Загрузка видео...

Не удалось загрузить видео

На главную

Sakura Bonsai by Hy4 Preview developed using their WorkBuddy client. Prompt is below (big one) I think is from Majid Manzarpour 👇 Thanks for the early access 🙏🏻 Tencent Hy Tencent AI WorkBuddy

40,712 просмотров • 2 месяцев назад •via X (Twitter)

Комментарии: 13

Фото профиля Ivan Fioravanti ᯅ
Ivan Fioravanti ᯅ2 месяцев назад

Prompt: BUILD: single-file Three.js diorama. One ultra-detailed cherry blossom bonsai in a tokonoma alcove. Macro scale, museum lighting, Japanese interface design done with restraint. This is a quiet piece, not a spectacle. The bar is "a still frame is indistinguishable from a product photograph, and the UI looks like it was designed by someone who has actually read Kenya Hara." === HARD CONSTRAINTS === One .html file. No build step, no bundler. three r170 via importmap (CDN ESM). Addons allowed from the same version. ZERO external assets. Every mesh, texture, LUT, font fallback and sound generated in code. No CDN fonts, no data-URI blobs pasted in. No truncation, no TODOs, no "// rest of implementation here". Ship the file. UPPERCASE section header comments dividing the file. Desktop 60fps, working mobile path. Report at the end: tri count, draw calls, techniques. === SCALE === Tree ~45cm tall. Pot ~35cm wide. Camera 0.4m to 2m out. Near plane 0.02, far plane 20. Single frustum: at this depth range there is no precision problem and no reason for split cameras or log depth. Spend the entire budget on surface detail, because at macro distance every surface is a hero surface. === THE TREE (this is 70% of the work) === Species: Prunus incisa 'Kojo-no-mai', the fuji cherry, which is what an actual sakura bonsai is. Not somei yoshino, which does not reduce. Style: moyogi (informal upright) with real movement, or shakan (slanting). Never a straight trunk. The trunk should change direction 3 or 4 times, each turn tightening as it rises. NEBARI: surface root flare spreading radially into the soil. This is the single strongest signal of an old bonsai and it is what everyone omits. Roots emerge visibly, radiate outward, taper into the akadama. TAPER: aggressive. Trunk caliper at nebari to apex should drop by 10x or more. A cylinder reads as a stick with flowers on it. RAMIFICATION: fine division at the tips, 2 or 3 branchlets per node, 5 or 6 orders deep. Age lives in the twigs, not the trunk. Kojo-no-mai grows in ZIGZAGS. Twigs change direction at each node. Build this into the skeleton generator, it is free character. PAD STRUCTURE: flowers and twigs mass into discrete horizontal pads with real negative space between them. You should see sky through the tree. A solid pink blob is the failure state. Bark: fissured, with horizontal lenticels, moss in the crotches, healed pruning scars, faint wire spiral marks on one or two older branches. At this distance use real displacement or parallax occlusion, not a normal map alone. Blossoms: 2500 to 6000, not 80k. Each one is on screen at 2cm across, so build a real flower: 5 notched petals, visible stamens, sepals, a bud state, a spent-and-browning state on a few percent. Per-instance hue jitter white through pale pink, per-instance open amount. Blooms before leaf-out. A handful of bronze new leaves at most. === POT, SOIL, STAGING === Glazed stoneware pot, shallow oval or soft rectangle, pale celadon or cream with crazing and pooling in the recesses. Feet, a proper lip profile, wire tie-down holes, visible drainage mesh. Soil: instanced akadama granules, angular and red-brown, mixed pumice. Moss: shell layers or instanced strands, two colour populations, thicker at the pot edge and the nebari crotches. Fallen petals on the soil, the pot rim, the stand and the floor. TOKONOMA: raised wooden floor, pale clay back wall, dark lacquered display stand (shoku), a hanging scroll (kakejiku) with a single ink gesture, tatami edge in frame. A low accent plant (kusamono) offset to one side. ASYMMETRY IS MANDATORY. Never centre the tree in the alcove or the frame. === LIGHT === Key: daylight through a shoji screen off to one side. Large area source, soft wrap, subtly cool. This one choice carries the entire Japanese read. Bounce: warm reflection off tatami and wood into the shadow side. Build the environment analytically (shoji panel + wall + floor as emissive areas), bake to a small cube, and project to SH9 on the CPU for ambient. Same integral drives the key colour. No PMREM: no GPU readback, works headless. PETAL TRANSLUCENCY. Wrap plus back-transmission with a red shift, so blossoms between the camera and the shoji glow and the ones facing you stay matte. Position the default camera so a good number of blossoms are backlit. Soft shadows with contact hardening. At 45cm the penumbra from a window is wide and soft, and twig shadows on the soil should be nearly diffuse while the nebari contact shadow stays tight. Optional single warm practical (paper lantern) for the night preset. === ANIMATION === Restraint. This is air moving in a room, not weather. One global breath field, sum of sines plus curl noise, very low amplitude, long period. Everything samples it. Vertex-shader sway weighted by distance from nebari and branch order. Tips move a few millimetres. The trunk does not move at all. Occasional single petal releases and falls: tumble plus glide on a slow descending spiral with a phase-flipped face normal so it flashes as it turns. One petal falling well beats a hundred falling badly. Petals land and stay. Accumulate over the session. Camera: slow drift with handheld noise, plus damped orbit constrained to the alcove (you can circle the tree, you cannot fly through the wall). === MACRO OPTICS AND POST === 90mm-equivalent macro. Aperture control from f/2.8 to f/11 that actually changes the CoC. Thin focus plane, visible falloff across the pot. Bokeh with a real iris shape and highlights that resolve into discs, not gaussian mush. Blossom specular in the out-of-focus foreground is the shot. Focus pull on interaction and a rack-focus preset. ACES filmic, exposure solved per lighting preset from measured luminance. Hand-rolled dual-filter bloom, threshold driven from exposure. Halation on petal highlights, fine grain applied AFTER the transfer curve and weighted to midtones, light vignette, near-zero chromatic aberration. TAA or 2x supersample. Twig geometry at this density crawls without it. === UX AND FEEL (first-class, not a control panel bolted on) === Design principles, in order: ma (negative space), restraint, asymmetry, one accent colour used exactly once. The screen is mostly empty. Default state shows the diorama and a single thin vertical rail at one edge. Nothing else. Palette: sumi black, washi off-white, three greys. Vermilion appears once, on the active state only. No pink UI. No gradients. Paper: a generated washi grain and a faint deckle edge on panel surfaces. Generate the texture in code to a canvas, no image files. Type: system Japanese stack (Hiragino Mincho, Yu Mincho, Noto Serif JP) with a serif latin fallback, and it must still look correct when none are present. Labels are kanji with small romaji beneath: 光 hikari, 風 kaze, 花 hana, 時 toki, 景 kei, 音 oto. One vertical-writing label group using writing-mode: vertical-rl. Motion: panels slide like fusuma from one edge, one at a time, never stacked. Easing cubic-bezier(0.22, 0.61, 0.36, 1), 450 to 700ms, long tail, NO bounce and no spring overshoot. Everything decelerates into place. Presets named by time: 朝 asa, 昼 hiru, 夕 yugure, 夜 yoru. Changing preset cross-fades the light over about 2 seconds rather than cutting. Photo mode (撮): all UI fades out, a hairline framing guide fades in, one keypress or long-press to enter and exit. Loading: a single sumi brush stroke drawn once, or an ink dot blooming into the paper. No spinner, no progress bar, no percentage. BANNED, because they are what "Japanese-inspired" usually degrades into: sakura-pink background gradients, emoji, glassmorphism, neon, torii icons, brush-script latin fonts, drop shadows on everything, rounded pills. === AUDIO (optional, off by default, one toggle) === Synthesized only. Room tone, a suikinkutsu water drip at irregular intervals, a furin chime that rings only when the breath field peaks. Quiet enough that the toggle feels like a decision, not a feature. === PERF === Instance everything: blossoms, granules, moss, petals. State a draw call budget and hit it. DPR clamp. FPS governor that walks render scale, then shadow resolution, then bloom quality, and never reduces blossom or twig count. === CARRY-FORWARD RULES (known failure modes, do not rediscover) === Swept-tube branches must start as a near-point ring buried inside the parent. A bare tube has no cap, and every shallow-angle child shows a dark angular wedge you can see straight into. Sway displacement must be a function of world position and weight ONLY. Never localT, branch depth or anchor: all three are discontinuous at exactly the joint the mesh has to stay welded across. Weight continuity alone will not save you. Compile the identical sway function into the bark, twig and blossom shaders and assert that it references none of the three. Bark lenticels and fissures author in tube UV space (u wraps, v runs along), not world triplanar, so "horizontal" tracks the branch whichever way it points. Chromatic aberration strength is in UV units. Clamp the sample UV or clamp-to-edge smears the border texel into parallel streaks. Bloom threshold in absolute radiance breaks at every exposure but the tuned one. Derive it from exposure. Grain before the transfer curve turns shadows into coloured confetti. Encode linear view distance in every material's alpha channel. One buffer then serves DoF CoC and any depth-aware upsample, with no prepass and no MRT. === VALIDATION === Before delivering: Static GLSL pass over every composed ShaderMaterial: duplicate uniform, varying and function declarations, redeclaration of three's injected builtins (modelMatrix, position, instanceMatrix), attributes in fragment stages, brace and paren balance, and any uXxx token with no declaration in that stage. This catches chunk-assembly bugs no frame driver will. Extract the module, shim three so math and geometry stay real but WebGLRenderer, PMREMGenerator and post addons are stubbed. Stub the renderer's render() to traverse and count visible meshes for a real draw-call number. In Node, define navigator with Object.defineProperty, not assignment. Drive a few thousand frames: no NaN in instance matrices, sway bounded, fallen-petal pool recycles instead of leaking, camera stays inside the alcove, shadow matrices finite, SH coefficients finite across all four presets. Assert the exposure solve lands every preset within range rather than trusting authored constants. Report what you ran and what it found. === FAILURE MODES TO CHECK YOURSELF AGAINST === Solid pink canopy with no negative space. Straight tapered trunk. No nebari. Flowers too small and too many, so it reads as noise. Pot that looks like a cereal bowl. Centred composition. UI that is a dark translucent panel with sliders. Wind that looks like outdoor weather indoors.

Фото профиля WorkBuddy
WorkBuddy2 месяцев назад

@majidmanzarpour @TencentHunyuan @TencentAI_News This is seriously impressive!! Love seeing what you built with Hy4 preview in WorkBuddy!

Фото профиля Joey
Joey2 месяцев назад

@majidmanzarpour @TencentHunyuan @TencentAI_News @WorkBuddy_AI Wow!! This is next level. Will play with it now.

Фото профиля Ivan Fioravanti ᯅ
Ivan Fioravanti ᯅ2 месяцев назад

@majidmanzarpour @TencentHunyuan @TencentAI_News @WorkBuddy_AI Very complex one, as a prompt I mean.

Фото профиля filipe
filipe2 месяцев назад

@majidmanzarpour @TencentHunyuan @TencentAI_News @WorkBuddy_AI that's seriously impressive i feel left out it's not yet on my workbuddy account!!

Фото профиля Ivan Fioravanti ᯅ
Ivan Fioravanti ᯅ2 месяцев назад

@majidmanzarpour @TencentHunyuan @TencentAI_News @WorkBuddy_AI Really? 👀

Фото профиля mzba
mzba2 месяцев назад

@majidmanzarpour @TencentHunyuan @TencentAI_News @WorkBuddy_AI Code name Nail: p

Фото профиля 花椰菜
花椰菜2 месяцев назад

@majidmanzarpour @TencentHunyuan @TencentAI_News @WorkBuddy_AI 好美

Фото профиля Michael Waitze
Michael Waitze2 месяцев назад

@majidmanzarpour @TencentHunyuan @TencentAI_News @WorkBuddy_AI Love seeing it work in production already, that's the proof point for these models. Actually, we covered the productivity capabilities enabling this here:

Фото профиля Mark
Mark2 месяцев назад

@majidmanzarpour @TencentHunyuan @TencentAI_News @WorkBuddy_AI Good update. Open productivity models expand practical options for operators.

Фото профиля BubbleBrain
BubbleBrain2 месяцев назад

@majidmanzarpour @TencentHunyuan @TencentAI_News @WorkBuddy_AI wow that is so beautiful!

Фото профиля 码良
码良2 месяцев назад

@majidmanzarpour @TencentHunyuan @TencentAI_News @WorkBuddy_AI Amazing

Фото профиля Eugene Pyvovarov
Eugene Pyvovarov2 месяцев назад

@majidmanzarpour @TencentHunyuan @TencentAI_News @WorkBuddy_AI wow, nice!

Похожие видео

🚨 I just built a game with an open-source AI model. And honestly… I didn’t expect it to be this capable. Tencent Hunyuan just released Hy4 preview, and it’s already pushing into the top tier of open-source models. Three major releases in six months. That pace is crazy. Here’s what Hy4 preview brings: → 770B total parameters → 49B active parameters → 1M+ token context window → Fully open-source But the numbers aren’t even the most interesting part. Hy4 preview was built around one goal: real-world productivity. Coding. Engineering. Office work. Science. Gaming. Finance. Security. And Tencent didn’t build it in isolation. Hy4 preview was co-designed alongside real products like WorkBuddy, using expertise and real-world data from across Tencent’s ecosystem. So I decided to test it the way I actually like testing AI models: I gave it a game idea and let WorkBuddy help turn it into a playable experience. 🎮 From the initial concept to the actual game logic, it was surprisingly smooth. And the benchmark results back up the hype: 163 internal experts 203 engineering tasks Hy4 preview — 2.99/4 Kimi K3 — 2.94/4 GLM 5.3 — 2.92/4 It also beats GLM 5.2 on benchmarks and comes remarkably close to GLM 5.3. Then comes the part I really like: 💰 ¥6/M input tokens 💰 ¥18/M output tokens 💰 ¥0.30/M cache hits Flagship-level capability without the flagship-level price. And right now, you can try Hy4 preview FREE through WorkBuddy for the next two weeks. If you’re curious what it can actually do, don’t just read the benchmarks. Build something with it. 🔗 Tencent Hy Tencent AI

Aryan Rakib

63,296 просмотров • 1 месяц назад

I wanted to see where Hy4 preview would stop, so I gave it a deliberately demanding prompt: “Build a working Cyber Incident Command Center.” Not a mockup. A real, responsive app with a live threat map, animated attack routes, incident simulation, filters, investigation panels, event logs, ownership controls, and working isolate/resolve actions. Hy4 preview planned the build, wrote the application, ran it locally, tested the interactions, checked the mobile layout, fixed issues, and opened the finished product for review. Then I clicked “Simulate.” A new incident appeared, the dashboard updated, I filtered it by severity, opened the evidence trail, isolated the affected system, and resolved it. The KPI counters changed with the action. That’s a much more useful coding demo to me than watching code appear in an editor. Hy4 preview is Hunyuan’s third major release in six months. It’s fully open-sourced, with 770B total parameters, 49B activated, and a 1M+ token context window. Tencent built it around productivity scenarios through close work with experts in software engineering, gaming, finance, and security, plus direct co-design with tools like WorkBuddy. In an internal blind test covering 203 engineering tasks and 163 Tencent experts, it scored 2.99/4—slightly ahead of Kimi K3 at 2.94 and GLM 5.3 at 2.92. And the pricing stays practical: $0.834/M input $2.501/M output $0.042/M cache hits If you want to test it yourself, Hy4 preview is free inside WorkBuddy for the next two weeks, for a limited time. Tencent Hy Tencent AI WorkBuddy

Clara Bennett

12,864 просмотров • 2 месяцев назад

I tested Tencent’s Hy4 preview model in WorkBuddy on a real frontend build, not a benchmark screenshot. I gave it one practical brief: create an original neon courier game in a single HTML file, with Canvas rendering, keyboard controls, collision detection, scoring, a countdown timer, a boost mechanic, sound effects, and reliable restart logic. The point was not to see whether it could describe a game. I wanted to see whether it could turn a creative idea into a coherent, playable result while handling the details that often break a quick prototype: state changes, movement, collisions, feedback, layout, and interaction flow. Tencent’s Hy4 preview is designed for stronger code understanding, generation, and editing, together with document and information processing, workflow automation, web and game development, cross-tool collaboration, and complex multi-step task completion. That makes it relevant not only for answering questions, but also for taking a task from brief to working output. The current preview version is faster and more effective than the first release. After the latest upgrade, the workflow takes fewer conversation rounds, uses fewer tokens, and reaches useful results more quickly. The difference is most noticeable in the handoff between idea, implementation, revision, and final testing: there is less waiting and less back-and-forth before the result becomes usable. For context, the reference materials describe Tencent’s Hy4 preview as a 770B-parameter model with 49B active parameters, a 1M-token context window, and Apache 2.0 weights. The practical question is how those capabilities translate into real work. In this case, I’m testing whether the model can produce a finished, playable game rather than just a promising code fragment. The one-minute video shows the complete process: selecting Tencent’s Hy4 preview model in WorkBuddy, submitting the build prompt, reviewing the generated result, playing the game, and checking the controls, collisions, score, timer, boost effect, sound, and restart flow. Try your own coding, frontend, document, or automation task in WorkBuddy: Tencent Hy Workbuddy Tencent AI

Rebecca Adson

61,820 просмотров • 1 месяц назад

I’ve been testing Hy4 preview in WorkBuddy, and the most interesting part is not simply the model size, it’s how much practical work it can handle with a relatively focused active parameter count. Hy4 preview brings together stronger code understanding, generation, and editing; improved document and information processing; workflow automation; web and game development; cross-tool collaboration; and more reliable completion of complex, multi-step tasks. In other words, it is designed for work that requires planning, tool use, iteration, and follow-through, not just a quick answer in a chat window. Compared with its initial release, the current Hy4 preview is noticeably faster and better-performing in practical workflows. Following an upgrade released yesterday, it can complete tasks with fewer conversation rounds and lower token usage, while reasoning more quickly and making the overall user experience feel smoother from the first instruction to the final result. For my test, I gave it a demanding Three.js game-prototyping task with a 770B-parameter model and 49B active parameters. The result was more revealing than a simple first-look demo: Hy4 preview handled the core logic, edge cases, and follow-up changes while maintaining the broader context of the project. That combination of capability, speed, context, and active compute is what makes its cost-effectiveness worth examining. A fair evaluation should use the same prompt and environment configuration across models, changing only the model itself. That makes it easier to assess task completion, planning quality, tool-calling stability, reasoning speed, token efficiency, and performance over longer workflows without confusing the result with different settings. If you want to test the model yourself, access Hy4 preview through WorkBuddy and see how it performs on a real coding, document, automation, or creative task: Tencent Hy Tencent AI WorkBuddy

Tyler Wayne

56,152 просмотров • 1 месяц назад