these to validate that these datasets work, and then finally try to use the generated datasets for some imitation learning via the pi0-lerobot repo I've been working on. I plan on making a blog post + providing more info on all of this in the near future so stay tuned

Pablo Vela

32,085 views • 1 year ago

Seedance V2 This used to take people month of study and practice on Adobe Flash, now it can be done with Seedance V2, not perfect but this is the worst it will be. 15 seconds, stylized 2D hand-drawn animation, overhead battlefield on aged yellow lined notebook paper, clear blue horizontal ruled lines and a red left margin line always visible, fine paper grain, pencil marks, ink strokes, minimal classroom-material aesthetic at the start. The entire video must preserve the same paper world from start to finish. No live action, no 3D rendering, no realistic human faces, no modern objects, no narration, no subtitles. Core concept: A childish classroom doodle of an ancient war gradually transforms into a legendary illustrated battlefield, then collapses back into scribbles after the climax. The escalation must feel smooth, intentional, and visually magical, as if imagination is taking over the page. Army design: Two opposing ancient armies drawn first as simple colored stick figures, one faction in red, one faction in blue. Dense infantry blocks with spears and swords, cavalry units with long lances, banner carriers, archers. At first they are crude doodles with simple line limbs and circular heads. As the battle intensifies, they evolve step by step into more detailed inked warriors with clearer armor silhouettes, horses, weapons, helmets, capes, and expressive movement, but still remain inside a hand-drawn 2D illustrated style on paper. Visual progression and timing: 0-3 seconds: Wide top-down view of a large notebook-paper battlefield. Rough stick-figure armies face each other across the page. The drawing feels playful and simple at first. The camera slowly glides forward over the paper as both sides begin charging. Tiny horses gallop, infantry rushes, arrows are sketched into existence and start falling. Everything still looks like rough schoolbook doodles. 3-7 seconds: The first major collision. Spears thrust, swords swing, cavalry crashes into cavalry, formations break apart. With each impact, the art style upgrades. Simple stick limbs become stronger ink lines, bodies gain armor shapes, horses gain muscular form, banners gain flowing detail, shadows and dust marks appear. The battlefield becomes denser, faster, more dramatic. Red and blue strokes smear across the page with the force of combat. 7-11 seconds: The battle reaches full transformation. The once-crude doodles are now a glorious hand-illustrated ancient war scene, still clearly drawn on notebook paper but far more detailed and cinematic. The camera pushes into a central duel between two opposing generals on horseback. Their weapons clash with a powerful burst of ink lines and paper tremor. Around them, infantry and cavalry continue fighting in layered motion, arrows rain down, fallen soldiers scatter across the ruled lines. 11-15 seconds: At the peak of the duel, one final strike lands. A shockwave ripples through the page. The detailed warriors, horses, banners, and battle effects suddenly break apart into loose pencil scribbles, sketch fragments, and drifting paper-line debris. The great war rapidly collapses back into childish rough doodles, then into scattered marks and unfinished lines, as if the imagination has burned out. End on the overhead notebook page with the battlefield reduced to messy hand-drawn remnants. Animation and motion: Smooth fluid motion, strong timing, readable silhouettes, at least 24fps feel. The escalation from crude doodle to epic illustrated warfare must be gradual and continuous, not abrupt. Impacts should feel sharp and rhythmic. Keep all action legible from overhead. Maintain strong contrast between the innocent notebook-paper setting and the seriousness of the war. Atmosphere: Starts playful and curious, grows intense and heroic, peaks as a mythic battlefield, then ends with a strange quiet after the collapse. The whole piece should feel like a child’s imagination turning into an epic war vision on paper.

Seedance V2 This used to take people month of study and practice on Adobe Flash, now it can be done with Seedance V2, not perfect but this is the worst it will be. 15 seconds, stylized 2D hand-drawn animation, overhead battlefield on aged yellow lined notebook paper, clear blue horizontal ruled lines and a red left margin line always visible, fine paper grain, pencil marks, ink strokes, minimal classroom-material aesthetic at the start. The entire video must preserve the same paper world from start to finish. No live action, no 3D rendering, no realistic human faces, no modern objects, no narration, no subtitles. Core concept: A childish classroom doodle of an ancient war gradually transforms into a legendary illustrated battlefield, then collapses back into scribbles after the climax. The escalation must feel smooth, intentional, and visually magical, as if imagination is taking over the page. Army design: Two opposing ancient armies drawn first as simple colored stick figures, one faction in red, one faction in blue. Dense infantry blocks with spears and swords, cavalry units with long lances, banner carriers, archers. At first they are crude doodles with simple line limbs and circular heads. As the battle intensifies, they evolve step by step into more detailed inked warriors with clearer armor silhouettes, horses, weapons, helmets, capes, and expressive movement, but still remain inside a hand-drawn 2D illustrated style on paper. Visual progression and timing: 0-3 seconds: Wide top-down view of a large notebook-paper battlefield. Rough stick-figure armies face each other across the page. The drawing feels playful and simple at first. The camera slowly glides forward over the paper as both sides begin charging. Tiny horses gallop, infantry rushes, arrows are sketched into existence and start falling. Everything still looks like rough schoolbook doodles. 3-7 seconds: The first major collision. Spears thrust, swords swing, cavalry crashes into cavalry, formations break apart. With each impact, the art style upgrades. Simple stick limbs become stronger ink lines, bodies gain armor shapes, horses gain muscular form, banners gain flowing detail, shadows and dust marks appear. The battlefield becomes denser, faster, more dramatic. Red and blue strokes smear across the page with the force of combat. 7-11 seconds: The battle reaches full transformation. The once-crude doodles are now a glorious hand-illustrated ancient war scene, still clearly drawn on notebook paper but far more detailed and cinematic. The camera pushes into a central duel between two opposing generals on horseback. Their weapons clash with a powerful burst of ink lines and paper tremor. Around them, infantry and cavalry continue fighting in layered motion, arrows rain down, fallen soldiers scatter across the ruled lines. 11-15 seconds: At the peak of the duel, one final strike lands. A shockwave ripples through the page. The detailed warriors, horses, banners, and battle effects suddenly break apart into loose pencil scribbles, sketch fragments, and drifting paper-line debris. The great war rapidly collapses back into childish rough doodles, then into scattered marks and unfinished lines, as if the imagination has burned out. End on the overhead notebook page with the battlefield reduced to messy hand-drawn remnants. Animation and motion: Smooth fluid motion, strong timing, readable silhouettes, at least 24fps feel. The escalation from crude doodle to epic illustrated warfare must be gradual and continuous, not abrupt. Impacts should feel sharp and rhythmic. Keep all action legible from overhead. Maintain strong contrast between the innocent notebook-paper setting and the seriousness of the war. Atmosphere: Starts playful and curious, grows intense and heroic, peaks as a mythic battlefield, then ends with a strange quiet after the collapse. The whole piece should feel like a child’s imagination turning into an epic war vision on paper.

Emily

35,023 views • 4 months ago

THIS IS WHY THE SOUTHERN LEBANESE ARE ANGRY 🇱🇧🇮🇱 The Full 14 Articles Lebanon Signed Read Like a SURRENDER, Not a Treaty Al-Jadeed obtained the complete Trilateral Framework signed in Washington, the document no government released in public. The clauses, in plain language: 1) DISARM FIRST (Art. 4): Lebanon commits to the "complete and verified disarmament" of every non-state armed group on its soil, Hezbollah unnamed but unmistakable, before anything is given back. 2) WITHDRAWAL IS THE REWARD (Art. 2, 5): The IDF only "progressively redeploys" out of two pilot zones, governed by a Security Annex that DOESN'T EVEN EXIST YET, and Israel's "no territorial ambitions" pledge is conditioned entirely on a disarmament IT ALONE gets to certify as complete. 3) SOVEREIGNTY ON PAPER (Art. 6): Lebanon holds the "exclusive sovereign authority" over war and peace, a clause built to outlaw the resistance by name without naming it. 4) AID ON A LEASH (Art. 9, 11): U.S. money to the Lebanese army is "strictly conditioned on verifiable milestones" and oversight. Beirut also pledges to choke off all funds to non-state groups, then keep reconstruction money out of their hands entirely. 5) RECONSTRUCTION, SEPARATELY GATED (Art. 10): Washington will "rally partners" to rebuild Lebanon, but on a SEPARATE track from the conditioned aid, leverage held in reserve. 6) THE TELL (Art. 14): The whole framework closes with both governments expressing "deep appreciation for the vision and leadership of President Donald J. Trump."

THIS IS WHY THE SOUTHERN LEBANESE ARE ANGRY 🇱🇧🇮🇱 The Full 14 Articles Lebanon Signed Read Like a SURRENDER, Not a Treaty Al-Jadeed obtained the complete Trilateral Framework signed in Washington, the document no government released in public. The clauses, in plain language: 1) DISARM FIRST (Art. 4): Lebanon commits to the "complete and verified disarmament" of every non-state armed group on its soil, Hezbollah unnamed but unmistakable, before anything is given back. 2) WITHDRAWAL IS THE REWARD (Art. 2, 5): The IDF only "progressively redeploys" out of two pilot zones, governed by a Security Annex that DOESN'T EVEN EXIST YET, and Israel's "no territorial ambitions" pledge is conditioned entirely on a disarmament IT ALONE gets to certify as complete. 3) SOVEREIGNTY ON PAPER (Art. 6): Lebanon holds the "exclusive sovereign authority" over war and peace, a clause built to outlaw the resistance by name without naming it. 4) AID ON A LEASH (Art. 9, 11): U.S. money to the Lebanese army is "strictly conditioned on verifiable milestones" and oversight. Beirut also pledges to choke off all funds to non-state groups, then keep reconstruction money out of their hands entirely. 5) RECONSTRUCTION, SEPARATELY GATED (Art. 10): Washington will "rally partners" to rebuild Lebanon, but on a SEPARATE track from the conditioned aid, leverage held in reserve. 6) THE TELL (Art. 14): The whole framework closes with both governments expressing "deep appreciation for the vision and leadership of President Donald J. Trump."

Ryan Rozbiani

61,924 views • 1 month ago