Загрузка видео...

Не удалось загрузить видео

Возникла проблема при загрузке этого видео. Это может быть связано с временными проблемами сети или видео может быть недоступно.

На главную

Adaptive Resolution for AO: splits the screen into 2x2 tiles and traces at 0.25🟩, 0.5🟨, or 1.0🟥resolution based on geometry uniformity within a tile / disocclusion. Then uses 2-frame temporal upscaling to approach near full-res quality. #gamedev #indiedev #unity3d #raytracing

Pavel

1,163 subscribers

11,534 просмотров • 8 месяцев назад •via X (Twitter)

Игры #gamedev #indiedev #unity3d #raytracing

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

There are two things worth pointing out about upscaling in #AssassinCreed Black Flag Resynced. 1. Incorrect resolution scaling with Balanced and Performance modes: Using Balanced or Performance with any upscaler causes the game's Resolution Scale (output resolution) to drop, which also lowers the internal render resolution. As a result, these modes apply incorrect scaling and the image quality is noticeably worse than it should be. For now, the workaround is to set the Resolution Scale to 100% and manually adjust the Max Dynamic Resolution (internal resolution) to match the expected scaling for Balanced or Performance mode. 2. The game defaults to the older DLSS 3 CNN model: By default, the game uses the older DLSS 3 Preset E. While it's still a fine option, it doesn't match the image quality of the DLSS 4 Transformer model (Preset K) or the newer DLSS 4.5 (Preset L and M). So I highly recommend forcing at least Preset K through the NVIDIA App (or another tool). Even on RTX 20/30 series GPUs, with Preset K, you can maintain the same performance as before while getting better image quality by using a lower upscaling mode, like Balanced instead of Quality. Hopefully Ubisoft Support fixes the scaling issue and adds an in-game option to choose between DLSS 3, DLSS 4, and DLSS 4.5.

There are two things worth pointing out about upscaling in #AssassinCreed Black Flag Resynced. 1. Incorrect resolution scaling with Balanced and Performance modes: Using Balanced or Performance with any upscaler causes the game's Resolution Scale (output resolution) to drop, which also lowers the internal render resolution. As a result, these modes apply incorrect scaling and the image quality is noticeably worse than it should be. For now, the workaround is to set the Resolution Scale to 100% and manually adjust the Max Dynamic Resolution (internal resolution) to match the expected scaling for Balanced or Performance mode. 2. The game defaults to the older DLSS 3 CNN model: By default, the game uses the older DLSS 3 Preset E. While it's still a fine option, it doesn't match the image quality of the DLSS 4 Transformer model (Preset K) or the newer DLSS 4.5 (Preset L and M). So I highly recommend forcing at least Preset K through the NVIDIA App (or another tool). Even on RTX 20/30 series GPUs, with Preset K, you can maintain the same performance as before while getting better image quality by using a lower upscaling mode, like Balanced instead of Quality. Hopefully Ubisoft Support fixes the scaling issue and adds an in-game option to choose between DLSS 3, DLSS 4, and DLSS 4.5.

BenchmarKing

38,817 просмотров • 2 месяцев назад

Wuthering Waves PS5 Performance Test • Test Setup PS5 Slim (brand new, clean, no dust) Room temperature: 22°C with AC • Map Issue Yes, it's real. The new map is locked to around 2 FPS when opened from the menu, making it extremely laggy. • Stuttering Problems The game has severe stuttering. In some areas, it even freezes for 4–5 seconds, making the experience frustrating. • My Observation From my testing, the game appears to be running at a very high internal resolution on PS5 instead of relying on aggressive upscaling. For example, DLSS Quality at 1440p renders the game internally at approximately 1707×960 before upscaling it to 2560×1440. On PS5, however, it looks like the game is rendering at or near 2560×1440 internally and then upscaling to 4K. If that's true, it's an extremely demanding target for a base PS5. • Is This Fixable? Yes. If the game is indeed using such a high internal resolution, lowering it and using better reconstruction or dynamic resolution scaling could significantly improve performance. Running at such a high internal resolution on a base PS5 doesn't make much sense, and the game has suffered from performance issues for nearly 12 months. • Final Thoughts Kuro Games needs to address this. At this point, we can't keep blaming players, their hardware, or the engine. If the high internal resolution is the cause, it should be adjusted. If it isn't, then something else in the game's optimization is seriously wrong. #WutheringWaves #Gacha

Kaito

50,500 просмотров • 2 месяцев назад

Rendering 3D is gonna be SO MUCH FASTER 🤯 Thanks to NVIDIA DLSS Technology 🔥 We were at the Blender 🔶 booth at SIGGRAPH 25 🇨🇦, and we saw the Demo for Blender Cycles viewport + DLSS which blew our minds. In Blender, DLSS (Deep Learning Super Sampling) uses AI to predict and reconstruct missing pixels by rendering at a lower resolution and then intelligently upscaling in real time, boosting frame rates while preserving near-native image quality. We have a FULL VLOG of SIGGRAPH coming up on our Youtube channel soon so if you wanna see all the Tech updates make sure you go and turn on notification 🔔

Rendering 3D is gonna be SO MUCH FASTER 🤯 Thanks to NVIDIA DLSS Technology 🔥 We were at the Blender 🔶 booth at SIGGRAPH 25 🇨🇦, and we saw the Demo for Blender Cycles viewport + DLSS which blew our minds. In Blender, DLSS (Deep Learning Super Sampling) uses AI to predict and reconstruct missing pixels by rendering at a lower resolution and then intelligently upscaling in real time, boosting frame rates while preserving near-native image quality. We have a FULL VLOG of SIGGRAPH coming up on our Youtube channel soon so if you wanna see all the Tech updates make sure you go and turn on notification 🔔

Bad Decisions Studio

26,057 просмотров • 1 год назад

zeroscope_v2 XL, A watermark-free Modelscope-based video model capable of generating high quality video at 1024 x 576 Model on Hugging Face : This model was trained with offset noise using 9,923 clips and 29,769 tagged frames at 24 frames, 1024x576 resolution. zeroscope_v2_XL is specifically designed for upscaling content made with zeroscope_v2_576w using vid2vid in the 1111 text2video extension by kabachuha. Leveraging this model as an upscaler allows for superior overall compositions at higher resolutions, permitting faster exploration in 576x320 (or 448x256) before transitioning to a high-resolution render. zeroscope_v2_XL uses 15.3gb of vram when rendering 30 frames at 1024x576

zeroscope_v2 XL, A watermark-free Modelscope-based video model capable of generating high quality video at 1024 x 576 Model on Hugging Face : This model was trained with offset noise using 9,923 clips and 29,769 tagged frames at 24 frames, 1024x576 resolution. zeroscope_v2_XL is specifically designed for upscaling content made with zeroscope_v2_576w using vid2vid in the 1111 text2video extension by kabachuha. Leveraging this model as an upscaler allows for superior overall compositions at higher resolutions, permitting faster exploration in 576x320 (or 448x256) before transitioning to a high-resolution render. zeroscope_v2_XL uses 15.3gb of vram when rendering 30 frames at 1024x576

AK

714,563 просмотров • 3 лет назад

Upscale-A-Video: Temporal-Consistent Diffusion Model for Real-World Video Super-Resolution paper page: Text-based diffusion models have exhibited remarkable success in generation and editing, showing great promise for enhancing visual content with their generative prior. However, applying these models to video super-resolution remains challenging due to the high demands for output fidelity and temporal consistency, which is complicated by the inherent randomness in diffusion models. Our study introduces Upscale-A-Video, a text-guided latent diffusion framework for video upscaling. This framework ensures temporal coherence through two key mechanisms: locally, it integrates temporal layers into U-Net and VAE-Decoder, maintaining consistency within short sequences; globally, without training, a flow-guided recurrent latent propagation module is introduced to enhance overall video stability by propagating and fusing latent across the entire sequences. Thanks to the diffusion paradigm, our model also offers greater flexibility by allowing text prompts to guide texture creation and adjustable noise levels to balance restoration and generation, enabling a trade-off between fidelity and quality. Extensive experiments show that Upscale-A-Video surpasses existing methods in both synthetic and real-world benchmarks, as well as in AI-generated videos, showcasing impressive visual realism and temporal consistency.

Upscale-A-Video: Temporal-Consistent Diffusion Model for Real-World Video Super-Resolution paper page: Text-based diffusion models have exhibited remarkable success in generation and editing, showing great promise for enhancing visual content with their generative prior. However, applying these models to video super-resolution remains challenging due to the high demands for output fidelity and temporal consistency, which is complicated by the inherent randomness in diffusion models. Our study introduces Upscale-A-Video, a text-guided latent diffusion framework for video upscaling. This framework ensures temporal coherence through two key mechanisms: locally, it integrates temporal layers into U-Net and VAE-Decoder, maintaining consistency within short sequences; globally, without training, a flow-guided recurrent latent propagation module is introduced to enhance overall video stability by propagating and fusing latent across the entire sequences. Thanks to the diffusion paradigm, our model also offers greater flexibility by allowing text prompts to guide texture creation and adjustable noise levels to balance restoration and generation, enabling a trade-off between fidelity and quality. Extensive experiments show that Upscale-A-Video surpasses existing methods in both synthetic and real-world benchmarks, as well as in AI-generated videos, showcasing impressive visual realism and temporal consistency.

AK

32,849 просмотров • 2 лет назад

[SIGGRAPH '25] TeGA: Texture Space Gaussian Avatars for High-Resolution Dynamic Head Modeling Note: On the left that's a 3DGS rendering! Contributions: 1. We propose a simple approach for rigging 3D Gaussians within the continuous tangent space of 3DMM face models, allowing Gaussians to move freely across mesh triangles. 2. We propose a novel CNN-based deformation model that is agnostic to the number of 3D Gaussians, naturally enabling adaptively densification of the representation to improve detail where most needed, with expression-dependent shading. 3. We show significant improvements over baseline SOTA methods and demonstrate the ability to render even extreme close-up images at high quality.

[SIGGRAPH '25] TeGA: Texture Space Gaussian Avatars for High-Resolution Dynamic Head Modeling Note: On the left that's a 3DGS rendering! Contributions: 1. We propose a simple approach for rigging 3D Gaussians within the continuous tangent space of 3DMM face models, allowing Gaussians to move freely across mesh triangles. 2. We propose a novel CNN-based deformation model that is agnostic to the number of 3D Gaussians, naturally enabling adaptively densification of the representation to improve detail where most needed, with expression-dependent shading. 3. We show significant improvements over baseline SOTA methods and demonstrate the ability to render even extreme close-up images at high quality.

MrNeRF

29,057 просмотров • 1 год назад

"Create a video project announcing Agent Skills, following Cursor brand rules. 1920×1080, 60fps. Typewriter text effect at about 1 character per frame, then hold for 3 seconds after typing on every title card. Line 1 types first, then line 2. Card 1: “Agent Skills are now available” on line 1 and “in Cursor” on line 2 Card 2: “Skills let agents discover and run” and line 2 “specialized prompts and code.” Then play the skills.mp4 recording full-screen, framed top-aligned for first 2 seconds, then zoom in (with easing) to the top-left by about 125%. Should be continuous zoom! Final card: “This video was made entirely in Cursor" on line 1 and "with Remotion skills.” on line 2, then hold; End with the existing end.mp4 animation full-screen. Render and open the file inside Cursor’s browser tool to let me see it here!" 1. Uses an existing screen recording eric zakariasson made. 2. Uses our existing end card animation. 3. Had a Cursor rule with our brand guidelines and colors.

edwin

73,538 просмотров • 8 месяцев назад

🔥 HONKAI: STAR RAIL IS JOINING THE NEXT GENERATION OF MOBILE GAMING Qualcomm has officially confirmed Honkai: Star Rail as 1 of the games supporting its new Adreno Neural Fusion technology. 👀📱 ANF brings AI directly into the mobile graphics pipeline with: ✅ AI Super Resolution ✅ AI Frame Generation ✅ Higher perceived frame rates ✅ Lower rendering workload ✅ Better power efficiency ✅ Potentially longer sustained gaming performance Instead of forcing the GPU to render every frame at full resolution, ANF can reconstruct a lower resolution image and generate additional frames with AI. Qualcomm says this approach can significantly reduce the GPU workload and power consumption on supported hardware. And Honkai: Star Rail is already among the confirmed games being prepared for it. Mobile gaming is getting ridiculous. 🔥 NTE, Silver Palace, Honkai: Star Rail and several other major games are now pushing these next generation Snapdragon graphics technologies. (Follow and like for more FroGGy news) #HonkaiStarRail #HSR #Gaming #RPG

🔥 HONKAI: STAR RAIL IS JOINING THE NEXT GENERATION OF MOBILE GAMING Qualcomm has officially confirmed Honkai: Star Rail as 1 of the games supporting its new Adreno Neural Fusion technology. 👀📱 ANF brings AI directly into the mobile graphics pipeline with: ✅ AI Super Resolution ✅ AI Frame Generation ✅ Higher perceived frame rates ✅ Lower rendering workload ✅ Better power efficiency ✅ Potentially longer sustained gaming performance Instead of forcing the GPU to render every frame at full resolution, ANF can reconstruct a lower resolution image and generate additional frames with AI. Qualcomm says this approach can significantly reduce the GPU workload and power consumption on supported hardware. And Honkai: Star Rail is already among the confirmed games being prepared for it. Mobile gaming is getting ridiculous. 🔥 NTE, Silver Palace, Honkai: Star Rail and several other major games are now pushing these next generation Snapdragon graphics technologies. (Follow and like for more FroGGy news) #HonkaiStarRail #HSR #Gaming #RPG

FroGGy🐸

46,017 просмотров • 11 дней назад

👇BEHOLD: The proof people asked for; 30 seconds of Seedance 2.5 480p → Topaz 4K using upscaling. Topaz 4K on top, 480p source below. Same frame. Watch the closeups from :04 onwards for the best comparison of the subject for where the most detail in the upscale is visible. (Best compared on full screen) 💻 The Process: This clip was upscaled using Topaz's "Rhea" model, which does 4x upscaling. With the resolution of our original 21:9 aspect ratio clip increased to: 3440x1498 The "Proteus" model is better for general upscaling of 720p; "Rhea" makes 480p footage look as good as 4K clips upscaled from 720p or 1080p. 💭Thoughts: The only downside for using Rhea is the render time if you aren't running on decent hardware and a noticable, slight grain on renders. Otherwise, I wholly recommend Rhea over Proteus if you want razor sharpness in 480p clips. For general upscaling and speed, stick with Proteus which also dials back the sharpness for the nice soft filmic look. Skin holds texture instead of going plastic, the cloth weave comes back, and the background details fill in. 🗒️Note: [Differences may seem subtle here due to X's video compression. I know a lot of these comparisons soften the low-res side on purpose to make the jump look bigger. Keeping it sincere here. Cost breakdown coming soon.]

👇BEHOLD: The proof people asked for; 30 seconds of Seedance 2.5 480p → Topaz 4K using upscaling. Topaz 4K on top, 480p source below. Same frame. Watch the closeups from :04 onwards for the best comparison of the subject for where the most detail in the upscale is visible. (Best compared on full screen) 💻 The Process: This clip was upscaled using Topaz's "Rhea" model, which does 4x upscaling. With the resolution of our original 21:9 aspect ratio clip increased to: 3440x1498 The "Proteus" model is better for general upscaling of 720p; "Rhea" makes 480p footage look as good as 4K clips upscaled from 720p or 1080p. 💭Thoughts: The only downside for using Rhea is the render time if you aren't running on decent hardware and a noticable, slight grain on renders. Otherwise, I wholly recommend Rhea over Proteus if you want razor sharpness in 480p clips. For general upscaling and speed, stick with Proteus which also dials back the sharpness for the nice soft filmic look. Skin holds texture instead of going plastic, the cloth weave comes back, and the background details fill in. 🗒️Note: [Differences may seem subtle here due to X's video compression. I know a lot of these comparisons soften the low-res side on purpose to make the jump look bigger. Keeping it sincere here. Cost breakdown coming soon.]

God's Way Foundation

23,702 просмотров • 1 месяц назад

remind-reid-tracker REMIND — RE-Identification with Memory for INDoor Navigation REMIND addresses a core challenge in visual tracking: re-identifying objects that disappear and reappear, look similar to one another, or are observed from changing viewpoints. Rather than relying on position or motion cues, REMIND builds appearance-based identity models per object using DINOv3 patch features, decomposed into: Global and part-level descriptors — K-means and attention-guided semantic parts extracted per detection Relational context (neighbor sets) — structural scene layout encoded as co-occurrence graphs of neighboring objects Known-set distance disambiguation — geometry-aware resolution of visually ambiguous groups Adaptive memory — per-object appearance, part, and background models that update over time The association pipeline runs a per-frame sequence of visual evidence building, context activation, global Hungarian assignment, and post-assignment guards, producing explicit uncertainty signals (ambiguous, provisional) alongside confident identity decisions. Evaluation outputs span case, object, frame, class, scene, and batch levels, with full internal telemetry for diagnostic analysis.

remind-reid-tracker REMIND — RE-Identification with Memory for INDoor Navigation REMIND addresses a core challenge in visual tracking: re-identifying objects that disappear and reappear, look similar to one another, or are observed from changing viewpoints. Rather than relying on position or motion cues, REMIND builds appearance-based identity models per object using DINOv3 patch features, decomposed into: Global and part-level descriptors — K-means and attention-guided semantic parts extracted per detection Relational context (neighbor sets) — structural scene layout encoded as co-occurrence graphs of neighboring objects Known-set distance disambiguation — geometry-aware resolution of visually ambiguous groups Adaptive memory — per-object appearance, part, and background models that update over time The association pipeline runs a per-frame sequence of visual evidence building, context activation, global Hungarian assignment, and post-assignment guards, producing explicit uncertainty signals (ambiguous, provisional) alongside confident identity decisions. Evaluation outputs span case, object, frame, class, scene, and batch levels, with full internal telemetry for diagnostic analysis.

Ryohei Sasaki@engineer

19,013 просмотров • 2 месяцев назад

$FAU Erlangen-Nürnberg presents TRIPS Trilinear Point Splatting for Real-Time Radiance Field Rendering paper page: Point-based radiance field rendering has demonstrated impressive results for novel view synthesis, offering a compelling blend of rendering quality and computational efficiency. However, also latest approaches in this domain are not without their shortcomings. 3D Gaussian Splatting [Kerbl and Kopanas et al. 2023] struggles when tasked with rendering highly detailed scenes, due to blurring and cloudy artifacts. On the other hand, ADOP [R\"uckert et al. 2022] can accommodate crisper images, but the neural reconstruction network decreases performance, it grapples with temporal instability and it is unable to effectively address large gaps in the point cloud. In this paper, we present TRIPS (Trilinear Point Splatting), an approach that combines ideas from both Gaussian Splatting and ADOP. The fundamental concept behind our novel technique involves rasterizing points into a screen-space image pyramid, with the selection of the pyramid layer determined by the projected point size. This approach allows rendering arbitrarily large points using a single trilinear write. A lightweight neural network is then used to reconstruct a hole-free image including detail beyond splat resolution. Importantly, our render pipeline is entirely differentiable, allowing for automatic optimization of both point sizes and positions. Our evaluation demonstrate that TRIPS surpasses existing state-of-the-art methods in terms of rendering quality while maintaining a real-time frame rate of 60 frames per second on readily available hardware. This performance extends to challenging scenarios, such as scenes featuring intricate geometry, expansive landscapes, and auto-exposed footage.$

FAU Erlangen-Nürnberg presents TRIPS Trilinear Point Splatting for Real-Time Radiance Field Rendering paper page: Point-based radiance field rendering has demonstrated impressive results for novel view synthesis, offering a compelling blend of rendering quality and computational efficiency. However, also latest approaches in this domain are not without their shortcomings. 3D Gaussian Splatting [Kerbl and Kopanas et al. 2023] struggles when tasked with rendering highly detailed scenes, due to blurring and cloudy artifacts. On the other hand, ADOP [R\"uckert et al. 2022] can accommodate crisper images, but the neural reconstruction network decreases performance, it grapples with temporal instability and it is unable to effectively address large gaps in the point cloud. In this paper, we present TRIPS (Trilinear Point Splatting), an approach that combines ideas from both Gaussian Splatting and ADOP. The fundamental concept behind our novel technique involves rasterizing points into a screen-space image pyramid, with the selection of the pyramid layer determined by the projected point size. This approach allows rendering arbitrarily large points using a single trilinear write. A lightweight neural network is then used to reconstruct a hole-free image including detail beyond splat resolution. Importantly, our render pipeline is entirely differentiable, allowing for automatic optimization of both point sizes and positions. Our evaluation demonstrate that TRIPS surpasses existing state-of-the-art methods in terms of rendering quality while maintaining a real-time frame rate of 60 frames per second on readily available hardware. This performance extends to challenging scenarios, such as scenes featuring intricate geometry, expansive landscapes, and auto-exposed footage.

AK

45,489 просмотров • 2 лет назад

The reason why Reflections in GTA 6 look noisy and blurry isn't because Rockstar Games doesn't know how to do them or that a $2 billion budget magically fixes this. It all comes down to the GPU in current gen consoles, these consoles don't have the power they need to render high resolution Ray Traced reflections in real-time and corners had to be cut. Here I'm running GTA 5 enhanced for PC at ultra settings to show you how good and clean the reflections look if you have high-end and recent hardware that have AI upscaling to make this possible. That being said, The PS5 Pro console is much more powerful compared to the base PS5 (the footage which GTA 6 was captured) offering higher resolution, 45% faster rendering, 2X-3X better Ray Tracing performance and PSSR upscaling. With the PS5 Pro, it will have significantly better looking and sharper reflections somewhat close to what GTA 5 Enhanced reflections look like on a gaming PC at high settings. However, the ultimate experience will still be on PC when GTA 6 launches, allowing for the highest quality Ray Tracing or even Path Tracing if Rockstar Games decides to implement it.

The reason why Reflections in GTA 6 look noisy and blurry isn't because Rockstar Games doesn't know how to do them or that a $2 billion budget magically fixes this. It all comes down to the GPU in current gen consoles, these consoles don't have the power they need to render high resolution Ray Traced reflections in real-time and corners had to be cut. Here I'm running GTA 5 enhanced for PC at ultra settings to show you how good and clean the reflections look if you have high-end and recent hardware that have AI upscaling to make this possible. That being said, The PS5 Pro console is much more powerful compared to the base PS5 (the footage which GTA 6 was captured) offering higher resolution, 45% faster rendering, 2X-3X better Ray Tracing performance and PSSR upscaling. With the PS5 Pro, it will have significantly better looking and sharper reflections somewhat close to what GTA 5 Enhanced reflections look like on a gaming PC at high settings. However, the ultimate experience will still be on PC when GTA 6 launches, allowing for the highest quality Ray Tracing or even Path Tracing if Rockstar Games decides to implement it.

NikTek

143,307 просмотров • 1 месяц назад

Meta just announced FlowVid Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis paper page: Diffusion models have transformed the image-to-image (I2I) synthesis and are now permeating into videos. However, the advancement of video-to-video (V2V) synthesis has been hampered by the challenge of maintaining temporal consistency across video frames. This paper proposes a consistent V2V synthesis framework by jointly leveraging spatial conditions and temporal optical flow clues within the source video. Contrary to prior methods that strictly adhere to optical flow, our approach harnesses its benefits while handling the imperfection in flow estimation. We encode the optical flow via warping from the first frame and serve it as a supplementary reference in the diffusion model. This enables our model for video synthesis by editing the first frame with any prevalent I2I models and then propagating edits to successive frames. Our V2V model, FlowVid, demonstrates remarkable properties: (1) Flexibility: FlowVid works seamlessly with existing I2I models, facilitating various modifications, including stylization, object swaps, and local edits. (2) Efficiency: Generation of a 4-second video with 30 FPS and 512x512 resolution takes only 1.5 minutes, which is 3.1x, 7.2x, and 10.5x faster than CoDeF, Rerender, and TokenFlow, respectively. (3) High-quality: In user studies, our FlowVid is preferred 45.7% of the time, outperforming CoDeF (3.5%), Rerender (10.2%), and TokenFlow (40.4%).

Meta just announced FlowVid Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis paper page: Diffusion models have transformed the image-to-image (I2I) synthesis and are now permeating into videos. However, the advancement of video-to-video (V2V) synthesis has been hampered by the challenge of maintaining temporal consistency across video frames. This paper proposes a consistent V2V synthesis framework by jointly leveraging spatial conditions and temporal optical flow clues within the source video. Contrary to prior methods that strictly adhere to optical flow, our approach harnesses its benefits while handling the imperfection in flow estimation. We encode the optical flow via warping from the first frame and serve it as a supplementary reference in the diffusion model. This enables our model for video synthesis by editing the first frame with any prevalent I2I models and then propagating edits to successive frames. Our V2V model, FlowVid, demonstrates remarkable properties: (1) Flexibility: FlowVid works seamlessly with existing I2I models, facilitating various modifications, including stylization, object swaps, and local edits. (2) Efficiency: Generation of a 4-second video with 30 FPS and 512x512 resolution takes only 1.5 minutes, which is 3.1x, 7.2x, and 10.5x faster than CoDeF, Rerender, and TokenFlow, respectively. (3) High-quality: In user studies, our FlowVid is preferred 45.7% of the time, outperforming CoDeF (3.5%), Rerender (10.2%), and TokenFlow (40.4%).

AK

123,729 просмотров • 2 лет назад

5001 lights in WebGPU (Shade). 1 directional shadowed light with CSM + 5000 sphere lights. Using clustered lighting, similar to Unreal, the screen is split into tiles of 32x32 pixels and then further split along the Z axis. Lights are culled and binned. Most of the time engines implement clustered lighting with a fixed number of partitions in XY, something like 40x20, let's say, this is generally fine, but it doesn't adapt to different aspect ratios and tends not to scale as well into higher screen resolutions. Finally, by splitting the XY in 32 pixel increments (or anything that's 2^N really) - you get better wave coherence, that is - pixel shader runs more efficiently. The engine supports up to 10,000 so-called "local lights" in the frame, which can be either be sphere lights (point lights with volume) or spot lights. When I say "in the frame" - I mean on-screen at the same time. There can be a million lights in the scene for example, but only 10,000 can be visible at a time. Can increase the limit, but I reckon 10,000 is already a pretty (un)reasonable limit.

5001 lights in WebGPU (Shade). 1 directional shadowed light with CSM + 5000 sphere lights. Using clustered lighting, similar to Unreal, the screen is split into tiles of 32x32 pixels and then further split along the Z axis. Lights are culled and binned. Most of the time engines implement clustered lighting with a fixed number of partitions in XY, something like 40x20, let's say, this is generally fine, but it doesn't adapt to different aspect ratios and tends not to scale as well into higher screen resolutions. Finally, by splitting the XY in 32 pixel increments (or anything that's 2^N really) - you get better wave coherence, that is - pixel shader runs more efficiently. The engine supports up to 10,000 so-called "local lights" in the frame, which can be either be sphere lights (point lights with volume) or spot lights. When I say "in the frame" - I mean on-screen at the same time. There can be a million lights in the scene for example, but only 10,000 can be visible at a time. Can increase the limit, but I reckon 10,000 is already a pretty (un)reasonable limit.

Alex Goldring

13,620 просмотров • 8 месяцев назад

This trader used Claude to build a Quant Bot and made +$823,796 on Polymarket 95,943 predictions in 103 days with a 52% win rate How is this account bringing in about $7,998 per day? The logic is simple: 1. It stays active in short crypto “Up / Down” markets at roughly 39 trades per hour 2. The strategy appears to blend temporal arbitrage, hedged directional exposure, and constant inventory rotation 3. It can take exposure on one outcome first, wait for the probabilities to move, then bring in the opposite side when that gives the position a cleaner hedge 4. As the signal changes, the bot keeps redistributing exposure instead of sitting on the same structure until resolution This trader’s Polymarket account: Most profitable trades: $3,636 → $10,868 (+$7,232 +198.9%) $4,278 → $11,165 (+$6,887 +161.0%) $5,771 → $12,130 (+$6,359 +110.2%) The PnL curve is built by running the same adaptive setup across nearly 100,000 entries and letting repeated small advantages accumulate into a much larger result

This trader used Claude to build a Quant Bot and made +$823,796 on Polymarket 95,943 predictions in 103 days with a 52% win rate How is this account bringing in about $7,998 per day? The logic is simple: 1. It stays active in short crypto “Up / Down” markets at roughly 39 trades per hour 2. The strategy appears to blend temporal arbitrage, hedged directional exposure, and constant inventory rotation 3. It can take exposure on one outcome first, wait for the probabilities to move, then bring in the opposite side when that gives the position a cleaner hedge 4. As the signal changes, the bot keeps redistributing exposure instead of sitting on the same structure until resolution This trader’s Polymarket account: Most profitable trades: $3,636 → $10,868 (+$7,232 +198.9%) $4,278 → $11,165 (+$6,887 +161.0%) $5,771 → $12,130 (+$6,359 +110.2%) The PnL curve is built by running the same adaptive setup across nearly 100,000 entries and letting repeated small advantages accumulate into a much larger result

Daniro

19,152 просмотров • 1 месяц назад

** 3D Engine Update 6 ** New Buffer System. 15-20% Faster !! Using Planes A and B interleaved horizontally we can get the best of all worlds with a near continuous frame buffer for fast Triangle & Quad writes and still use DMA for fast sending of the RAM frame buffer to VRAM. Normally you get one benefit without the other leading to tradeoffs. There is a caveat ofcourse, I lose a spare plane , however thats not a large concern for what im trying to achieve, i still have the full sprite load if i need to simulate ground etc - we can also multiplex sprites if needed. So each plane is full of half fill tiles ( 4 pixels set in each tile row ) , which are then slightly offset to make the full image. Also changed to H40 mode for faster DMA albiet centered view , plenty of Hud options for side panels etc. What is quite cool is the layout in VRAM is very similar to what you see on screen just with some striping . VRAM layout to the right. Theres no more VRAM used verse a normal double buffer , the trick is here to send the buffer in one frame always as theres no room for a double buffer in vram. I have to implement partial ofscreen drawing yet hence the snapping from front to back you see still . #SGDK #SegaGenesis #SegaMegadrive

3D Engine Update 6 New Buffer System. 15-20% Faster !! Using Planes A and B interleaved horizontally we can get the best of all worlds with a near continuous frame buffer for fast Triangle & Quad writes and still use DMA for fast sending of the RAM frame buffer to VRAM. Normally you get one benefit without the other leading to tradeoffs. There is a caveat ofcourse, I lose a spare plane , however thats not a large concern for what im trying to achieve, i still have the full sprite load if i need to simulate ground etc - we can also multiplex sprites if needed. So each plane is full of half fill tiles ( 4 pixels set in each tile row ) , which are then slightly offset to make the full image. Also changed to H40 mode for faster DMA albiet centered view , plenty of Hud options for side panels etc. What is quite cool is the layout in VRAM is very similar to what you see on screen just with some striping . VRAM layout to the right. Theres no more VRAM used verse a normal double buffer , the trick is here to send the buffer in one frame always as theres no room for a double buffer in vram. I have to implement partial ofscreen drawing yet hence the snapping from front to back you see still . #SGDK #SegaGenesis #SegaMegadrive

Shannon Birt

12,758 просмотров • 5 месяцев назад

*** SOTA RETROCON S.O.T.A UPDATE *** Whilst the Versao LTDA team is busy at Retrocon showing off our latest Sega Genesis SOTA update, here is a small snippet of how the game is progressing. This video has 3 parts, our menu screen, rain - wind sub section of level 1, then the boss. Everything is still wip including my gameplay skill lol ( and i thought the controller would help .. ) Some challenges overcome : The Menu screen uses shadow and highlight transparency modes in two ways, the background mode for the light rays and then the sprite mode for the mist - it has a nice overlapping effect too, due to using the forground plane for the light rays (s & h background mode) the entire forground is made up of sprites - all the words are sprites . SaviorMarks sent me a screenshot of an issue with boss leftover Mist sprites affecting the menu screen and I thought that looks really cool - lets do that haha . Sometimes bugs are keepers !! The rain section uses more than 80 sprites, a sprite multiplexor handles the rain whilst normal non plexed sprites make up the moon / clouds in the far background , the lamp posts and lamp light in the mid forground and Fences in the front forground . Without the multiplexor there was too much drop out on the rain at times. The rain is still blinked in alternating X positions due to the pixels per line limit , but it gives a great result still. The lamp lights use Shaded Active Dither as S&H here would not overlap the enemies / player . The Boss was a challenge with its size and full frame updates, up to 650 tiles are changed per animation frame into vram. Custom Dual priority DMA routines and extended DMA timings are used (still NTSC 224 pixels tall screen size but most of the vblank was used for updating tiles ) . These methods were used to update the bosses dual display buffers. I needed to have dual priority DMA buffers as the player / waterfall etc had to update their frames whilst the boss is constantly streaming tiles. There are also S&H transparent Mist sprites used for the boss sequence. It seems if you want everything to break at once target a demo for a retrocon event haha, we had lots of issues mainly with vram getting stomped and as all animating sprites need to be DMA streamed to fit into VRAM as the forgrounds and backgrounds take up a lot of tiles. Some new tools which show the VRAM allocation for the sprite engine on a frame by frame basis helped and a self defragging VRAM system helped quite a bit also. Thankfully things are in a good state now apart from a couple of small bugs that persist. Thanks to SaviorMarks and @MichelinFabio for putting up with and testing all the bugged updates haha !! #SegaGenesis #SGDK @MichelinFabio SaviorMarks #VersaoLTDA

* SOTA RETROCON S.O.T.A UPDATE * Whilst the Versao LTDA team is busy at Retrocon showing off our latest Sega Genesis SOTA update, here is a small snippet of how the game is progressing. This video has 3 parts, our menu screen, rain - wind sub section of level 1, then the boss. Everything is still wip including my gameplay skill lol ( and i thought the controller would help .. ) Some challenges overcome : The Menu screen uses shadow and highlight transparency modes in two ways, the background mode for the light rays and then the sprite mode for the mist - it has a nice overlapping effect too, due to using the forground plane for the light rays (s & h background mode) the entire forground is made up of sprites - all the words are sprites . SaviorMarks sent me a screenshot of an issue with boss leftover Mist sprites affecting the menu screen and I thought that looks really cool - lets do that haha . Sometimes bugs are keepers !! The rain section uses more than 80 sprites, a sprite multiplexor handles the rain whilst normal non plexed sprites make up the moon / clouds in the far background , the lamp posts and lamp light in the mid forground and Fences in the front forground . Without the multiplexor there was too much drop out on the rain at times. The rain is still blinked in alternating X positions due to the pixels per line limit , but it gives a great result still. The lamp lights use Shaded Active Dither as S&H here would not overlap the enemies / player . The Boss was a challenge with its size and full frame updates, up to 650 tiles are changed per animation frame into vram. Custom Dual priority DMA routines and extended DMA timings are used (still NTSC 224 pixels tall screen size but most of the vblank was used for updating tiles ) . These methods were used to update the bosses dual display buffers. I needed to have dual priority DMA buffers as the player / waterfall etc had to update their frames whilst the boss is constantly streaming tiles. There are also S&H transparent Mist sprites used for the boss sequence. It seems if you want everything to break at once target a demo for a retrocon event haha, we had lots of issues mainly with vram getting stomped and as all animating sprites need to be DMA streamed to fit into VRAM as the forgrounds and backgrounds take up a lot of tiles. Some new tools which show the VRAM allocation for the sprite engine on a frame by frame basis helped and a self defragging VRAM system helped quite a bit also. Thankfully things are in a good state now apart from a couple of small bugs that persist. Thanks to SaviorMarks and @MichelinFabio for putting up with and testing all the bugged updates haha !! #SegaGenesis #SGDK @MichelinFabio SaviorMarks #VersaoLTDA

Shannon Birt

10,319 просмотров • 1 год назад

Added a requested feature by many: Connect apps to windows in workspaces. For the interested ones, here is a full list of the changes made, written by AI: - Pin an app to a workspace so it opens there wherever you launch it from, using a Hyprland window rule - Pin an app to a numbered place, or to several places at once - Put several apps in one place; whichever is open takes it - Search every app installed on the machine, not just what is running - Read names and launch commands from desktop entries - One button opens everything the workspace is missing, one window per empty place - Terminal apps like nvim and btop open in your terminal under a class the pin can match - A pin corrects itself when a launch opens a window under a different class, as webapps and terminal apps do - A pin remembers the command that opened it and the readable name of the app - Split a slot the other way, so a column becomes a top and a bottom - Split that half again, three levels deep: columns, then rows, then columns - Drag the divider at every level, including the one inside a split - Extra windows become real places instead of a temporary stack while you work in the panel - Editing a shipped layout makes a copy called custom-xxxx and leaves the original alone - Capture builds a layout plus pins from the windows already on screen - Drag one tile onto another and the two swap their apps - Hold a tile near an edge and that half lights up; drop it there and the place splits, your app takes the half, and the place it came from disappears - Hover a tile for two arrows that split it either way - Right-click a tile to split, merge, remove, even out, clear its apps, or put an app in it - Right-click a workspace to hand it back to Hyprland, capture it, or clear its apps - App names are drawn inside the tiles they live in - The tile you are carrying follows the cursor with its name and target - A layout per monitor, sitting between the per-workspace choice and the profile default - Clicking a workspace takes you to it - The panel scrolls when it grows taller than the screen - Single-key shortcuts are gone; Esc closes, everything else is pointed at - Sixteen command-line commands: status, workspace, json, profiles, apply, layouts, set, reset, pin, unpin, capture, launch, open, close, show, hide, toggle - Full command-line documentation in docs/cli.md

Bjarne Øverli

39,206 просмотров • 29 дней назад

INTERVIEW: Why $NEAR is Growing: The AI-Native Money Thesis - Intents, Privacy, & Tokenomics | Sal Ternullo NEAR Protocol is showing up more and more - Integrated in Zodl, right before the $ZEC explosion - Used in Venice to create private AI inferece - Infinex uses Near Intents for creating a unified wallet What's going on over at NEAR Protocol, and how does Near adoption impact $NEAR? Sal Ternullo, CEO of SVRN, joins us to explain why he thinks the original NEAR thesis finally coming into focus. They dig into NEAR Intents, AI money, tokenomics, privacy, fee capture, agentic commerce, and why SVRN is trying to commercialize the NEAR ecosystem rather than simply hold the asset. [TIMESTAMPS] 0:00 Intro 2:23 Why NEAR is Everywhere Lately 3:58 Pivot or Return to Roots? 7:41 The “AI Money” Thesis 9:54 Intents, Zashi, and Real Product Market Fit 11:38 Tokenomics, Buybacks, and Value Accrual 14:45 NEAR’s Different Approach From Ethereum 18:00 Fees, Scaling, and First-Party Apps 23:04 Cross-Chain Liquidity as NEAR’s Moat 28:27 The Contrarian Investor Case for NEAR 31:43 Privacy, Enterprises, and AI Agents 39:38 What SVRN Is Building for NEAR 46:00 Investor Sentiment and Final Thoughts

INTERVIEW: Why $NEAR is Growing: The AI-Native Money Thesis - Intents, Privacy, & Tokenomics | Sal Ternullo NEAR Protocol is showing up more and more - Integrated in Zodl, right before the $ZEC explosion - Used in Venice to create private AI inferece - Infinex uses Near Intents for creating a unified wallet What's going on over at NEAR Protocol, and how does Near adoption impact $NEAR? Sal Ternullo, CEO of SVRN, joins us to explain why he thinks the original NEAR thesis finally coming into focus. They dig into NEAR Intents, AI money, tokenomics, privacy, fee capture, agentic commerce, and why SVRN is trying to commercialize the NEAR ecosystem rather than simply hold the asset. [TIMESTAMPS] 0:00 Intro 2:23 Why NEAR is Everywhere Lately 3:58 Pivot or Return to Roots? 7:41 The “AI Money” Thesis 9:54 Intents, Zashi, and Real Product Market Fit 11:38 Tokenomics, Buybacks, and Value Accrual 14:45 NEAR’s Different Approach From Ethereum 18:00 Fees, Scaling, and First-Party Apps 23:04 Cross-Chain Liquidity as NEAR’s Moat 28:27 The Contrarian Investor Case for NEAR 31:43 Privacy, Enterprises, and AI Agents 39:38 What SVRN Is Building for NEAR 46:00 Investor Sentiment and Final Thoughts

Bankless

53,730 просмотров • 4 месяцев назад

Path Tracing on PS5 Pro even as a demo and at 30fps, pretty much confirms at least some games will use PT on PS6, which I said months ago but some said I was crazy for suggesting. PC gamers won't want to admit it, but next-gen the gap between console and higher end PC may shrink in terms of real world differences, bolstered by more competent RT and AI upscaling tech in the next-gen consoles. Keep in mind Valve already confirmed what Steam data already showed, that "the steam machine is equal or better than ~70% of what people have at home", and the Steam Machine is reportedly similar in perfomance to the BASE PS5. In regards to PT in F1 on PS5 Pro, I say drop the base resolution further (there's examples of PSSR 2 looking great even at 864p) and give us 40fps or higher. Also, this is a reason the future/RDNA4 RT tech and custom AI hardware in PS5 Pro was noteworthy. The "2-3x" PS5's RT performance Cerny quoted clearly wasn't just PR bluster. Outside of this demo, it's already allowing WAY better RT than PS5 in some games (Cyberpunk, F1, AC Shadows etc), with better image quality and/or perfomance to boot. Such demo's show there's potentially more to extract, especially for quality modes. #PS5Pro #PS6

Path Tracing on PS5 Pro even as a demo and at 30fps, pretty much confirms at least some games will use PT on PS6, which I said months ago but some said I was crazy for suggesting. PC gamers won't want to admit it, but next-gen the gap between console and higher end PC may shrink in terms of real world differences, bolstered by more competent RT and AI upscaling tech in the next-gen consoles. Keep in mind Valve already confirmed what Steam data already showed, that "the steam machine is equal or better than ~70% of what people have at home", and the Steam Machine is reportedly similar in perfomance to the BASE PS5. In regards to PT in F1 on PS5 Pro, I say drop the base resolution further (there's examples of PSSR 2 looking great even at 864p) and give us 40fps or higher. Also, this is a reason the future/RDNA4 RT tech and custom AI hardware in PS5 Pro was noteworthy. The "2-3x" PS5's RT performance Cerny quoted clearly wasn't just PR bluster. Outside of this demo, it's already allowing WAY better RT than PS5 in some games (Cyberpunk, F1, AC Shadows etc), with better image quality and/or perfomance to boot. Such demo's show there's potentially more to extract, especially for quality modes. #PS5Pro #PS6

NIB

193,851 просмотров • 5 месяцев назад