Video wird geladen...

Video konnte nicht geladen werden

Beim Laden dieses Videos ist ein Problem aufgetreten. Dies könnte an einem vorübergehenden Netzwerkproblem liegen oder das Video ist möglicherweise nicht verfügbar.

skin-tokens.cpp is out. Auto-rigging, running locally on CPU or Vulkan, from the same group that ported Kimodo to C++. Drop in a static mesh, get back a skeleton and skin weights, export a rigged GLB. Non-humanoid meshes too, the demo ships a giraffe. With Trellis2 for the mesh and... show more

Stefan 3D AI

5,774 subscribers

12,348 Aufrufe • vor 13 Tagen •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

10 Kommentare

Profilbild von Stefan 3D AI

Stefan 3D AIvor 13 Tagen

Repo: Original SkinTokens by VAST AI: And kimodo.cpp for the motion side:

Profilbild von Peter

Petervor 13 Tagen

Everything locally.... That's awesome. I've been trying for two weeks to get trellis or pixal3d working in comfyui, maybe today I will figure it out.

Profilbild von Digital Dimension

Digital Dimensionvor 13 Tagen

Nice, will try it with something I have done as a front-end for trellis.2!

Profilbild von cetusian

cetusianvor 13 Tagen

i'd try this tonight if it takes a messy mesh. what size before it falls over

Profilbild von alex okita

alex okitavor 13 Tagen

This is a pretty neat project! very cool to see how far along animation is coming, soon the full 3d game generation pipeline will be in place.

Profilbild von xenstudio

xenstudiovor 13 Tagen

do you happen to know if they plan on advancing kimodo.cpp to feature parity with the nvidia python tooling?

Profilbild von Jessy Leite

Jessy Leitevor 13 Tagen

Super cool!

Profilbild von Mr Pixel | Driftania 🏝

Mr Pixel | Driftania 🏝vor 13 Tagen

local auto-rig that still trips over its own ankles at the end is the most honest demo possible.

Profilbild von Stefan 3D AI

Stefan 3D AIvor 13 Tagen

That shows kimodo animations, there is an issue now. But I think author will fix that soon and I will make video

Profilbild von Mr Pixel | Driftania 🏝

Mr Pixel | Driftania 🏝vor 13 Tagen

rough handoff footage ages better than a polished lie. 🙂

Ähnliche Videos

kimodo.cpp is live. NVIDIA's Kimodo, the best AI animation model right now, running natively on CPU or Vulkan. Type a sentence, get character animation, locally on your own machine. Before, you needed the whole PyTorch and CUDA stack standing up before you could generate a single motion. Now it is a native C++/GGML build that loads GGUF models and comes with a local web demo. No Python environment to fight. It gives you SMPL-X, and from there you retarget onto pretty much any skeleton you want, like Mixamo. Great piece of work Richard Palethorpe

kimodo.cpp is live. NVIDIA's Kimodo, the best AI animation model right now, running natively on CPU or Vulkan. Type a sentence, get character animation, locally on your own machine. Before, you needed the whole PyTorch and CUDA stack standing up before you could generate a single motion. Now it is a native C++/GGML build that loads GGUF models and comes with a local web demo. No Python environment to fight. It gives you SMPL-X, and from there you retarget onto pretty much any skeleton you want, like Mixamo. Great piece of work Richard Palethorpe

Stefan 3D AI

107,037 Aufrufe • vor 21 Tagen

Ferris State HC Tony Annese - Mesh (QB/Ball Carrier) & Motion Fundamentals QB Mesh Fundamentals - Regardless of play, the first thing the QB needs to do to make sure he has a good mesh is stab the ball back off his hip towards the ball carrier as far as he can. - His arms should be straightened with his eyes at the HOK. - The ball should move with the back through his track. - If the ball is pulled the decision must be made by the time the ball gets to the QB's front hip. Ball Carrier Mesh Fundamentals - The potential ball carrier is the one that is responsible for his track. He needs to make sure his is at the mesh point that the QB is creating. - He must maintain a pocket that is firm enough to secure the ball but loose enough that the QB can pull the ball if the read is a pull read. - Do not deviate from the track until you fully clear the mesh. - If you do not get the ball then track away from the HOK and make him chase you. Motion Fundamentals - All of our motions start up at FULL SPEED! It is incredibly important to make every motion at full speed so the QB has a consistent timing. - The QB is responsible for sending the motion back on every play except for the triple option. - This allows us to control each motion easier. - Every Jet motion should be timed as if the jet back was getting the ball regardless of play. - The motion back will change his depth based on play. - This allows us to use motion with a multitude of schemes. - The motion stops at different spots on different plays and each perimeter player must understand where he should stop for each play.

Ferris State HC Tony Annese - Mesh (QB/Ball Carrier) & Motion Fundamentals QB Mesh Fundamentals - Regardless of play, the first thing the QB needs to do to make sure he has a good mesh is stab the ball back off his hip towards the ball carrier as far as he can. - His arms should be straightened with his eyes at the HOK. - The ball should move with the back through his track. - If the ball is pulled the decision must be made by the time the ball gets to the QB's front hip. Ball Carrier Mesh Fundamentals - The potential ball carrier is the one that is responsible for his track. He needs to make sure his is at the mesh point that the QB is creating. - He must maintain a pocket that is firm enough to secure the ball but loose enough that the QB can pull the ball if the read is a pull read. - Do not deviate from the track until you fully clear the mesh. - If you do not get the ball then track away from the HOK and make him chase you. Motion Fundamentals - All of our motions start up at FULL SPEED! It is incredibly important to make every motion at full speed so the QB has a consistent timing. - The QB is responsible for sending the motion back on every play except for the triple option. - This allows us to control each motion easier. - Every Jet motion should be timed as if the jet back was getting the ball regardless of play. - The motion back will change his depth based on play. - This allows us to use motion with a multitude of schemes. - The motion stops at different spots on different plays and each perimeter player must understand where he should stop for each play.

James Light

19,145 Aufrufe • vor 8 Monaten

Weekend project: A reusable deployment pipeline defined in C# that can take any aspire project and deploy it to a VM with ssh and docker. - Use of C# >>>> Shell scripts interleaved in YAML 🤮. - The pipeline code that runs over a model. This same code can work for any aspire app without changes. It's a dynamic pipeline template. - It can run locally or in the CI - This was built on top of upcoming APIs in Aspire 9.4 (you can build your own pipeline and deploy anywhere too!) #dotnet #aspire

Weekend project: A reusable deployment pipeline defined in C# that can take any aspire project and deploy it to a VM with ssh and docker. - Use of C# >>>> Shell scripts interleaved in YAML 🤮. - The pipeline code that runs over a model. This same code can work for any aspire app without changes. It's a dynamic pipeline template. - It can run locally or in the CI - This was built on top of upcoming APIs in Aspire 9.4 (you can build your own pipeline and deploy anywhere too!) #dotnet #aspire

David Fowler

23,700 Aufrufe • vor 1 Jahr

GPT-5.6 Sol Ultra was given a full set of character references and asked to rebuild the character in Blender, rig it, create three animations, export a real GLB, and build a complete Three.js showcase around it. The entire pipeline was completed in 2 hours and 11 minutes. The final model still loses some of the original proportions, costume structure, and material fidelity, but going from reference images to a rigged, animated 3D character running in the browser is already a very strong result. Cc: Givros

GPT-5.6 Sol Ultra was given a full set of character references and asked to rebuild the character in Blender, rig it, create three animations, export a real GLB, and build a complete Three.js showcase around it. The entire pipeline was completed in 2 hours and 11 minutes. The final model still loses some of the original proportions, costume structure, and material fidelity, but going from reference images to a rigged, animated 3D character running in the browser is already a very strong result. Cc: Givros

Token Gremlin

37,583 Aufrufe • vor 2 Monaten

A voice agent powered by gpt-oss. Running locally on my macBook. Demo recorded in a Waymo with WiFi turned off. I'm still on my space game voice AI kick, obviously. Code link below. For conversational voice AI, you want to set the gpt-oss reasoning behavior to "low". (The default is "medium".) Notes on how to do that and a jinja template you can use are in the repo. The LLM in the demo video is the big, 120B version of gpt-oss. You can use the smaller, 20B model for this, of course. But OpenAI really did a cool thing here designing the 120B model to run in "just" 80GB of VRAM. And the llama.cpp mlx inference is fast: ~250ms TTFT. Running a big model on-device feels like a time warp into the future of AI.

A voice agent powered by gpt-oss. Running locally on my macBook. Demo recorded in a Waymo with WiFi turned off. I'm still on my space game voice AI kick, obviously. Code link below. For conversational voice AI, you want to set the gpt-oss reasoning behavior to "low". (The default is "medium".) Notes on how to do that and a jinja template you can use are in the repo. The LLM in the demo video is the big, 120B version of gpt-oss. You can use the smaller, 20B model for this, of course. But OpenAI really did a cool thing here designing the 120B model to run in "just" 80GB of VRAM. And the llama.cpp mlx inference is fast: ~250ms TTFT. Running a big model on-device feels like a time warp into the future of AI.

kwindla

202,211 Aufrufe • vor 1 Jahr

Let me share a long-awaited behind-the-scenes video of how Simulon works. It all starts with setting up a scene. You set the location, lighting conditions and capture the light map with a few turns. After that, you can capture the environment mesh using the Lidar scanner. That's the moment the fun begins. You can bring in your #3D models with animations and capture a video of up to 50" using the #AR preview on screen. After that is done, you send the footage to the cloud and in a couple of minutes you get back an incredible #CGI shot. And this is just the beginning. Follow Gabriele Romagnoli to see how Simulon integrates with #Blender and in a full #VFX workflow. A special thanks to Divesh Naidoo for sharing early access and for building something 🤯

Let me share a long-awaited behind-the-scenes video of how Simulon works. It all starts with setting up a scene. You set the location, lighting conditions and capture the light map with a few turns. After that, you can capture the environment mesh using the Lidar scanner. That's the moment the fun begins. You can bring in your #3D models with animations and capture a video of up to 50" using the #AR preview on screen. After that is done, you send the footage to the cloud and in a couple of minutes you get back an incredible #CGI shot. And this is just the beginning. Follow Gabriele Romagnoli to see how Simulon integrates with #Blender and in a full #VFX workflow. A special thanks to Divesh Naidoo for sharing early access and for building something 🤯

Gabriele Romagnoli

142,698 Aufrufe • vor 2 Jahren

This guy is lucky to still be alive. In this crazy video from Mexico, a mountaineer records buzzing from a corona discharge (a form of St. Elmo’s Fire), and it means a lightning strike is imminent. Extreme voltage builds up between cumulonimbus clouds and the peak. The buzzing is the sound of electricity actively leaking into the atmosphere (often accompanied by tingling skin or hair standing on end), and is the final warning sign that lightning is about to strike that exact spot. The buzzing at the end of the video is quite extraordinary. Original video link:

This guy is lucky to still be alive. In this crazy video from Mexico, a mountaineer records buzzing from a corona discharge (a form of St. Elmo’s Fire), and it means a lightning strike is imminent. Extreme voltage builds up between cumulonimbus clouds and the peak. The buzzing is the sound of electricity actively leaking into the atmosphere (often accompanied by tingling skin or hair standing on end), and is the final warning sign that lightning is about to strike that exact spot. The buzzing at the end of the video is quite extraordinary. Original video link:

Iain Cameron

496,748 Aufrufe • vor 1 Monat

We wanted to show the performance and the insane power of the Static Hermes native FFI. But we don't yet support the all language features needed to compile React Native (soon though!), so we improvised. We imported the bindings for DearImGui (a popular C++ imperative game UI library that renders meshes), and Sokol (an amazing lightweight 3D backend in C) and wrote a small demo. "Imperative UI" means that the entire UI and the state of the screen is recreated on every frame, which suits our purposes to demonstrate SH very well. The demo runs 100% compiled native code. JavaScript runs in every frame and recreates the entire state of the screen. It is portable to all platforms, though the screen layout is not optimized for phone. We wrote the demo in C++ and in JS, to compare, and they both ran with the same performance. In the attached videos you can see the demo running at the stable screen refresh rate - 120Hz on my Mac, 60Hz on a old Moto G5 phone. It includes some widgets, an n-body simulation of planets, a playable game demo, etc.

We wanted to show the performance and the insane power of the Static Hermes native FFI. But we don't yet support the all language features needed to compile React Native (soon though!), so we improvised. We imported the bindings for DearImGui (a popular C++ imperative game UI library that renders meshes), and Sokol (an amazing lightweight 3D backend in C) and wrote a small demo. "Imperative UI" means that the entire UI and the state of the screen is recreated on every frame, which suits our purposes to demonstrate SH very well. The demo runs 100% compiled native code. JavaScript runs in every frame and recreates the entire state of the screen. It is portable to all platforms, though the screen layout is not optimized for phone. We wrote the demo in C++ and in JS, to compare, and they both ran with the same performance. In the attached videos you can see the demo running at the stable screen refresh rate - 120Hz on my Mac, 60Hz on a old Moto G5 phone. It includes some widgets, an n-body simulation of planets, a playable game demo, etc.

Tzvetan Mikov

826,205 Aufrufe • vor 2 Jahren

Requiring skin in the game, why shared investment builds trust and filters quality | Markus explains "But I would prefer to see some skin in the game. So, like, if there is a 1,000,000 investment or a project that demands 1,000,000, then we shouldn't give them 1,000,000, but something like 700,000 and 300,000 needs to come from the project owner itself. So, skin in the game. That gives me the trust that he really believes in his project. This is, in my opinion, one of the strongest free filters and selections, so you don't end up with 1,000 proposals, maybe just 50." 🧵 #CardanoShorts

Requiring skin in the game, why shared investment builds trust and filters quality | Markus explains "But I would prefer to see some skin in the game. So, like, if there is a 1,000,000 investment or a project that demands 1,000,000, then we shouldn't give them 1,000,000, but something like 700,000 and 300,000 needs to come from the project owner itself. So, skin in the game. That gives me the trust that he really believes in his project. This is, in my opinion, one of the strongest free filters and selections, so you don't end up with 1,000 proposals, maybe just 50." 🧵 #CardanoShorts

Cardano Community

13,469 Aufrufe • vor 1 Jahr

✨ I made a little demo of #vibejam sponsor Tripo and how you can use it to generate assets for your games In this case I'm generating an apartment flat that's half destroyed and open for my drone sim to fly around, take cover and shoot at enemies from First I write the prompt and select Nano Banana to generate an image first, I get a few different ones, and I select the one I like Then that one is converted into a 3d object called a Smart Mesh Smart Mesh is different than regular image-to-3d models because usually you get issues with uneven polygon distribution, too dense geometry and poor edge flow, so while you think you save time with generating it with AI you end up doing lots of editing after to fix the topology which kinda defeats the point Smart Mesh doesn't have any of this, it's fast and clean low-poly models you can drop into your game Anyway, after generating the model I can then see it in 3d, and if I don't like it, regenerate it, or if I like it start texturing it Once that's done, I export it as a GLB, FBX or OBJ and drop it in (usually I use GLB as ThreeJS seems to prefer that), in my case I just drop the URL into Cursor and let it figure out the placement of the asset into my game!

✨ I made a little demo of #vibejam sponsor Tripo and how you can use it to generate assets for your games In this case I'm generating an apartment flat that's half destroyed and open for my drone sim to fly around, take cover and shoot at enemies from First I write the prompt and select Nano Banana to generate an image first, I get a few different ones, and I select the one I like Then that one is converted into a 3d object called a Smart Mesh Smart Mesh is different than regular image-to-3d models because usually you get issues with uneven polygon distribution, too dense geometry and poor edge flow, so while you think you save time with generating it with AI you end up doing lots of editing after to fix the topology which kinda defeats the point Smart Mesh doesn't have any of this, it's fast and clean low-poly models you can drop into your game Anyway, after generating the model I can then see it in 3d, and if I don't like it, regenerate it, or if I like it start texturing it Once that's done, I export it as a GLB, FBX or OBJ and drop it in (usually I use GLB as ThreeJS seems to prefer that), in my case I just drop the URL into Cursor and let it figure out the placement of the asset into my game!

@levelsio

72,338 Aufrufe • vor 4 Monaten

Big news I found a way to convert Gaussian Splatting into a standard mesh while preserving all the visual advantages of Gaussian Splatting, such as reflections, shadows, and lighting that changes depending on the viewing angle. The result is excellent performance, even on an old laptop or a VR headset. And of course, the mesh can be animated like any regular 3D mesh, while still keeping the visual benefits of GS. I’ll soon show a full-body character animated using this technique.

Big news I found a way to convert Gaussian Splatting into a standard mesh while preserving all the visual advantages of Gaussian Splatting, such as reflections, shadows, and lighting that changes depending on the viewing angle. The result is excellent performance, even on an old laptop or a VR headset. And of course, the mesh can be animated like any regular 3D mesh, while still keeping the visual benefits of GS. I’ll soon show a full-body character animated using this technique.

Alex

38,119 Aufrufe • vor 3 Monaten

A fun way to try the new Claude Fable. I asked it to train a bot using RL to navigate 3D worlds created with SpAItial AI. This is all running in the browser including RL training! Link to demo and code in comments What you can do: -> Train a bot from scratch, watch it learn in minutes, then drop it into a world it has never seen and it keeps navigating obstacles. -> The worlds are real scenes generated from a text prompt or a photo via the SpAItial AI API (you get a splat + a collision mesh). -> Mesh is made into a 2D walkability + height grid that the RL environment runs on. -> The trick to generalization: the policy only ever sees egocentric inputs. A 16-ray lidar, the goal's body frame, and its speed. No global coordinates, no map. So it learns navigation, not a memorized world.

A fun way to try the new Claude Fable. I asked it to train a bot using RL to navigate 3D worlds created with SpAItial AI. This is all running in the browser including RL training! Link to demo and code in comments What you can do: -> Train a bot from scratch, watch it learn in minutes, then drop it into a world it has never seen and it keeps navigating obstacles. -> The worlds are real scenes generated from a text prompt or a photo via the SpAItial AI API (you get a splat + a collision mesh). -> Mesh is made into a 2D walkability + height grid that the RL environment runs on. -> The trick to generalization: the policy only ever sees egocentric inputs. A 16-ray lidar, the goal's body frame, and its speed. No global coordinates, no map. So it learns navigation, not a memorized world.

RO2⚡️

12,797 Aufrufe • vor 3 Monaten

Microsoft open-sourced a 4B model that turns any image into a production-ready 3D asset in 3 seconds. It’s called TRELLIS.2, a fully textured, physically accurate 3D models with PBR textures out of the box. → Full PBR (base color, roughness, metallic, opacity) → Handles hair, cloth, glass, non-manifold geometry → Exports .glb ready for Unity/Unreal/Blender → Runs locally, ships in 3 seconds It's not a demo or a research preview. The full training codebase is public. You can fine-tune it on your own asset library and get a model that generates in your studio's exact style. 100% Open Source

Microsoft open-sourced a 4B model that turns any image into a production-ready 3D asset in 3 seconds. It’s called TRELLIS.2, a fully textured, physically accurate 3D models with PBR textures out of the box. → Full PBR (base color, roughness, metallic, opacity) → Handles hair, cloth, glass, non-manifold geometry → Exports .glb ready for Unity/Unreal/Blender → Runs locally, ships in 3 seconds It's not a demo or a research preview. The full training codebase is public. You can fine-tune it on your own asset library and get a model that generates in your studio's exact style. 100% Open Source

Superman

283,528 Aufrufe • vor 1 Monat

This is what AI subscription companies are afraid of. A guy with a screwdriver, a $700 pile of components, and one free weekend. They built their entire business model on you never figuring out that the same models run locally. On hardware you own. No rate limits. No data leaving the room. No invoice at the end of the month. He figured it out on a Saturday. By Sunday night six agents were online. Research, writing, QA, publishing, monitoring. All running locally for $23 a month in electricity. By the following month he had eight clients paying $2,500 each to run the same system for their businesses. $20,000 a month. Built on a PC he assembled himself in a weekend. They're still sending him renewal emails. He stopped opening them.

This is what AI subscription companies are afraid of. A guy with a screwdriver, a $700 pile of components, and one free weekend. They built their entire business model on you never figuring out that the same models run locally. On hardware you own. No rate limits. No data leaving the room. No invoice at the end of the month. He figured it out on a Saturday. By Sunday night six agents were online. Research, writing, QA, publishing, monitoring. All running locally for $23 a month in electricity. By the following month he had eight clients paying $2,500 each to run the same system for their businesses. $20,000 a month. Built on a PC he assembled himself in a weekend. They're still sending him renewal emails. He stopped opening them.

Superior

23,036 Aufrufe • vor 2 Monaten

This is Segment Anything for #GaussianSplatting running on the web... but it comes with a caveat. This is SplatEdit, a powerful #3DGS editing software built on top of SuperSplat. Last week, it added a segmentation selection feature powered by Segment Anything, running locally in the browser. Just upload a splat, click on the "mug" icon and select entire objects in the scene. The problem is that everything after the object is also selected (as you can see in the video). To fix that, just use the brush tool to subtract the unwanted portion of the scene (hold CTRL while brushing). I feel we are getting closer and closer to having more editing control on these amazing fuzzy clouds 😉

This is Segment Anything for #GaussianSplatting running on the web... but it comes with a caveat. This is SplatEdit, a powerful #3DGS editing software built on top of SuperSplat. Last week, it added a segmentation selection feature powered by Segment Anything, running locally in the browser. Just upload a splat, click on the "mug" icon and select entire objects in the scene. The problem is that everything after the object is also selected (as you can see in the video). To fix that, just use the brush tool to subtract the unwanted portion of the scene (hold CTRL while brushing). I feel we are getting closer and closer to having more editing control on these amazing fuzzy clouds 😉

Gabriele Romagnoli

14,767 Aufrufe • vor 2 Monaten

Here's what's coming next for the Fondant Custom Normals Addon: Adjustable 3D anime face proxy mesh! If you've tried the proxy method before, you know that it can be difficult to get good results as you need the right topology and shape for good Normals, and also for it to fit to the face properly for a clean transfer. The solution is to use two proxies: one for the shape, and one for the fitting. But that gets cumbersome! And it is still often unintuitive to get the correct shading shapes. The next major feature of the addon will be aiming to solve these problems by providing a convenient proxy mesh that you can fit to your character's face with some rig controls, and then handles the rest under the hood with Geometry Nodes. All the benefits of complexity without having to deal with it yourself! This clip shows the general idea, but is early stages. I'll be sharing more as I build it out.

Here's what's coming next for the Fondant Custom Normals Addon: Adjustable 3D anime face proxy mesh! If you've tried the proxy method before, you know that it can be difficult to get good results as you need the right topology and shape for good Normals, and also for it to fit to the face properly for a clean transfer. The solution is to use two proxies: one for the shape, and one for the fitting. But that gets cumbersome! And it is still often unintuitive to get the correct shading shapes. The next major feature of the addon will be aiming to solve these problems by providing a convenient proxy mesh that you can fit to your character's face with some rig controls, and then handles the rest under the hood with Geometry Nodes. All the benefits of complexity without having to deal with it yourself! This clip shows the general idea, but is early stages. I'll be sharing more as I build it out.

aVersionOfReality

107,211 Aufrufe • vor 1 Jahr

I’m running for Congress to fight for the community that raised me. As an Army Combat Veteran and one of the first women in Special Operations leadership, I’ve never backed down from a tough fight. Our country is at a crossroads. I’m running for Congress to fight for you. Check out our first ad of the campaign. ⬇️

I’m running for Congress to fight for the community that raised me. As an Army Combat Veteran and one of the first women in Special Operations leadership, I’ve never backed down from a tough fight. Our country is at a crossroads. I’m running for Congress to fight for you. Check out our first ad of the campaign. ⬇️

Cait Conley

14,174 Aufrufe • vor 4 Monaten

Running Gemma 4 12B locally at 28T/s. Facing issues with vision in LM Studio, but text performance is quite solid. Asked it to create a landing page for a tech event. Video is fast forwarded for the demo.

Running Gemma 4 12B locally at 28T/s. Facing issues with vision in LM Studio, but text performance is quite solid. Asked it to create a landing page for a tech event. Video is fast forwarded for the demo.

AshutoshShrivastava

15,697 Aufrufe • vor 3 Monaten

I gave GPT-5.6 Sol Ultra a full set of character references: front T-pose, four-side turnaround, idle, walk and attack poses. The goal was not just to generate a nice mage image. It had to rebuild the character in Blender, rig it, create 3 animations, export a real GLB and then build a complete Three.js showcase around it. The full goal was completed in 2h11. The pipeline works: the model loads in the browser, the camera can orbit around it, Idle / Walk / Attack are playable, and the page displays the real mesh, material and animation data. But the before/after also shows exactly where image-to-3D still struggles. The overall identity is there, but the proportions became much more chibi and blocky than the original reference. The hat lost its clean folded silhouette and turned into a striped curved tube. The face and eyes are too large and too bright, while the original had a darker hood with much smaller cyan eyes. The robe layers, shoulder shapes and front ornaments were also simplified too aggressively. The back is clearly the weakest part: the cape became a large smooth cone, most of the costume structure disappeared, and the head turns into a glowing purple sphere instead of preserving the dark hooded silhouette. Some gold trims are uneven, the hands are still rough, several parts intersect, and the materials look much more saturated and plastic than the reference. So no, this is not a perfect “one prompt 3D character” result. But going from a few generated references to a rigged, animated GLB running inside a custom web showcase in 2h11 is already a pretty interesting baseline. Now the real work starts: fixing the silhouette, rebuilding the back, cleaning the materials and improving the animations.

I gave GPT-5.6 Sol Ultra a full set of character references: front T-pose, four-side turnaround, idle, walk and attack poses. The goal was not just to generate a nice mage image. It had to rebuild the character in Blender, rig it, create 3 animations, export a real GLB and then build a complete Three.js showcase around it. The full goal was completed in 2h11. The pipeline works: the model loads in the browser, the camera can orbit around it, Idle / Walk / Attack are playable, and the page displays the real mesh, material and animation data. But the before/after also shows exactly where image-to-3D still struggles. The overall identity is there, but the proportions became much more chibi and blocky than the original reference. The hat lost its clean folded silhouette and turned into a striped curved tube. The face and eyes are too large and too bright, while the original had a darker hood with much smaller cyan eyes. The robe layers, shoulder shapes and front ornaments were also simplified too aggressively. The back is clearly the weakest part: the cape became a large smooth cone, most of the costume structure disappeared, and the head turns into a glowing purple sphere instead of preserving the dark hooded silhouette. Some gold trims are uneven, the hands are still rough, several parts intersect, and the materials look much more saturated and plastic than the reference. So no, this is not a perfect “one prompt 3D character” result. But going from a few generated references to a rigged, animated GLB running inside a custom web showcase in 2h11 is already a pretty interesting baseline. Now the real work starts: fixing the silhouette, rebuilding the back, cleaning the materials and improving the animations.

Givros

24,498 Aufrufe • vor 2 Monaten