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.

Marigold V2: Repurposes a pretrained diffusion transformer into single-step dense predictors: depth, see-through depth, surface normals, albedo, and other dense modalities. Works in consumer GPU. Apache-2.0 license

Arun Kurian

1,180 subscribers

40,493 views • 1 month ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

6 Comments

Sergio Caires29 days ago

Wow. 😲 👍

Lewis_Orton29 days ago

impressive, how's the temporal stability

Aaron M Harris29 days ago

Can it work to convert video frames?

Brian Shields29 days ago

@bilawalsidhu

Mellen Y. Pu1 month ago

amazing! are they OOD samples?

Ghoshan1 month ago

@grok what can it predict?

Related Videos

Sweet! Marigold V2 is in ComfyUI now. turns regular 2D images into ultra-detailed 3D depth maps. - individual hairs, animal fur, thin foliage - handles surface normals/material properties in a single step - super fast

Sweet! Marigold V2 is in ComfyUI now. turns regular 2D images into ultra-detailed 3D depth maps. - individual hairs, animal fur, thin foliage - handles surface normals/material properties in a single step - super fast

Wildminder

31,757 views • 1 month ago

Today, we are open-sourcing Hunyuan World 1.1 (WorldMirror), a universal feed-forward 3D reconstruction model. 🚀🚀🚀 While our previously released Hunyuan World 1.0 (open-sourced, lite version deployable on consumer GPUs) focused on generating 3D worlds from text or single-view images, Hunyuan World 1.1 significantly expands the input scope by unlocking video-to-3D and multi-view-to-3D world creation. Highlights: 🔹Any Input, Maximized Flexibility and Fidelity: Flexibly integrates diverse geometric priors (camera poses, intrinsics, depth maps) to resolve structural ambiguities and ensure geometrically consistent 3D outputs. 🔹Any Output, SOTA Results：This elegant architecture simultaneously generates multiple 3D representations: dense point clouds, multi-view depth maps, camera parameters, surface normals, and 3D Gaussian Splattings. 🔹Single-GPU & Fast Inference: As an all-in-one, feed-forward model, Hunyuan World 1.1 runs on a single GPU and delivers all 3D attributes in a single forward pass, within seconds. 🌐Project Page: 🔗Github： 🤗Hugging Face： ✨Demo: 📄Technical Report:

Today, we are open-sourcing Hunyuan World 1.1 (WorldMirror), a universal feed-forward 3D reconstruction model. 🚀🚀🚀 While our previously released Hunyuan World 1.0 (open-sourced, lite version deployable on consumer GPUs) focused on generating 3D worlds from text or single-view images, Hunyuan World 1.1 significantly expands the input scope by unlocking video-to-3D and multi-view-to-3D world creation. Highlights: 🔹Any Input, Maximized Flexibility and Fidelity: Flexibly integrates diverse geometric priors (camera poses, intrinsics, depth maps) to resolve structural ambiguities and ensure geometrically consistent 3D outputs. 🔹Any Output, SOTA Results：This elegant architecture simultaneously generates multiple 3D representations: dense point clouds, multi-view depth maps, camera parameters, surface normals, and 3D Gaussian Splattings. 🔹Single-GPU & Fast Inference: As an all-in-one, feed-forward model, Hunyuan World 1.1 runs on a single GPU and delivers all 3D attributes in a single forward pass, within seconds. 🌐Project Page: 🔗Github： 🤗Hugging Face： ✨Demo: 📄Technical Report:

Tencent Hy

168,848 views • 11 months ago

Depth-aware light injection in TypeGPU I got a 448x448 monocular depth model down to ~8 ms on my M4 Pro across ~250 dispatches, which is fast enough to use in realtime :D Since the inference is written directly in TypeGPU, I can just feed the depth buffer straight into the lighting pass. It never has to leave the GPU or go through any extra synchronization/interop step Inference, lighting and draw all go through the same command encoder.

Depth-aware light injection in TypeGPU I got a 448x448 monocular depth model down to ~8 ms on my M4 Pro across ~250 dispatches, which is fast enough to use in realtime :D Since the inference is written directly in TypeGPU, I can just feed the depth buffer straight into the lighting pass. It never has to leave the GPU or go through any extra synchronization/interop step Inference, lighting and draw all go through the same command encoder.

Konrad Reczko

10,107,190 views • 1 month ago

We're reintroducing and open-sourcing project "See-through". Given a single anime illustration, it automatically decomposes the character into fully-inpainted semantic layers with depth ordering. One image in, layered PSD out. (1/n) Repo:

We're reintroducing and open-sourcing project "See-through". Given a single anime illustration, it automatically decomposes the character into fully-inpainted semantic layers with depth ordering. One image in, layered PSD out. (1/n) Repo:

柔情猫娘

582,614 views • 6 months ago

I've been playing more with this depth map motion transfer workflow, and if you're trying to recreate really specific body movements from a reference video, this works surprisingly well. Here's the workflow: 1. Grab your reference video, then create your character and environment references. 2. Run the video through Depth Anything on fal to generate a depth map: 3. Drop that depth map into Seedance Omni as your motion reference along with your character and environment references. You don't even need the original video. I also tested this against using the original video as the motion reference in Seedance, and there's definitely a noticeable difference. One thing I found, though, is that this works much better for body movement than facial performance since the depth map can't really capture subtle facial motion. I'm running more tests to see if there's a good way to control both body movement and facial performance. If you have any workflows I should try, do share in the comments!

I've been playing more with this depth map motion transfer workflow, and if you're trying to recreate really specific body movements from a reference video, this works surprisingly well. Here's the workflow: 1. Grab your reference video, then create your character and environment references. 2. Run the video through Depth Anything on fal to generate a depth map: 3. Drop that depth map into Seedance Omni as your motion reference along with your character and environment references. You don't even need the original video. I also tested this against using the original video as the motion reference in Seedance, and there's definitely a noticeable difference. One thing I found, though, is that this works much better for body movement than facial performance since the depth map can't really capture subtle facial motion. I'm running more tests to see if there's a good way to control both body movement and facial performance. If you have any workflows I should try, do share in the comments!

Curious Refuge

16,709 views • 2 months ago

running a 405-billion-parameter model used to mean a $30,000 GPU rack. one developer made it run on a single 8GB card the kind that's been sitting in gaming laptops for years. it's called AirLLM, and the trick is almost stupidly simple: instead of loading the whole model, it keeps one layer on the GPU at a time and streams the rest through. / Llama 3.1 405B on 8GB / DeepSeek-V3, all 671B of it, on about 12GB / no quantization, no distillation, no pruning the full-precision model / works with Llama, Qwen, DeepSeek, Mistral, Gemma, and more it's Apache 2.0 and already past 22,600 stars. → bookmark it.

running a 405-billion-parameter model used to mean a $30,000 GPU rack. one developer made it run on a single 8GB card the kind that's been sitting in gaming laptops for years. it's called AirLLM, and the trick is almost stupidly simple: instead of loading the whole model, it keeps one layer on the GPU at a time and streams the rest through. / Llama 3.1 405B on 8GB / DeepSeek-V3, all 671B of it, on about 12GB / no quantization, no distillation, no pruning the full-precision model / works with Llama, Qwen, DeepSeek, Mistral, Gemma, and more it's Apache 2.0 and already past 22,600 stars. → bookmark it.

dunik

151,882 views • 2 months ago

New | Texas DPS K-9 Stark in action 2/3/26: Texas DPS tracking K-9 Stark and his handler tracked for over a mile through dense brush along the Rio Grande in Starr Co, assisting the U.S. Border Patrol. The collaboration resulted in the apprehension of 14 illegal immigrants who were attempting to evade capture by hiding in thick brush. In this video, you’ll see K-9 Stark’s determination and tracking capabilities as he works through some of the most challenging terrain along the border. Our DPS tracking K-9s continue to lead the way—helping secure the border and keeping Texas and the nation safe from public safety and national security threats. #OperationLoneStar 2.0

New | Texas DPS K-9 Stark in action 2/3/26: Texas DPS tracking K-9 Stark and his handler tracked for over a mile through dense brush along the Rio Grande in Starr Co, assisting the U.S. Border Patrol. The collaboration resulted in the apprehension of 14 illegal immigrants who were attempting to evade capture by hiding in thick brush. In this video, you’ll see K-9 Stark’s determination and tracking capabilities as he works through some of the most challenging terrain along the border. Our DPS tracking K-9s continue to lead the way—helping secure the border and keeping Texas and the nation safe from public safety and national security threats. #OperationLoneStar 2.0

Chris Olivarez

11,879 views • 8 months ago

Bach’s Goldberg Variations like you’ve rarely heard them before. 🎼 Originally written for keyboard, this monumental work is built on a single aria and 30 intricate variations — a masterpiece of structure, counterpoint, and architectural precision. In this performance, Thibaut Garcia and Antoine Morinière reimagine it for two classical guitars, completely reshaping how the music breathes. What was once a dense keyboard texture becomes an intimate dialogue between two voices — every line more transparent, every variation more personal. Filmed inside the Abbey of Solesmes, the natural acoustics add a sense of stillness and depth that feels almost meditative, revealing a new emotional side of Bach’s design. A familiar masterpiece, heard through a completely different lens. 🎻✨

Bach’s Goldberg Variations like you’ve rarely heard them before. 🎼 Originally written for keyboard, this monumental work is built on a single aria and 30 intricate variations — a masterpiece of structure, counterpoint, and architectural precision. In this performance, Thibaut Garcia and Antoine Morinière reimagine it for two classical guitars, completely reshaping how the music breathes. What was once a dense keyboard texture becomes an intimate dialogue between two voices — every line more transparent, every variation more personal. Filmed inside the Abbey of Solesmes, the natural acoustics add a sense of stillness and depth that feels almost meditative, revealing a new emotional side of Bach’s design. A familiar masterpiece, heard through a completely different lens. 🎻✨

🎶𝗖𝗹𝗮𝘀𝘀𝗶𝗰𝗮𝗹 𝗠𝗲𝗹𝗼𝗱𝗶𝗲𝘀 ✨

15,443 views • 6 months ago

Have you used quantization with an open source machine learning library, and wondered how quantization works? How can you preserve model accuracy as you compress from 32 bits to 16, 8, or even 2 bits? In our new short course, Quantization in Depth, taught by Hugging Face's Marc Sun and @younesbelkada, you'll learn to implement variants of linear quantization, such as asymmetric and symmetric modes, from scratch. You'll also quantize at different granularities (per-tensor, per-channel, per-group) to maintain performance. You’ll then construct a quantizer to compress any open source deep learning model’s dense layers to 8-bit precision. Finally, you’ll practice quantizing weights into 2 bits by packing four 2-bit weights into a single 8-bit integer. If you've ever run a large open source model on your laptop, you've likely benefited from someone's work in quantization. Come learn how this key technique works under the hood! Please sign up here:

Have you used quantization with an open source machine learning library, and wondered how quantization works? How can you preserve model accuracy as you compress from 32 bits to 16, 8, or even 2 bits? In our new short course, Quantization in Depth, taught by Hugging Face's Marc Sun and @younesbelkada, you'll learn to implement variants of linear quantization, such as asymmetric and symmetric modes, from scratch. You'll also quantize at different granularities (per-tensor, per-channel, per-group) to maintain performance. You’ll then construct a quantizer to compress any open source deep learning model’s dense layers to 8-bit precision. Finally, you’ll practice quantizing weights into 2 bits by packing four 2-bit weights into a single 8-bit integer. If you've ever run a large open source model on your laptop, you've likely benefited from someone's work in quantization. Come learn how this key technique works under the hood! Please sign up here:

Andrew Ng

198,671 views • 2 years ago

New short course on Pretraining LLMs! Developed with Upstage and taught by their CEO Sung Kim and CSO Lucy Park. While prompting or fine-tuning existing models works well for many general language tasks, pretraining is valuable for specialized domains or languages with limited representation in current models. This course walks you through the LLM pretraining pipeline: 1. Data preparation: Learn to source, clean, and prepare training data using HuggingFace. 2. Model architecture: Configure transformer networks, including modifying existing models. 3. Training: Set up and run training using open-source libraries. 4. Evaluation: Benchmark performance using popular evaluation strategies. As an example use case, you'll also compare the output of a base model with its fine-tuned and further pretrained variants, to see the impact of pretraining on a model's ability to write Python. The course also explores an innovative technique called depth up-scaling, which Upstage used to train their Solar model family, reducing pretraining compute costs by up to 70%. This technique works by first duplicating layers of a smaller pretrained model to form a larger model, and then further pretraining the result. Sign up here!

New short course on Pretraining LLMs! Developed with Upstage and taught by their CEO Sung Kim and CSO Lucy Park. While prompting or fine-tuning existing models works well for many general language tasks, pretraining is valuable for specialized domains or languages with limited representation in current models. This course walks you through the LLM pretraining pipeline: 1. Data preparation: Learn to source, clean, and prepare training data using HuggingFace. 2. Model architecture: Configure transformer networks, including modifying existing models. 3. Training: Set up and run training using open-source libraries. 4. Evaluation: Benchmark performance using popular evaluation strategies. As an example use case, you'll also compare the output of a base model with its fine-tuned and further pretrained variants, to see the impact of pretraining on a model's ability to write Python. The course also explores an innovative technique called depth up-scaling, which Upstage used to train their Solar model family, reducing pretraining compute costs by up to 70%. This technique works by first duplicating layers of a smaller pretrained model to form a larger model, and then further pretraining the result. Sign up here!

Andrew Ng

85,781 views • 2 years ago

Wow. Recreating the Shawshank Redemption prison in 3D from a single video, in real time (!) Just read the MASt3R-SLAM paper and it's pretty neat. These folks basically built a real-time dense SLAM system on top of MASt3R, which is a transformer-based neural network that can do 3d reconstruction and localization from uncalibrated image pairs. The cool part is they don't need a fixed camera model -- it just works with arbitrary cameras -- think different focal lengths, sensor sizes, even handling zooming in video (FMV drone video anyone?!). If you've done photogrammetry or played with NeRFs you know that is a HUGE deal. They've solved some tricky problems like efficient point matching and tracking, plus they've figured out how to fuse point clouds and handle loop closures in real-time. Their system runs at about 15 FPS on a 4090 and produces both camera poses and dense geometry. When they know the camera calibration, they get SOTA results across several benchmarks, but even without calibration, they still perform well. What's interesting is the approach -- most recent SLAM work has built on DROID-SLAM's architecture, but these folks went a different direction by leveraging a strong 3D reconstruction prior. Seems to give them more coherent geometry, which makes sense since that's what MASt3R was designed for. For anyone who cares about monocular SLAM and 3D reconstruction, this feels like a significant step toward plug-and-play dense SLAM without calibration headaches -- perfect for drones, robots, AR/VR -- the works!

Wow. Recreating the Shawshank Redemption prison in 3D from a single video, in real time (!) Just read the MASt3R-SLAM paper and it's pretty neat. These folks basically built a real-time dense SLAM system on top of MASt3R, which is a transformer-based neural network that can do 3d reconstruction and localization from uncalibrated image pairs. The cool part is they don't need a fixed camera model -- it just works with arbitrary cameras -- think different focal lengths, sensor sizes, even handling zooming in video (FMV drone video anyone?!). If you've done photogrammetry or played with NeRFs you know that is a HUGE deal. They've solved some tricky problems like efficient point matching and tracking, plus they've figured out how to fuse point clouds and handle loop closures in real-time. Their system runs at about 15 FPS on a 4090 and produces both camera poses and dense geometry. When they know the camera calibration, they get SOTA results across several benchmarks, but even without calibration, they still perform well. What's interesting is the approach -- most recent SLAM work has built on DROID-SLAM's architecture, but these folks went a different direction by leveraging a strong 3D reconstruction prior. Seems to give them more coherent geometry, which makes sense since that's what MASt3R was designed for. For anyone who cares about monocular SLAM and 3D reconstruction, this feels like a significant step toward plug-and-play dense SLAM without calibration headaches -- perfect for drones, robots, AR/VR -- the works!

Bilawal Sidhu

704,409 views • 1 year ago

🚨 BREAKING: A research lab just released a 15B model that generates multilingual talking human videos with synced audio, beats every competitor in human evaluation, and runs in 38 seconds on one GPU. It's called daVinci-MagiHuman. The key insight is that every other model in this category stacks cross-attention, multi-stream pipelines, and separate conditioning branches to handle video and audio together. This one throws all of that out and uses a single unified self-attention stream across all modalities. Super-resolution happens in latent space rather than pixel space so there's no extra VAE decode-encode round trip. The turbo VAE decoder cuts decoding overhead even further. The distilled version runs in 8 steps with no CFG at all. Visual quality, text alignment, and word error rate all beat Ovi 1.1 and LTX 2.3 on the benchmark table. 100% Opensource. Apache 2.0. Repo and research paper links are in the comments.

🚨 BREAKING: A research lab just released a 15B model that generates multilingual talking human videos with synced audio, beats every competitor in human evaluation, and runs in 38 seconds on one GPU. It's called daVinci-MagiHuman. The key insight is that every other model in this category stacks cross-attention, multi-stream pipelines, and separate conditioning branches to handle video and audio together. This one throws all of that out and uses a single unified self-attention stream across all modalities. Super-resolution happens in latent space rather than pixel space so there's no extra VAE decode-encode round trip. The turbo VAE decoder cuts decoding overhead even further. The distilled version runs in 8 steps with no CFG at all. Visual quality, text alignment, and word error rate all beat Ovi 1.1 and LTX 2.3 on the benchmark table. 100% Opensource. Apache 2.0. Repo and research paper links are in the comments.

Ihtesham Ali

45,621 views • 6 months ago

Liquid Chrome 3D Icons / Imagine Image 2.0 Upload your image and use this prompt: Transform the uploaded logo into a premium 3D render, preserve the original logo shape, proportions, silhouette, structure and recognizable design, completely disregard the original logo colors, render the entire logo in a unified neutral liquid chrome material, mirror-polished metallic surface, flowing molten-metal appearance, smooth reflective contours, subtle fluid deformation, fine hairline scratches, tiny surface abrasions, scattered micro-scuffs, subtle signs of wear, slightly imperfect polished metal, realistic metal reflections, deep metallic depth, sharp specular highlights, polished 3D rendering, dramatic studio lighting, controlled reflections, subtle ambient occlusion, monochromatic metallic appearance, clean high-contrast presentation, centered composition, isolated single logo, solid pure black background, premium product-render aesthetic, ultra high resolution

Dogan Ural

20,991 views • 1 month ago

been playing with hermes agent paired with qwen 3.5 dense 27B on my single 3090 since last night. there is something about this harness that caught me and i think i know what it is. i've now run five qwen configs on consumer hardware: 35B MoE (3B active) -- 112 tok/s flat across 262K context, 1x 3090 27B dense -- 35 tok/s, zero degradation across the same range, 1x 3090 qwopus 27B (opus distilled) -- 35.7 tok/s, same architecture, different brain 80B coder -- 46 tok/s on 2x 3090s, oneshotted a 564 line particle sim 80B coder -- 1.3 tok/s on 1x 3090, bleeding through RAM because it didn't fit but it still ran with same benchmarks. same prompts. same quant where possible. every config is documented. i know these models. and hermes agent is the first harness that feels like it respects that work. tool calls show inline with execution time. nvidia-smi 0.2s. write_file 0.7s. you see exactly what the agent is doing and how long each step takes. no mystery. no black box. no tool call failures so far and i've been pushing it. most agent frameworks feel like you're watching a spinner and hoping. hermes shows the work. that transparency changes how you trust the output. once you use it you see the UX decisions are not accidental. Teknium 🪽 and the nous team built this like engineers who actually use their own tools. 80 skills. 29 tools. persistent memory. context compression. runs clean on a single consumer GPU.

been playing with hermes agent paired with qwen 3.5 dense 27B on my single 3090 since last night. there is something about this harness that caught me and i think i know what it is. i've now run five qwen configs on consumer hardware: 35B MoE (3B active) -- 112 tok/s flat across 262K context, 1x 3090 27B dense -- 35 tok/s, zero degradation across the same range, 1x 3090 qwopus 27B (opus distilled) -- 35.7 tok/s, same architecture, different brain 80B coder -- 46 tok/s on 2x 3090s, oneshotted a 564 line particle sim 80B coder -- 1.3 tok/s on 1x 3090, bleeding through RAM because it didn't fit but it still ran with same benchmarks. same prompts. same quant where possible. every config is documented. i know these models. and hermes agent is the first harness that feels like it respects that work. tool calls show inline with execution time. nvidia-smi 0.2s. write_file 0.7s. you see exactly what the agent is doing and how long each step takes. no mystery. no black box. no tool call failures so far and i've been pushing it. most agent frameworks feel like you're watching a spinner and hoping. hermes shows the work. that transparency changes how you trust the output. once you use it you see the UX decisions are not accidental. Teknium 🪽 and the nous team built this like engineers who actually use their own tools. 80 skills. 29 tools. persistent memory. context compression. runs clean on a single consumer GPU.

Sudo su

100,259 views • 7 months ago

no prompt engineering, no agentic harness, no tool calls. just me being lazy in llama.cpp's web ui and gemma 4 31b dense taking the task seriously. i typed "create gpu marketplace cards with hardware specs and prices per hour" and the model went and coded this ui, one shot, navy bg, glassmorphism cards, neon accent buttons, realistic pricing tiers per architecture. it even wrote a "why this looks premium" explanation under the code. for context this is a q4 quant of google's 31b dense thinking model, running on a rtx 5090 mobile 24gb in the rog scar 18 at around 15 tok/s sustained, same vram tier as a 3090 or 4090 desktop, so whatever you see here translates directly to your card at home. the whole interaction was me not trying and the model reasoning harder than the prompt deserved. that tells me more about where local ai is at in april 2026 than any leaderboard score. next test drops gemma 4 into hermes agent, autonomous tool calling, multi step reasoning, real agentic loop instead of a chat window. let's see what the same model does when it gets the right environment. more experiments coming anon. octopus invaders queued. same hardware, different tasks, all published here on x and all translatable to your 24gb card. for now the video below shows it coding live, gpu going brrr.

no prompt engineering, no agentic harness, no tool calls. just me being lazy in llama.cpp's web ui and gemma 4 31b dense taking the task seriously. i typed "create gpu marketplace cards with hardware specs and prices per hour" and the model went and coded this ui, one shot, navy bg, glassmorphism cards, neon accent buttons, realistic pricing tiers per architecture. it even wrote a "why this looks premium" explanation under the code. for context this is a q4 quant of google's 31b dense thinking model, running on a rtx 5090 mobile 24gb in the rog scar 18 at around 15 tok/s sustained, same vram tier as a 3090 or 4090 desktop, so whatever you see here translates directly to your card at home. the whole interaction was me not trying and the model reasoning harder than the prompt deserved. that tells me more about where local ai is at in april 2026 than any leaderboard score. next test drops gemma 4 into hermes agent, autonomous tool calling, multi step reasoning, real agentic loop instead of a chat window. let's see what the same model does when it gets the right environment. more experiments coming anon. octopus invaders queued. same hardware, different tasks, all published here on x and all translatable to your 24gb card. for now the video below shows it coding live, gpu going brrr.

Sudo su

28,939 views • 5 months ago

first impressions of qwen 3.5 27B dense on a single RTX 3090. 35 tok/s. from 4K all the way to 300K+ context. no speed drop. hermes 4.3 started at 35 and degraded to 15 as context filled. qwen dense holds. MoE held 112 flat. 3x faster but only 3B of 35B active per token. architecture tradeoff. Q4_K_M on 16.7GB. native context 262K. pushed past training limit to 376K before VRAM ceiling on 24GB. tried q8 KV cache at 262K, speed collapsed to 11 tok/s. q4_0 KV is the sweet spot. flash attention mandatory. built in reasoning mode. the model thinks step by step before it answers. full chain of thought surviving Q4 quant. 1,799+ token thinking chains with self correction loops. on a single consumer GPU. gave it one prompt: "build a realtime particle galaxy simulation in one HTML file." 3,340 tokens. 95 seconds. one shot. ran on first load. full reasoning and coding in the video below. optimal config if you want to skip the hours of testing: llama-server -ngl 99 -c 262144 -fa on --cache-type-k q4_0 --cache-type-v q4_0 this is just the warmup. octopus invaders is next: 10 files, 3,400+ lines, zero steering. the prompt hermes quit at 22%. already more impressed than expected. full results coming soon.

first impressions of qwen 3.5 27B dense on a single RTX 3090. 35 tok/s. from 4K all the way to 300K+ context. no speed drop. hermes 4.3 started at 35 and degraded to 15 as context filled. qwen dense holds. MoE held 112 flat. 3x faster but only 3B of 35B active per token. architecture tradeoff. Q4_K_M on 16.7GB. native context 262K. pushed past training limit to 376K before VRAM ceiling on 24GB. tried q8 KV cache at 262K, speed collapsed to 11 tok/s. q4_0 KV is the sweet spot. flash attention mandatory. built in reasoning mode. the model thinks step by step before it answers. full chain of thought surviving Q4 quant. 1,799+ token thinking chains with self correction loops. on a single consumer GPU. gave it one prompt: "build a realtime particle galaxy simulation in one HTML file." 3,340 tokens. 95 seconds. one shot. ran on first load. full reasoning and coding in the video below. optimal config if you want to skip the hours of testing: llama-server -ngl 99 -c 262144 -fa on --cache-type-k q4_0 --cache-type-v q4_0 this is just the warmup. octopus invaders is next: 10 files, 3,400+ lines, zero steering. the prompt hermes quit at 22%. already more impressed than expected. full results coming soon.

Sudo su

122,051 views • 7 months ago

📖THE SEEDANCE 2.0 PROMPT FORMAT THAT NOBODY IS USING BUT EVERYONE SHOULD Most people use Seedance 2.0 wrong. They type a normal prompt and wonder why the result looks generic. The difference between these two visuals isn't a better idea — it's a better prompt format. Normal prompt gives you a good result. JSON prompt gives you a cinematic one. Here's why it works: A normal prompt describes what you want. A JSON prompt tells Seedance exactly how to build the scene — camera movement, lighting angle, atmosphere, texture, depth, timing. You're not asking for a video. You're engineering one. The crystal dolphin jumping above an ancient courtyard with floating jellyfish — same concept, two completely different outputs. The JSON version has golden hour, dramatic clouds, volumetric light, and a sense of scale the normal prompt never reached. The glowing lotus fountain — same story. Normal prompt gives you a clean render. JSON gives you a full cinematic frame with layered depth, warm lantern light, and a composition that looks like it belongs in a feature film. The workflow that produces this: • Step 1 — Write your scene idea in plain language first •Step 2 — Ask Claude to convert it into a structured JSON prompt with fields for: subject, environment, camera, lighting, mood, motion, color grade • Step 3 — Paste the JSON into Seedance 2.0 • Step 4 — Take the best clip into Premiere Pro for final color correction and sound design The tools are free or cheap. The knowledge of how to prompt is what separates the results. This is the gap most creators never close. 📥 BACK TOMORROW. There will be new life hacks 🔖 The full guide is pinned below. Save it before you need it.

📖THE SEEDANCE 2.0 PROMPT FORMAT THAT NOBODY IS USING BUT EVERYONE SHOULD Most people use Seedance 2.0 wrong. They type a normal prompt and wonder why the result looks generic. The difference between these two visuals isn't a better idea — it's a better prompt format. Normal prompt gives you a good result. JSON prompt gives you a cinematic one. Here's why it works: A normal prompt describes what you want. A JSON prompt tells Seedance exactly how to build the scene — camera movement, lighting angle, atmosphere, texture, depth, timing. You're not asking for a video. You're engineering one. The crystal dolphin jumping above an ancient courtyard with floating jellyfish — same concept, two completely different outputs. The JSON version has golden hour, dramatic clouds, volumetric light, and a sense of scale the normal prompt never reached. The glowing lotus fountain — same story. Normal prompt gives you a clean render. JSON gives you a full cinematic frame with layered depth, warm lantern light, and a composition that looks like it belongs in a feature film. The workflow that produces this: • Step 1 — Write your scene idea in plain language first •Step 2 — Ask Claude to convert it into a structured JSON prompt with fields for: subject, environment, camera, lighting, mood, motion, color grade • Step 3 — Paste the JSON into Seedance 2.0 • Step 4 — Take the best clip into Premiere Pro for final color correction and sound design The tools are free or cheap. The knowledge of how to prompt is what separates the results. This is the gap most creators never close. 📥 BACK TOMORROW. There will be new life hacks 🔖 The full guide is pinned below. Save it before you need it.

Zentrix⌚️

137,112 views • 3 months ago

🌷 The Egg Carton Trick for Planting Bulbs—Fast & Perfectly Spaced Forget digging dozens of tiny holes. This simple, low-cost gardening trick lets you plant many bulbs at once, evenly spaced, with almost no measuring and far less effort. How it works: Step 1 Dig a single wide trench to the correct depth (roughly three times the height of the bulb). Level the bottom so everything sits evenly. Step 2 Place cardboard egg cartons directly on the soil, side by side. Drop one bulb into each cup, pointed tip facing up. Step 3 Cover everything with the soil you removed earlier—about 4–6 inches above the bulbs. Water thoroughly so the soil settles and the cartons begin breaking down. Step 4 Over time, the cardboard naturally decomposes. When spring arrives, you’re left with a dense, evenly spaced blanket of flowers. ⚠️ Important material reminder: Use only paper or cardboard egg cartons Never use plastic or Styrofoam cartons For faster root growth, cut a small slit in the bottom of each egg cup 🌼 Bulbs that work best: Ideal size: tulips, daffodils, crocus, grape hyacinths (muscari), scilla Not recommended: very large bulbs like giant alliums 🍂 When to plant: Best planted in fall, once soil temperatures drop below 60°F (15°C) Water immediately after planting to speed up decomposition and root contact

🌷 The Egg Carton Trick for Planting Bulbs—Fast & Perfectly Spaced Forget digging dozens of tiny holes. This simple, low-cost gardening trick lets you plant many bulbs at once, evenly spaced, with almost no measuring and far less effort. How it works: Step 1 Dig a single wide trench to the correct depth (roughly three times the height of the bulb). Level the bottom so everything sits evenly. Step 2 Place cardboard egg cartons directly on the soil, side by side. Drop one bulb into each cup, pointed tip facing up. Step 3 Cover everything with the soil you removed earlier—about 4–6 inches above the bulbs. Water thoroughly so the soil settles and the cartons begin breaking down. Step 4 Over time, the cardboard naturally decomposes. When spring arrives, you’re left with a dense, evenly spaced blanket of flowers. ⚠️ Important material reminder: Use only paper or cardboard egg cartons Never use plastic or Styrofoam cartons For faster root growth, cut a small slit in the bottom of each egg cup 🌼 Bulbs that work best: Ideal size: tulips, daffodils, crocus, grape hyacinths (muscari), scilla Not recommended: very large bulbs like giant alliums 🍂 When to plant: Best planted in fall, once soil temperatures drop below 60°F (15°C) Water immediately after planting to speed up decomposition and root contact

Plants and gardening

301,733 views • 7 months ago

AI VIDEOS HAVE CRACKED CONVERSATIONAL HAND GESTURES AND LOW ANGLE ANATOMY A girl with sleek auburn hair talks candidly into a low angle camera framed in a white apartment doorway, wearing a lavender plunge sports bra, high cut grey lounge shorts, and a layered gold heart pendant, gesturing naturally with open hands as she speaks. Nothing about the articulation or physical depth feels synthetic. Until recently, conversational hand gestures from an upward perspective were an instant failure point for generative video. When hands move rapidly in front of the torso, gesturing, pausing, and turning palms toward the lens, diffusion models struggle with spatial depth: fingers usually warp, nails blur into skin, and the hands frequently clip right through the chest fabric. Add a plunging neckline and tight elastic waistbands that must expand and contract with natural breathing, and earlier pipelines would produce severe texture smearing. Now, every gesture respects physical clearance and perspective. The fingers maintain sharp anatomical separation, the gold pendant stays anchored to the sternum without drifting, and the high-cut jersey shorts hold their structural tension over shifting hips. The real breakthrough isn’t hyper stylized cinematic shots, it’s that AI can now render an unscripted, 5 second casual vlogging story without a single glitch giving away the prompt.

AI VIDEOS HAVE CRACKED CONVERSATIONAL HAND GESTURES AND LOW ANGLE ANATOMY A girl with sleek auburn hair talks candidly into a low angle camera framed in a white apartment doorway, wearing a lavender plunge sports bra, high cut grey lounge shorts, and a layered gold heart pendant, gesturing naturally with open hands as she speaks. Nothing about the articulation or physical depth feels synthetic. Until recently, conversational hand gestures from an upward perspective were an instant failure point for generative video. When hands move rapidly in front of the torso, gesturing, pausing, and turning palms toward the lens, diffusion models struggle with spatial depth: fingers usually warp, nails blur into skin, and the hands frequently clip right through the chest fabric. Add a plunging neckline and tight elastic waistbands that must expand and contract with natural breathing, and earlier pipelines would produce severe texture smearing. Now, every gesture respects physical clearance and perspective. The fingers maintain sharp anatomical separation, the gold pendant stays anchored to the sternum without drifting, and the high-cut jersey shorts hold their structural tension over shifting hips. The real breakthrough isn’t hyper stylized cinematic shots, it’s that AI can now render an unscripted, 5 second casual vlogging story without a single glitch giving away the prompt.

aitifa uncut

60,196 views • 26 days ago

Today, we’re launching Gemma 4, our most intelligent open models to date. Built with the same breakthrough technology as Gemini 3, Gemma 4 brings advanced reasoning to your personal hardware and devices. Here’s what Gemma 4 unlocks for developers: — Intelligence-per-parameter: Our 31B (Dense) and 26B (MoE) models deliver state-of-the-art performance for their size, outcompeting models 20x their size on Arena.ai — Commercial flexibility: Released under a permissive Apache 2.0 license for complete developer flexibility and digital sovereignty — Agentic workflows: Native support for function-calling and structured JSON output allows you to build reliable, autonomous agents — Multimodal edge AI: The E2B and E4B models bring native vision, audio, and low latency to mobile and IoT devices — Long-context reasoning: Up to 256K context windows allow you to process entire repositories or large documents in a single prompt Whether you're building global applications in 140+ languages or local-first AI code assistants, Gemma 4 is built to be your foundation. Explore in Google AI Studio or download the weights on Hugging Face, Kaggle, and ollama.

Today, we’re launching Gemma 4, our most intelligent open models to date. Built with the same breakthrough technology as Gemini 3, Gemma 4 brings advanced reasoning to your personal hardware and devices. Here’s what Gemma 4 unlocks for developers: — Intelligence-per-parameter: Our 31B (Dense) and 26B (MoE) models deliver state-of-the-art performance for their size, outcompeting models 20x their size on Arena.ai — Commercial flexibility: Released under a permissive Apache 2.0 license for complete developer flexibility and digital sovereignty — Agentic workflows: Native support for function-calling and structured JSON output allows you to build reliable, autonomous agents — Multimodal edge AI: The E2B and E4B models bring native vision, audio, and low latency to mobile and IoT devices — Long-context reasoning: Up to 256K context windows allow you to process entire repositories or large documents in a single prompt Whether you're building global applications in 140+ languages or local-first AI code assistants, Gemma 4 is built to be your foundation. Explore in Google AI Studio or download the weights on Hugging Face, Kaggle, and ollama.

Google AI

1,671,077 views • 6 months ago