Ludo.ai
Ludo.ai
Pricing
Features
Blog
Docs
What's New
Contact

You've got questions? We've got answers!

Explore our comprehensive documentation for in-depth information about Ludo.ai and its powerful features.
API & MCPAudio GeneratorSprite GeneratorVideo Generator3D Asset GeneratorFAQUser Account and SubscriptionProjectImage GeneratorGame IdeatorAsk Ludo

3D Asset Generator

  • Introduction

    The 3D Generator is a powerful tool within the Ludo.ai platform that allows you to create 3D assets for your games using text prompts, images, or a combination of both. It leverages an advanced AI model to generate models with PBR (Physically Based Rendering) textures, including albedo, metallic, and roughness maps. Generated (or uploaded) models can then be rigged with a skeleton and animated.

    Key Features:

    • Text-to-3D: Create models directly from descriptive text prompts.
    • Image-to-3D: Transform 2D images into 3D models.
    • Texture Generation: Add new PBR textures to your own models or change the texture of previously generated models.
    • Rigging: Add a skeleton and skin weights to any model — characters, creatures, props, plants — so it can be posed and animated. Choose between a general-purpose rig and standard humanoid skeletons, and export the bones under the joint naming convention your engine or 3D editor expects (Mixamo, Unreal, Godot, and more).
    • Animation from presets: Apply a ready-made animation from Ludo's preset library — walk cycles, attacks, idles, reactions — directly onto a rigged humanoid character. These are real, full-length motions retargeted onto your model's own skeleton, so they are reliable and game-ready. This is the route to try first.
    • Animation from prompts (experimental): Describe a motion in plain words and get short, custom animation clips. Works on any rigged asset — characters, creatures, props, doors, plants — but it is hit or miss.
    • Interactive 3D Viewer: View your generated models from any angle in a fully interactive 3D space, inspect the skeleton, pose it by hand, and play saved animations.
    • Download Options: Download models in various formats (GLB, STL, PLY, OBJ, OFF), with rigging and saved animations included in the GLB.
    • Favorites System: Save your best creations for easy access.
    • In-Platform Image Editor: Modify your input images before 3D transformation (accessible via "Open in Editor").

  • Getting Started

    Creating your first 3D model in Ludo.ai takes just a few steps.

    1. Access the 3D Generator: Log into your Ludo.ai account and select "3D Asset Generator" from the main menu.
    2. Choose a Generation Mode: Select the tab for the type of generation you want to perform: "Text to 3D", "Image to 3D", "Texture Generation", or "Rig & Animate".
    3. Provide Input:
      • Text to 3D: Enter a text description of the 3D model you want to create.
      • Image to 3D: Click "Choose reference image" and upload or choose an image. For best results, use a clear image of a single subject.
      • Texture Generation: Choose a reference model and describe the new texture you want to apply.
      • Rig & Animate: Choose a reference model (generated or uploaded), then pick a Rig type and a Joint naming convention to add a skeleton to it.
    4. Set Advanced Options (Optional): Before generating, you can adjust the Advanced options panel to control mesh complexity (max triangles, adaptive decimation) and texture quality (type, resolution). This allows you to create models that are optimized for your game's performance needs, from simple low-poly shapes to more detailed assets.
    5. Click "Start New Generation": Ludo.ai will process your input and create the 3D model.
    6. View, Refine, and Download: Use the interactive viewer to inspect your model, then download it or use the contextual options to refine it further.

    Working on an existing model: rigging and animation are steps you apply to a model you already have. Every 3D model in your history has an action bar with "Rig" (or "Re-rig" and "Animate" once it is rigged), alongside "Optimize Model", "Re-texture" and "Generate New Model". A typical full workflow is: generate a model → rig it → animate it → download the GLB.

  • Text-to-3D

    This mode creates a 3D model based solely on your text input. The process involves two steps: first generating a 2D concept image from your prompt, and then converting that image into a 3D model.

    How to Use

    1. Select the "Text to 3D" tab.
    2. Enter your text prompt in the input box. Be as specific as possible.
    3. Click "Start New Generation".
    4. Ludo.ai will generate a 2D preview image based on your prompt. If you're not satisfied, you can edit the prompt and click "Start New Generation" again.
    5. Once you have a 2D image you like, click the "Generate 3D Object" button on that image to begin the conversion.
    6. The final 3D model will be generated. You can then view, refine, or download the result.

    Example Prompts

    • "A low-poly, cartoonish treasure chest with gold trim and a skull emblem."
    • "A futuristic spaceship with glowing blue engines and a sleek, metallic hull."
    • "A black cartoon lego stormtrooper"
    • "A whimsical mushroom house with glowing windows."

    Tips

    • The prompt should focus on a single subject. The generator cannot generate scenes or environments.
    • Use adjectives to describe the desired shape, color, material, style, and any specific features. For example, instead of "sword," try "ornate elven sword with a glowing blue blade and a golden hilt."
    • Experiment with different wording. Rephrasing or adding/removing keywords can significantly impact the initial 2D image and the final 3D model.

  • Image-to-3D

    This mode uses a 2D image as the primary input for creating a 3D model. The AI analyzes the image, automatically removes the background, and attempts to extrapolate its 3D form.

    How to Use

    1. Select the "Image to 3D" tab.
    2. Click "Choose reference image" and upload a 2D image from your computer or choose one from your favorites in Ludo.
    3. (Optional) Click "Open in Editor" to make adjustments to the image before generating the 3D model. You can use masking tools and text prompts to guide the AI's edits.
    4. Click "Generate".
    5. Ludo.ai will process the input image, remove the background, and then generate a 3D model based on it.
    6. View, refine, or download the 3D model.

    Best Practices

    • Use images with a single, clearly defined subject. The object should be fully visible.
    • The subject can be over a background (it will be automatically removed), but there should be a clear distinction between the subject and background for best results.
    • The AI will not be able to process images of scenes or environments, and the results will be unpredictable.
    • Your uploaded images should have a resolution of at least 512x512 for optimal quality.
    • Use the "Open in Editor" feature to remove or change unwanted elements before generating the 3D model.

  • Texture Generation

    This generation mode allows you to upload your own 3D asset or choose one of your previously generated assets, and generate new PBR textures for it. The mesh (shape) of the original model will not be modified.

    How to Use:

    1. Select the "Texture Generation" tab.
    2. Click on "Choose reference model" to select a 3D model, either one you've uploaded or a previous generation.
    3. Enter a text prompt describing the desired texture in the "Describe the model texture here" box.
    4. Note that the texture of the input model (if any) will be ignored. Your prompt should describe the new texture, not modifications to the existing one.
    5. Click "Start New Generation".
    6. Ludo.ai will generate a new 3D model, keeping the original mesh but applying the new texture.

    Tips:

    • For best results, your prompt should describe both the asset and its texture. For example: "A cartoonish dinosaur with purple and green skin."
    • However, describing only the texture should still produce good results. For example: "purple and green skin."
    • Be specific and use adjectives to define color, material, and style.

  • How to Upload 3D Models

    The "Texture Generation" and "Rigging" modes allow you to upload your own 3D assets — to re-texture them, or to add a skeleton to them. The uploaded 3D files must contain a single subject. Files with multiple models or entire scenes will produce unexpected and low-quality results.

    Limitations

    We have the following restrictions on the 3D models that can be uploaded:

    • File Types: obj, ply, stl, 3ds, gltf, glb, fbx, off, dae, ctm, x3d.
    • File Content: The file should contain only the subject 3D model.
    • Mesh Complexity: For best performance, meshes should have fewer than 100,000 vertices.
    • File Size: Maximum file size is 20MB.

    Preparing Your Model for Upload (Using Blender)

    You can use a free tool like Blender to reliably prepare and export your model in a supported format:

    1. Open your model in Blender (File > Open or File > Import > ...).
    2. In the editor, delete any undesirable objects (like cameras, lights, or background elements), keeping only the main subject.
    3. Export the model to the glTF format (File > Export > glTF 2.0 (.glb/.gltf)).
    4. In the export options panel (usually on the right), apply the following settings:
      • Include > Limit to > Check the Renderable Objects box.
      • Transform > Check the +Y Up box.
      • Data > Materials > Materials > Select No Export.
      • Data > Uncheck Shape Keys, Skinning, and Compression.
      • Animation > Uncheck Use Animation.
    5. Select your file path and click the export button.

    Any existing skeleton in the uploaded file is not used: rigging always generates a new one from the mesh.

  • Re-topology

    The Optimize Model option (also known as re-topology) allows you to change the complexity of an existing 3D model, optimizing it for your needs. You can simplify a high-poly mesh or refine a low-poly one. You can also adjust the type and resolution of texture maps.

    The new object will have a shape and texture that is very similar to the original one, but with adjusted complexity and textures.

    To use it, click the "Optimize Model" button (wrench icon) on a 3D generation, choose the new options in the dialog (Max triangles, Adaptive decimation, Texture type, Resolution — see the "Advanced Generation & Optimize Model Options" section for details), and click "Process".

    When finished, the new result will replace the original. You can do this process again for further adjustments. Note that the data of the object is never destroyed during this process. You can re-topology to a low triangle count and then to a higher triangle count and it will still work as expected.

    Rigged and animated models: Optimize Model preserves them. The skeleton and skin weights are adapted to the new mesh, and any animations you saved for that model still play. It is therefore safe to rig and animate a model first and adjust its triangle count afterwards. Re-texturing keeps the rig and animations as well.

  • Advanced Generation & Optimize Model Options

    To give you precise control over the quality and performance of your 3D models, the 3D Generator features an "Advanced" options panel. You can find this panel below the main prompt area. These settings allow you to define the complexity of the model's geometry (mesh) and the quality of its textures, ensuring the generated asset is optimized for your game engine or use case.

    The panel is divided into two main categories:

    Mesh Quality

    This group of settings controls the geometric complexity of your 3D model. Mesh Quality is only available for the "Text to 3D" and "Image to 3D" tabs — in "Texture Generation" the original mesh is preserved, so these controls are hidden.

    • Max triangles: Sets the maximum number of triangles (polygons) for the final mesh. The slider ranges from 1,000 to 200,000. A lower number (e.g., 5,000) creates a low-poly model ideal for performance, while higher values allow for more complex and detailed shapes.
    • Adaptive decimation: This slider (range 0.0 to 1.0) controls how aggressively the AI simplifies the mesh.
      • A value of 0.0 disables it, and the AI will aim for the "Max triangles" value.
      • Higher values allow the AI to tolerate more simplification, resulting in a mesh with fewer triangles while preserving detail.

    Texture

    These settings define the quality and type of the texture maps generated for your model.

    • Type: (Default: PBR) Determines the kind of texture maps to create:
      • PBR: Generates a full set of Physically Based Rendering (PBR) maps (albedo, metallic, roughness). This is best for realistic materials.
      • Color: Generates an albedo (color) map only.
      • None: Generates no texture, resulting in a plain, untextured mesh. This will result in a significantly faster generation.
    • Resolution: Sets the size of the generated texture maps. You can choose between 1024 (1K) and 2048 (2K). Higher resolutions provide sharper details but result in larger file sizes.

    How to Use These Options

    These settings can be used in two different workflows:

    1. For New Generations (Text-to-3D, Image-to-3D, Texture Generation): Set these options before you click "Start New Generation." This ensures the model is created with your desired complexity and texture settings from the start, saving you time on optimization later.

    2. For Optimize Model: The "Optimize Model" option applies these settings to a previously generated model. Choose your reference model, click the "Optimize Model" button (wrench icon) on a generation, adjust the settings to your new requirements, and click "Process." This will create a new version of your model with the updated topology and texture resolution, which is perfect for creating different Levels of Detail (LODs) for your game.

  • Viewing and Interacting with Results

    After generating a 3D model, it will appear in the results area with several interaction options:

    • Asset Preview: The main preview area displays the generated asset.
    • View Modes (Toggle): Click the small thumbnail in the corner of the preview to switch between:
      • Video View: A rotating video preview of the rendered, textured asset.
      • Image View: The 2D source/concept image used for the generation.
      • Clicking on the rendered model itself opens the interactive 3D viewer in a full-screen overlay (see below).
    • Animation Badge: Rigged models with saved animations show a small counter under the view toggle. Clicking it opens the Animate window for that model.
    • Add to Favorites (Heart): Saves the current model to your Ludo.ai Favorites.
    • Contextual Menu (...): Provides:
      • Download 3D Object: Opens a dialog to download the current 3D model in your chosen format (GLB, STL, PLY, OBJ, OFF).
      • Delete: Removes the asset from your generation history.
    • Action Bar (below the asset): Quick actions on the current model:
      • Optimize Model: Opens the Optimize Model dialog to change mesh complexity and texture settings (see the "Re-topology" section).
      • Animate: Available on rigged models — opens the animation window (see "Animating 3D Models").
      • Rig / Re-rig: Rig appears on models without a skeleton and Re-rig on rigged ones; both open the "Rig & Animate" tab with the model pre-selected (see "Rigging").
      • Re-texture: Opens the Texture Generation tab with the current model pre-selected.
      • Generate New Model: Pre-populates the form with the same source image and advanced parameters, ready to launch a new generation with a different result.
      • Create 3D Model: Available on results that are still 2D images — runs the image-to-3D conversion on the current concept.
      • Edit Image: Available on 2D results — opens the integrated Image Editor to refine the source image before generating a 3D model.

    The Interactive 3D Viewer

    Click a rendered model to open it full screen. There you can click-and-drag to rotate the model freely, and use the controls in the corner:

    • Wire-frame toggle — inspect the model's underlying mesh and topology (the triangle count is also shown).
    • Skeleton toggle — only on rigged models. Draws the bones over the mesh.
    • Pose — only while the skeleton is shown. Drag a joint and the mesh deforms with it, which is the quickest way to check whether a rig is any good. Posing is a preview only: nothing is saved, and Reset pose returns the model to its rest pose.
    • Animation selector — lists the model's saved animation clips. Pick one to play it on the model; choose "No animation" to go back to the static pose.

  • Downloading Models

    Ludo.ai's 3D Generator allows you to download your generated models in several popular 3D file formats:

    • GLB: A binary format for glTF that includes the model's geometry and PBR textures (albedo, metallic, roughness). If the model is rigged, the GLB also contains its skeleton and skin weights, and every animation you have saved for it. This is the recommended format for use in modern game engines and 3D software.
    • STL: A common format for 3D printing, containing only surface geometry (no color or texture information).
    • PLY: A flexible format that can store color and other data per vertex, often used for 3D scanned data.
    • OBJ: A widely supported legacy format for representing 3D geometry.
    • OFF: A simple text-based format for describing polyhedral objects.

    Important Note: Only the GLB format will include the PBR textures you see in the viewer, along with any rigging and animations. All other formats will export only the mesh (the shape of the model without colors, textures, skeleton or animations).

    Animations: only clips you have saved in the Animate window are packed into the GLB — unsaved candidates are ignored. You can download from the Animate window ("Download animated GLB") or from the model's Download option; both produce the same file. The download dialog tells you what the GLB will contain ("Includes textures", "Includes textures and rigging", or "Includes textures, rigging and animations").

    When choosing a format, consider the following:

    • For use in game engines (like Unity, Unreal, Godot) or 3D software: GLB is the best choice.
    • For 3D printing: STL is the most widely accepted format.

    Note on Unity Integration: To directly import GLB files with PBR materials into a Unity project, you will likely need to install the Unity glTFast package from the Unity Package Manager. You can find it here.

  • Tips for Effective 3D Generation

    To get the most out of the 3D Generator, it's important to use the right workflow for your goal. Here are some best practices:

    1. Start with a Single, Clear Subject The generator is designed to create individual assets, not complex scenes. For all generation modes, ensure your prompt or input image focuses on one primary object (e.g., "a treasure chest," "a character model," "a sci-fi rifle").

    2. Master the "Text-to-3D" Two-Step Flow Remember that Text-to-3D is a two-step process: Prompt -> 2D Image -> 3D Model. The quality of your final 3D model is highly dependent on the 2D image it's based on.

      • If the 2D image isn't right (e.g., it's at a bad angle, or the design is wrong), don't click "Create 3D Model" yet.
      • Instead, edit your text prompt and click "Start New Generation" again to get a better 2D concept first.
      • If you are overall happy with the image but want to adjust details, you can also refine it with the "Edit Image" action.

    3. Optimize Your "Image-to-3D" Inputs When using Image-to-3D, your source image is the blueprint.

      • Use clear, high-resolution images (at least 512x512) where the subject is distinct from the background.
      • Use the "Open in Editor" feature in the reference-image selector to clean up your image, remove unwanted background elements, or make adjustments before generating the 3D model.

    4. Use the Right Tool for Iteration Use the specialized actions to refine your model:

      • To change the texture only: Use the "Re-texture" action (or the "Texture Generation" tab). This will keep your mesh and apply a new PBR texture based on your prompt.
      • To get a new shape and texture from the same image: Use the "Generate New Model" action. This pre-populates the form with the same source image and advanced parameters, so you can quickly run another generation — great for getting a few options from one concept.
      • To optimize the model: Use the "Optimize Model" action. This is the perfect tool to change the mesh complexity and texture size of an existing model to make it game-ready.

    5. Use "Advanced" Options for Game-Ready Assets Don't ignore the "Advanced" panel. Set these options before generating to save time on optimization.

      • For Performance: Set "Max triangles" to a lower value (e.g., 10,000) and use "Adaptive decimation" to create efficient, low-poly assets.
      • For Quality: Increase "Resolution" to 2048 for hero assets.
      • For Engine Needs: If your engine doesn't need PBR, switch "Texture" "Type" to "Color" (for albedo only) or "None" (for a plain mesh).

    6. Get the Model Right Before You Rig and Animate Rigging and animation build on the mesh you already have, so settle the shape first — a mesh problem cannot be fixed by a better rig or a better prompt.

      • Generate the model in a neutral pose. For a humanoid that means an A-pose or a T-pose, whichever suits the character. A model generated mid-action (arms crossed, sword raised, leg lifted) rigs worse and drags that pose into all of its animations, since every clip starts from the model's rest pose.
      • Keep the limbs apart — this is the big one. The arms must not touch the body and the legs must not touch each other. Where two parts touch they become a single connected surface in the mesh, so the rigger can't separate them and any animation that moves one will drag or tear the other. The same applies to a tail against a leg, wings against the back, or a prop held flush against the chest. Prefer a source image with clear daylight between limbs and body — this cannot be fixed later by re-rigging.

    7. Pick the Rig Type That Matches the Asset The rig type is the single most consequential choice in the whole rig-and-animate flow.

      • Complete humanoid? Use "Humanoid" — it is what makes the animation preset library reliable. Remember that "humanoid" is about the body plan, not the species: a robot, an orc or an upright fox all count.
      • Fingers clearly modelled and separated? Use "Humanoid with fingers". If the hands are mittens or blocks, don't — plain "Humanoid" gives a better result.
      • Anything else — a non-humanoid, a humanoid missing limbs, or a humanoid with extras like a tail or wings — use "General".
      • Set Joint naming for your target tool (Mixamo for Unity's humanoid avatars, Unreal for the UE mannequin, Godot, Blender's Rigify, and so on). It changes nothing about the animation — only the bone names in the exported file — but it saves you remapping work on import.
      • Check the rig before animating. Open the viewer, turn on the skeleton, and drag a joint with Pose. Re-rig if it looks wrong — the copy is saved as a new asset, so you never lose the original.

    8. Try a Preset Before You Try a Prompt For a rigged humanoid, the From Preset tab gives you a real, full-length, game-ready motion in one run — walk cycles, attacks, idles, reactions. Prompt-based animation is experimental and short.

      • Use presets whenever your asset is a humanoid and the library has something close to what you need.
      • Use prompts for non-humanoids (creatures, doors, chests, trees, machines) or for motions the library doesn't cover.
      • Animate in small batches when prompting. Every run gives four candidates; save the good ones and generate again for the rest. Ask for one short action per clip.
      • Adjust the asset afterwards if needed. "Optimize Model" and "Re-texture" keep the rig and saved animations, so a rigged, animated asset can still be made lighter for your engine or given a new texture.

    9. Be Descriptive in All Prompts This applies to all modes, not just the initial generation.

      • Initial Prompt: "A low-poly, cartoonish treasure chest with gold trim and a golden skull emblem".
      • Texture Prompt: "A rusty, pitted metal texture with glowing blue magical runes".
      • Animation Prompt: the exception — keep it short and plain ("the lid opening"), since the text is rewritten behind the scenes.

    10. Download the Right Format For game development, always download in GLB format. The GLB file is the only format that packages the model's geometry (mesh) along with its PBR texture maps (albedo, metallic, roughness), its rigging, and its saved animations. Other formats like STL or OBJ will only export the mesh, without any colors or textures.

  • Troubleshooting

    If a 3D generation fails or the result looks wrong, these are the most common causes and fixes.

    Model not generating:

    • Check your internet connection: Ensure you have a stable internet connection, as the 3D Generator requires it to function.
    • Try again: Try generating again with the same inputs.

    Errors with uploaded models:

    • There are limitations around the models you can upload to Ludo. Try preparing/exporting your model before upload using Blender, as described in "How to Upload 3D Models".

    Results are not what you expected:

    • Rephrase your prompt: Try different wording, add more descriptive details, or be more specific about certain aspects.
    • Use a reference image: If you already have a particular design in mind, provide a visual reference for the AI.
    • Iterate: Use the "Generate New 3D Model", "Re-texture", and "Re-topology" features to explore variations and refine the model.

    The rig looks wrong (odd skeleton, mesh tearing when posed):

    • Rigging varies from run to run and can fail. Run it again — each attempt produces a different skeleton, and re-rigging saves the result as a new asset, so nothing is lost.
    • Check you picked the right Rig type. A humanoid template forced onto a body that isn't a complete humanoid — one missing a limb, or with a tail, wings or extra arms — will rig badly. Use General for those.
    • Conversely, if a complete humanoid rigged with General gives you a skeleton that doesn't suit it, re-rig with Humanoid.
    • Only use Humanoid with fingers when the fingers are clearly modelled and separated. On mitten-shaped hands it makes things worse.
    • Use the viewer's skeleton toggle and Pose tool to check a rig before animating it.

    The animation preset gave a poor result:

    • Presets only work on humanoids. If the asset is a quadruped, a vehicle, a prop or a plant, use the From Prompt tab instead.
    • Check the rig type. Presets are most reliable on models rigged as "Humanoid" or "Humanoid with fingers". A humanoid rigged with General can work, but not dependably — Re-rig it with the Humanoid type and try again.
    • Check the rig itself with the Pose tool. A preset applies a full-body motion, so weak skin weights or a misplaced joint are obvious straight away.
    • If the character slides across the floor or drifts, set In place to Yes. If a looping motion jumps at the seam, set Crop loop to Yes.
    • If the clip is trimmed shorter than you expected, set Crop loop to No to keep the full recorded motion.

    The prompt-based animation isn't what I asked for:

    • Prompt-based animation is experimental and hard to steer. Generate another batch — each one gives four different takes.
    • If the asset is a humanoid, check the preset library first. A walk, run, attack or idle from a preset will beat anything you can prompt for.
    • Try different wording. There are no keywords or prompt tricks (your text is rewritten behind the scenes), but a different way of describing the motion sometimes lands better.
    • Check the request is within reach: prompt-based clips are always 16 frames (a little over a second), cover a single action, and always start from the model's static pose. Walk cycles and multi-part animations are not possible — see "Animating with Prompts".
    • If every candidate is distorted or barely moves, the rig or the mesh is the likely cause — re-rig the model, and check whether the moving part is fused to the rest of the body.

    The bone names are wrong for my engine:

    • Joint naming is chosen at rigging time, not at animation time. Re-rig the model with the convention your engine expects (Mixamo, Unreal, Godot, Rigify, HumanIK, VRoid). The skeleton and the animations are unaffected — only the labels change.
    • Naming only applies to joints in the standard humanoid template. Models rigged with General, and any joints outside the template, keep their generic names.

    Slow generation times:

    • Server load: High server traffic can sometimes lead to slower generation times. Your generation will be added to the list of results even if you go to a different page in Ludo or close the browser. Make sure to reload the page if you don't see your result.

    Download issues:

    • Browser settings: Check your browser settings to make sure downloads are allowed and that you know where downloaded files are saved.
    • Download taking too long: Model export can take some time until the model starts to download. Wait at least 10 seconds.
    • Missing rigging or animations: only GLB carries them, and only animations you have saved are included.

    Unexpected errors:

    • Refresh the page: Sometimes a simple page refresh can resolve minor glitches.
    • Log out and back in: If refreshing doesn't work, try logging out of your Ludo.ai account and then logging back in.

    If problems persist, please contact Ludo.ai support or join our Discord server for assistance: https://discord.gg/FmTPyugsrR. When reporting an issue, provide as much detail as possible, including your inputs if relevant, the steps you took leading to the problem, and any error messages you encountered.

  • Use Cases in Game Development

    The 3D Generator fits into game development at several stages:

    • Rapid Prototyping: Quickly create placeholder 3D assets to test gameplay mechanics, level layouts, or overall game feel without investing significant time in detailed modeling.
    • Concept Visualization: Bring your game ideas to life in 3D early in the development process. This helps visualize the game's aesthetic and can be used for pitching or team communication.
    • Asset Customization: Generate variations of existing assets to add diversity to your game world. For example, you could create multiple versions of a base character model with different clothing, accessories, or color schemes.
    • Inspiration and Exploration: Experiment with different styles, themes, and concepts to spark new ideas and push creative boundaries. The 3D Generator can be a powerful tool for brainstorming and exploring uncharted territory.
    • Character and Object Design: Create unique characters, creatures, props, and environmental elements. The AI can generate designs that you might not have considered otherwise.
    • Level Design: Generate 3D models of buildings, structures, or natural formations to populate your game environments. This can be especially helpful for quickly blocking out levels or creating variations of environment assets.

  • Image Editing with the Integrated Image Editor

    Before generating 3D models from images, you can refine your 2D input using Ludo.ai's integrated Image Editor. This powerful tool allows you to make precise adjustments to your images, ensuring they are optimized for the 3D generation process.

    Accessing the Image Editor

    Upload an Image: In the 3D Generator, select the Image-to-3D tab and then click on "Choose reference image" and upload your desired image.

    Open in Editor: Once the image is uploaded, click on the "Open in Editor" button located next to the image preview.

  • 3D Asset Generation in Ask Ludo

    In addition to the dedicated 3D Generator tool, you can begin the 3D asset creation process directly within Ludo.ai's Ask Ludo chat interface (both the standalone version and the one integrated into the Game Concept tool). This workflow is designed for a more conversational and iterative approach to asset ideation.

    The process has been updated for a more controlled, two-step workflow:

    How to Use It:

    1. Access Ask Ludo: Open the Ask Ludo chat interface.
    2. Request a 3D Model: Type a natural language request for the asset you want. Be descriptive to get a better visual.
      • Examples:
        • "Generate a 3D model of a wooden treasure chest."
        • "I need a 3D asset of a futuristic motorcycle."
        • "Create a low-poly 3D model of a fantasy character."
    3. Receive a 2D Image: The AI will process your request and generate a high-quality 2D image of the requested asset, not a 3D model directly. This image serves as a visual concept or "render" of your idea.
    4. Convert to 3D:
      • Interact with the generated image to reveal its context menu.
      • Select the "Generate 3D Asset" option.
      • This action will send the image to the 3D Asset Generator, which will then convert it into a true 3D model that you can view, refine, and download.

    Benefits of this Workflow:

    • Conversational Ideation: Brainstorm and refine the look of your asset conversationally.
    • Faster Visual Iteration: Quickly generate multiple 2D concepts in the chat before committing to the more time-intensive 3D conversion.
    • Seamless Integration: Move smoothly from a chat-based idea to a functional 3D asset within the Ludo.ai ecosystem.

  • Rigging

    Rigging adds a skeleton (a hierarchy of bones) and skin weights (which part of the mesh each bone moves) to a 3D model. A rigged model can be posed, animated with the Animate tool, and exported as a GLB that game engines read as a proper skinned mesh.

    Rigging is not limited to characters. Any asset can be rigged — the AI works out a skeleton from the shape of the mesh:

    • Characters, creatures and monsters
    • A tree or plant, so branches and leaves can sway
    • A treasure chest or door, so the lid or panel can swing open
    • A banner, flag or cloth
    • Props with moving parts — a lamp arm, a crane, a windmill, a claw

    How to Use

    1. Select the "Rig & Animate" tab. You can also click "Rig" in the action bar of any existing 3D model, which opens the tab with that model already selected.
    2. Click "Choose reference model" and pick a previously generated model, one of your favorites, or upload your own file (same formats and limits as in "How to Upload 3D Models").
    3. Choose a Rig type (see below). This is the choice that matters — it decides what kind of skeleton you get and whether the asset can use the animation preset library.
    4. Choose a Joint naming convention (see below). This does not change the skeleton at all; it only affects the bone names in the exported file.
    5. Click "Start New Generation".
    6. When it finishes, the model is replaced by its rigged version and the preview video shows the rig moving. Rigged models get two new actions in their action bar: "Animate" and "Re-rig".

    Rigging only adds a skeleton and weights — the mesh and textures are untouched.

    The Model's Pose Matters

    A neutral pose always rigs better. The AI has to work out where the joints are from the shape of the mesh, and a model caught mid-action — arms crossed, sword raised, leg lifted, crouching — hides exactly the information it needs. It also carries that pose into every animation, since animation starts from the model's rest pose.

    For a humanoid, that means an A-pose or a T-pose — whichever suits the character better. Both work equally well; pick the one your model falls into naturally.

    Two rules matter more than the exact pose:

    • The arms must not touch the body. Keep a clear gap between each arm and the torso, all the way down.
    • The legs must not touch each other. Keep the feet apart, with daylight between the legs.

    Where two body parts touch in the mesh, they are a single connected surface. The rigger cannot tell where one ends and the other begins, so it places joints badly and assigns the shared surface to the wrong bone — and any animation that moves one part will drag, stretch or tear the other. This is the single most common cause of a bad rig, and no amount of re-rigging fixes it: it has to be fixed in the model.

    The same applies to non-humanoids: keep a tail clear of the legs, wings clear of the back, and any prop clear of the body that holds it.

    If your model is posed badly or has fused limbs, it is worth regenerating it — from a source image with clear gaps — before rigging.

    Rig Types

    • General — the AI builds whatever skeleton suits the asset. This is the only option for anything that is not a humanoid: creatures, quadrupeds, vehicles, plants, props, doors, chests. It is also the safest choice for anything unusual.
    • Humanoid — pins a standard humanoid skeleton (22 joints: spine, head, two arms, two legs, no individual fingers). Choose this when the asset is a complete humanoid. This is the rig type that makes the animation preset library work reliably.
    • Humanoid with fingers — the same standard skeleton with individual finger joints (52 joints, five fingers per hand). Choose it only when the fingers are clearly modelled and separate from each other. On a mitten-shaped or blocky hand there is nothing for the finger joints to attach to, and the result is worse than the plain Humanoid rig.

    "Humanoid" does not mean "human". It means the body plan: an upright figure with a head, a torso, two arms and two legs. A fox standing on two legs, a robot, an orc, a cartoon mushroom with arms and legs, an anthropomorphic animal — all of these are humanoids for rigging purposes, and all of them can use the humanoid rig types and the animation presets.

    Choosing between them:

    • Complete humanoid, hands modelled as mittens or simple blocks → Humanoid
    • Complete humanoid with distinct, separated fingers → Humanoid with fingers
    • Incomplete humanoid — a missing arm, a missing leg, no head, a torso that ends in a floating base → General
    • Humanoid with extras — a tail, wings, a second pair of arms, tentacles → General. The humanoid templates have no bones for the extra parts, so those parts end up rigid or attached to the wrong bone.
    • Anything that isn't a humanoid at all → General

    Forcing a humanoid template onto a body that doesn't match it produces a bad rig — the model gets bones it has no geometry for, or geometry with no bones to move it. When in doubt, use General and check the result.

    Joint Naming

    Joint naming has no effect on animation. Neither the preset library nor prompt-based animation cares what the bones are called — they work the same under any convention. It exists purely so the exported GLB drops into your game engine or 3D editor with bone names those tools already recognise, which is what lets them auto-map, retarget, or apply their own animation libraries to your model.

    The skeleton itself is identical in every case; only the labels change.

    • SMPL (default) — the naming used by the SMPL/SMPL-X body model, a standard in research and motion-capture tooling. Fine if you're not importing into a specific game-engine pipeline.
    • Mixamo (mixamorig:Hips…) — what Unity's humanoid auto-mapper and most Blender retargeting add-ons expect. A good default if you plan to use Mixamo animations or Unity's Humanoid avatar system.
    • HumanIK — the same names without the prefix. For Maya, MotionBuilder, and plain FBX pipelines.
    • Unreal (pelvis, spine_01, clavicle_l…) — matches the Unreal Engine mannequin.
    • Godot (Hips, LeftUpperArm…) — matches Godot's SkeletonProfileHumanoid, so Godot retargets on import.
    • Rigify (spine.001, upper_arm.L…) — Blender's Rigify metarig.
    • VRoid / VRM (J_Bip_C_Hips…) — for VRM avatar pipelines.

    Naming only applies to joints Ludo can identify as part of the standard humanoid template. Joints outside it — and rigs built with the General type — keep their generic names.

    Rigging Doesn't Always Work

    Rigging is a generative process, so the result varies from run to run and it can fail. The two common failure modes are:

    • A bad skeleton — joints in odd places, missing limbs, or far more (or fewer) bones than the asset needs.
    • Bad skin weights — parts of the mesh follow the wrong bone, so the model tears or stretches when it moves.

    If that happens, simply run it again. Every run produces a different skeleton, and a second or third attempt often lands on a good one. If a humanoid template keeps failing on an asset, that is usually a sign the asset isn't a clean humanoid — switch to General.

    Checking a Rig Before You Animate

    The fastest way to judge a rig is to try it:

    1. Click the model to open the full-screen 3D viewer.
    2. Turn on the skeleton toggle to see the bones over the mesh.
    3. Turn on Pose and drag the joints. The mesh deforms exactly as it will when animated, so a bad skeleton or bad weights show up immediately (parts moving that shouldn't, limbs tearing).
    4. Use Reset pose to return to the rest pose. Posing is a preview only — nothing is saved and the model itself is never modified.

    If the mesh deforms sensibly here, animations will have something good to work with. This check is especially worth doing before using the animation presets: a preset applies a real, full-body motion, so a weak rig shows up immediately.

    Re-rigging

    Clicking "Re-rig" on an already-rigged model generates a fresh skeleton and saves it as a new asset. The original model — with its rig and any animations you saved for it — is left untouched. This makes it safe to keep re-rigging until you get a skeleton you like, to try a different rig type or joint naming, and to keep more than one rig of the same model.

    Rigging and the Other Tools

    Optimize Model and Re-texture both keep the rig: the skeleton and skin weights carry over to the new model, and animations saved for it still work. You can rig and animate an asset first and still change its mesh complexity or its texture afterwards.

  • Animating 3D Models

    Once a model is rigged, you can animate it. Ludo offers two different routes, and they behave very differently — pick the right one before you spend credits.

    From Preset applies a ready-made motion from Ludo's animation library, retargeted onto your model's own skeleton. It is reliable: the clip is real recorded motion, not a generative guess, so the result is the actual animation played by your character. Clips are full length — long enough for genuine walk cycles and attacks — and each run produces one clip. It only works on humanoid characters.

    From Prompt generates a custom motion from a few words you write. It works on any rigged asset — characters, creatures, props, doors, plants — and can produce motions no preset covers. But it is experimental and hit or miss, clips are about one second long (16 frames), and each run gives you four candidates to choose between.

    Try a preset first. If your asset is a humanoid and the library has a motion close to what you need, that is by far the best result you will get. Fall back to prompts when the asset isn't a humanoid, or when you need something the library doesn't have.

    Requirements

    The model must be rigged first (see the "Rigging" section). Un-rigged models show a "Rig" action instead of "Animate".

    Opening the Animate Window

    Click "Animate" in the action bar of a rigged model (or click the animation counter badge on the model card). A full-screen view opens with the model on the left and the animation form on the right. The form has two tabs — From Preset and From Prompt — described in their own sections below.

    Everything below the tabs is shared between the two routes: the saved-animation list, and the download button.

    Saved vs. Unsaved Clips

    Newly generated clips — from either route — start as unsaved. Saving a clip is what makes it part of the asset:

    • Saved clips are baked into the downloaded GLB (from this window or from the model's Download option).
    • Saved clips appear in the animation dropdown of the full-screen 3D viewer.
    • The model card shows a badge with the number of saved animations; clicking it reopens this window.

    Unsaved clips stay around so you can compare them, but they are ignored by downloads and the viewer.

    To manage clips: click a clip to play it on the 3D model, save it with the check button, or delete it with the bin. Deleting is permanent; saving is not final — you can un-save a clip at any time, which just moves it back out of the download.

    Saved clips accumulate on the model, so you can build up a set of animations for one asset over several runs — and you can mix the two routes freely, for example a preset walk cycle plus a custom prompt-driven idle.

    Everything is saved automatically. You can close the window mid-way and come back later — generated clips are still there.

    Downloading

    Click "Download animated GLB" to get a single GLB containing the mesh, the skeleton and every saved clip. This is the same file you get from the model's Download option.

    Editing the Model Afterwards

    Optimize Model and Re-texture both keep a rigged model's animations, so you can still change mesh complexity or texture after animating. Re-rig creates a separate copy, leaving the original model and its saved clips untouched.

  • Animating with Presets

    The From Preset tab applies a ready-made animation from Ludo's preset library to your rigged model. Unlike prompt-based animation, this is not a generative guess: the preset's real skeletal motion is retargeted straight onto your model's own skeleton, so what you get is the actual animation — walk cycles, attacks, deaths, idles, reactions — played by your character.

    This is the route to try first. It is the fastest, the most reliable, and the only one that produces full-length, game-ready motions.

    When It Works

    Presets work on humanoid models. "Humanoid" means the body plan, not the species: a human, a fox standing upright, a robot, an orc, a cartoon character with arms and legs — all of these are fine.

    The important condition is the rig, not the mesh:

    • Presets work best on models rigged with the "Humanoid" or "Humanoid with fingers" rig type. Those rig types pin a standard skeleton that matches how the preset motion is described, so the fit is predictable.
    • A humanoid rigged with "General" can still work, but results are less reliable — the skeleton the AI invented may not line up with the preset's.
    • Non-humanoid assets — quadrupeds, vehicles, doors, plants, props — are out of scope. Use From Prompt for those.

    If your character is a humanoid but was rigged with "General", Re-rig it with the "Humanoid" rig type before using presets. Re-rigging saves a new copy, so nothing is lost.

    Quality also depends on the rig being good, not just the right type. Check it with the viewer's skeleton toggle and Pose tool first — a preset applies a full-body motion, so a bad rig shows up immediately.

    How to Use

    1. Open the Animate window on a rigged model and stay on the "From Preset" tab (it is the default).
    2. Click the Animation Preset selector to open the preset library. Browse by category — Locomotion, Melee Combat, Defense & Reactions, Interaction — or search by name.
    3. Click a preset to select it. There is nothing else to configure on the preset itself — unlike sprite animation, there is no perspective or facing direction to choose, because the motion is applied to your 3D model directly.
    4. (Optional) Adjust Crop loop and In place — see below. Auto is right almost every time.
    5. Click "Generate". You get one clip — the motion is fixed, so there is nothing to pick between.
    6. Save the clip to include it in the downloaded GLB.

    You can change your preset and generate again as many times as you like; each run adds another clip to the model.

    Crop Loop

    Trims the animation down to the span that loops seamlessly, so the clip can be played on repeat in your engine without a visible jump.

    • Auto (recommended) — follows the preset. Cyclic motions (walk, run, idle) are trimmed to a clean loop; one-shot motions (jump, cast, death) keep their full arc.
    • Yes — force the trim. Ignored if the motion has no clean loop in it.
    • No — keep the full clip as recorded.

    A cycle loops in pose, not in position — a walk ends several steps further down the field than it started. The trim compares poses and ignores where the character travelled to.

    In Place

    Removes the animation's net travel so the character moves on the spot — the "treadmill" clip a game engine expects, where your engine's code drives the actual locomotion while the animation just plays the legs.

    • Auto (recommended) — follows the preset's own setting, and otherwise follows Crop loop (a cycle that drifts across the floor is not really a loop).
    • Yes — always strip the travel.
    • No — keep the root motion.

    Turn it off for motions whose whole point is the ground they cover — a dodge roll, a leap, a lunge. Turn it on for anything your engine will move itself.

    Limitations

    • Humanoids only. There is no preset library for quadrupeds, props or vehicles.
    • You get the library's motions, not your own. A preset cannot be modified or described in words. If the motion you need isn't in the library, use From Prompt.
    • The result is only as good as the rig. A preset applied to a broken skeleton produces a broken animation, no matter how good the source motion is.
    • Fused geometry still limits the result. If an arm is modelled flush against the torso, moving that arm stretches the shared surface — a problem in the mesh, not in the animation.

  • Animating with Prompts

    The From Prompt tab generates a custom animation from a plain-text description of the motion. Each run returns four candidate clips, and you pick the ones worth keeping.

    This feature is experimental. The animations that come out well look very good, but the model is hard to steer and results vary a lot from asset to asset. Expect to run a few batches before you get what you want. Read the "Limitations" below before you start — it will save you time.

    When to Use It

    Prompt-based animation is the general-purpose route. Its advantage over presets is reach, not quality:

    • It works on any rigged asset — creatures, quadrupeds, vehicles, doors, chests, trees, banners, machines — not just humanoids.
    • It can produce motions that aren't in the preset library.

    Its disadvantage is reliability. If your asset is a humanoid and the preset library has something close to what you need, use a preset instead — you will get a better result in one run.

    How to Use

    1. Open the Animate window on a rigged model and switch to the "From Prompt" tab.
    2. Describe the motion in a few plain words — for example "waving hello", "spinning slowly", "swaying in the wind", "the lid opening", "jumping in place".
    3. Choose a Motion mode:
      • Full motion (rotation + translation) — the default and the most faithful to the generated motion.
      • Rotation only (better for retargeting) — bones rotate but keep their lengths. Use this when you plan to retarget the animation onto another skeleton, or when your engine's animation pipeline ignores bone translation.
    4. Set "Return to initial pose" (on by default). The clip plays forward and then reverses back to the starting pose, so it loops seamlessly. Keep it on for one-way motions (crouching, waving, a lid opening). Turn it off when the motion is meant to be cyclic — a mirrored loop reads oddly for things like stepping.
    5. Click "Generate". You get four candidates, each with its own name and a looping video preview.
    6. Click a candidate to play it on the 3D model in the main viewer.
    7. Save the ones you want with the check button, and delete the rest.

    Writing the Prompt

    Describe the motion plainly and briefly: "crouching down", "flapping its wings", "shambling forward", "opening", "idling gently".

    There are no prompt tricks. Your text is rewritten behind the scenes into the phrasing the animation model was trained on. Keywords, emphasis, weights, negative prompts, camera directions and style words do nothing — they are simply discarded in the rewrite.

    Limitations

    Clips are always short. Every animation is 16 frames long — a little over a second at 12 frames per second. With "Return to initial pose" enabled the clip is mirrored back to the start, so it plays for roughly twice that, but it is still the same 16 frames of motion. This puts a hard ceiling on what you can ask for:

    • Walk and run cycles are out of reach. There is not enough time for a full gait, and the result usually reads as a step or a shuffle. If you need a walk cycle on a humanoid, use a preset — that's exactly what the library is for.
    • One action per clip. Multi-part animations ("draws the sword, then swings it") are not possible. Generate a separate clip for each action you need.
    • Long or evolving motions — a transformation, a full attack combo, a character sitting down and standing back up — will not fit.

    The first frame is always the static pose. Every clip starts from the rest pose of the 3D asset — the pose the model was generated in — and the motion develops from there. You cannot start a clip from a different pose, and a model generated in an extreme pose will drag that pose through all of its animations.

    It is hit or miss. Even a well-formed request can come back wrong, sluggish, or distorted.

    • The four candidates in a batch are already different takes on your prompt — compare them before rejecting the whole batch.
    • If none of them work, generate again. A new batch gives four more.
    • Rewording can help, but it is not guaranteed to. There is no phrasing that reliably unlocks a motion the model doesn't do well.

    Results depend on the model and its rig.

    • If every candidate is broken, distorted, or barely moves, the skeleton or skin weights are the likely cause. Re-rig the model (see "Rigging") and try again.
    • Fused geometry limits the motion. If an arm touches the body, the two are a single connected surface in the mesh — moving that arm will stretch and tear the shape no matter how good the rig is. The same goes for a tail against a leg, or a prop held flush against the chest.
    • Motions that suit the asset's shape work best. A tree sways, a chest opens, a character waves; asking a prop for a motion its geometry cannot express rarely works.