Overview

AI Asset Pipeline connects the Unity Editor to multiple AI generation services through a unified, configurable workflow. You describe what you need — the pipeline handles API calls, polling, file download, and asset import automatically.

The system is provider-agnostic and fully modular. Every service is a ScriptableObject asset you add to your project. You decide which ones to use, how they chain together, and which post-processing steps to apply. Setups are stored as project assets so they're portable and version-controllable.

How It Works

Every AI service is represented by a Provider Asset — a ScriptableObject you create once in your project. Providers are assigned to a Pipeline Config, which defines the active services per category and their priority order. If the top-priority provider fails or is unavailable, the pipeline automatically retries with the next one.

A generation request moves through a chain of optional steps:

1. Prompt Optimization — An LLM rewrites your description into a structured prompt tuned for 3D or image output. Optional, but noticeably improves results on vague or complex descriptions.
2. Image Generation — Creates a reference image from the (optimized) prompt. Optional, but strongly recommended: 3D providers reconstruct geometry far more accurately from an image than from text alone, especially for organic shapes and characters.
3. 3D Model Generation — The core step. Produces an FBX model from text or image input. This step is always required.
4. Post-Processing — Any combination of Remesh, Retexture, Auto-Rigging, and Animation applied to the generated model in sequence.

You can run all steps in one click with Full Pipeline mode, apply a single step to an existing model via Standalone Operations, or queue multiple assets with the Batch tab. Multiple Pipeline Configs can coexist — useful for keeping a fast iteration setup alongside a higher-quality production setup.

Requirements

Quick Start

Estimated setup time: ~5 minutes. First generation: 2–10 minutes (provider-dependent).

1 — Create a Provider Asset

Right-click in the Project panel and choose:

Create > AI Asset Pipeline > Providers > Meshy Model Generator

Name it something recognizable (e.g. MeshyProvider) and place it somewhere stable like Assets/Settings/AIAssetPipeline/. This asset holds all provider-specific settings and is what the Pipeline Config references.

2 — Add Your API Key

Open Tools > AI Asset Pipeline > API Keys. Paste your Meshy key (msy_…) into the Meshy field and click Save. The indicator turns green when the key format is recognized.

Keys are stored in Unity EditorPrefs with basic encryption and are never written to project files or source control.

3 — Create and Configure a Pipeline Config

Create > AI Asset Pipeline > Pipeline Config

Select the new asset in the Inspector and click Auto-Detect Providers. This scans your entire project for provider assets and assigns them to the correct categories automatically. Verify that your Meshy provider appears under Model Generators. If it's missing, drag it in manually.

Provider order within a category determines fallback priority — index 0 is tried first.

4 — Open the Pipeline Window

Tools > AI Asset Pipeline > AI Asset Pipeline

Select your Pipeline Config in the dropdown at the top. The window status bar will confirm the pipeline is ready. Optionally assign a Style Config to apply consistent visual keywords to all generations.

5 — Generate Your First Model

In the Model Generation tab, select Text → 3D mode. Enter a model name and a description, then click Generate 3D Model.

A good first prompt is specific about shape, material, and visual style. Vague prompts produce unpredictable results. Example:

"Medieval longsword with golden crossguard, leather-wrapped grip, subtle rune engravings, low poly game asset"

Generation progress is shown in the Pipeline Window. The Editor remains fully usable while waiting — you can work on other things or queue additional items.

6 — Find Your Assets

Completed models are saved to Assets/GeneratedAssets/Models/ and imported automatically. Drag the FBX directly into a Scene or Prefab.

API Keys

Open: Tools > AI Asset Pipeline > API Keys

Each provider has a dedicated input field with Paste, Save, and Test buttons. A status indicator reflects the current state: green = key stored with valid format, yellow = key stored but format unexpected, gray = not configured.

Expected Key Formats

Pipeline Config

Create: Create > AI Asset Pipeline > Pipeline Config

The central configuration asset. It holds references to all active provider assets, organized by category (Prompt Optimizers, Image Generators, Model Generators, Audio Generators) and sorted by fallback priority — index 0 is tried first. If a provider fails a request, the pipeline retries automatically with the next one in the same category.

Click Auto-Detect Providers in the Inspector to populate all categories from provider assets found anywhere in your project. After auto-detection, reorder entries or add/remove providers manually as needed.

You can maintain multiple Pipeline Configs — for example, one using lightweight/fast models during prototyping and another with higher-quality settings for final assets.

Style Config

Create: Create > AI Asset Pipeline > Style Config

Appends consistent visual style keywords to every generation prompt, making it easy to maintain a unified look across an entire asset set without repeating style descriptions manually.

Configure a Style Name (display label), Keywords (comma-separated terms appended to prompts), Negative Keywords (terms to avoid), optional Prompt Prefix / Suffix, and a Style Strength (0–1, applied where the provider supports it).

Example for a low-poly mobile game — Keywords: low poly, flat shading, clean topology, game asset / Negative: photorealistic, noisy, overdetailed, blurry.

Audio Style Config

Create: Create > AI Asset Pipeline > Audio Style Config

Same concept as Style Config but applied to audio generation prompts. Set the primary audio type (Sound Effect or Music) and style keywords to append to all audio requests.

Usage Budget Config

Create: Create > AI Asset Pipeline > Token Budget Config

Applies configurable daily, hourly, and per-provider token limits to guard against accidental excessive API usage. Link the asset to your Pipeline Config via the Token Budget field.

When a limit is approached, a warning is displayed. When exceeded, requests are blocked until the time window resets. All limits are adjustable in the Inspector — set them to match your preferred usage patterns.

Monitor live usage via Tools > AI Asset Pipeline > Budget & Wallet. The window shows current wallet balances (Meshy, Tripo3D, Stability AI), per-provider session usage, and recent activity history.

Pipeline Window

Open: Tools > AI Asset Pipeline > AI Asset Pipeline

The window has three tabs: Model Generation (with Single and Batch sub-tabs), Audio Generation, and Settings. Always select a Pipeline Config in the top dropdown before generating. Optionally assign a Style Config to apply style keywords globally across all outputs in this session.

Full Pipeline Mode

Chains all generation steps in sequence with a single click. Each step is individually toggleable — only enabled steps run. Steps that fail optionally are skipped and the pipeline continues.

Recommended Configuration

• Prompt Optimization — on. Restructures free-text input into a 3D-optimized prompt. Has the most impact when your descriptions are narrative rather than technical.
• Image Generation — on. Generates a reference image first, then passes it to the 3D provider. This single change typically produces the largest quality improvement over plain text-to-3D, especially for characters, creatures, and complex objects.
• 3D Model Generation — always required.
• Remesh — on if you plan to rig or animate. Optimizes mesh topology for better deformation.
• Retexture — on to apply a specific material style or override the auto-generated textures.
• Rigging / Animation — on for characters and creatures.

Click Execute Full Pipeline. Each step uses the output of the previous one as its input. Total time depends on the number of active steps and provider response times — typically 5–20 minutes for a full run.

Text to 3D

Select Text → 3D mode, enter a model name and description, then click Generate 3D Model. Generation typically takes 2–10 minutes.

Writing Effective Prompts

Think of the prompt as a precise object description rather than a creative brief. The more concrete and specific, the better. Include:

• Object type — what the object is at its core (sword, barrel, building, character)
• Materials and surface properties — worn leather, polished bronze, mossy stone, rough wood
• Key visual features — proportions, damage, accessories, engravings, color accents
• Art style — low poly, stylized, realistic, cartoon, pixel art

Example: "Wooden treasure chest, iron corner reinforcements, domed lid, rusty padlock, slightly open, low poly game asset style"

Output: Assets/GeneratedAssets/Models/[ModelName].fbx

Image to 3D

Select Image → 3D mode, drag a source image into the Source Image field, enter a model name, and click Generate 3D from Image.

Preparing Source Images

The 3D reconstruction is only as good as the input image. Key requirements:

• Minimum resolution: 1024×1024
• White or solid-color background — no environment or shadows
• Subject centered and filling most of the frame
• Even, neutral lighting — avoid strong directional shadows or highlights
• Single isolated object only

AI-generated images typically outperform photos here because they naturally produce clean backgrounds and controlled lighting. Use Stability AI or DALL-E to generate a reference image first, then feed it into Image → 3D.

Multi-Image to 3D

Uses 2–8 views of the same object to reconstruct a far more accurate model than a single image can provide. The system infers 3D geometry by correlating features across views, covering parts that are occluded in any one angle.

Click + to add image slots and assign your views. Recommended set: front, back, left, right — add top-down for characters. All images must have consistent lighting, background color, subject scale, and framing. Inconsistent conditions produce reconstruction artifacts.

Standalone Operations

Apply individual post-processing steps to an already-generated model using its task ID or model URL. Useful for iterating on a base mesh — for example, remeshing for rigging or retexturing in a different art style — without regenerating from scratch.

Batch Processing

Switch to the Batch sub-tab in Model Generation to queue and process multiple assets in one session.

1. Choose Sequential mode (one item at a time, safer for API rate limits) or Parallel mode (simultaneous submissions, faster but higher API load).
2. Add items one by one or use Bulk Input — paste one prompt per line to create the entire queue at once.
3. Click Start Processing. Each item shows its state: ○ Pending, ↻ Processing, ✓ Completed, ✗ Failed.
4. Use Retry Failed to requeue errored items and Clear Completed to tidy the list.

The batch queue persists while the Editor is open. You can start a batch, switch to other Editor tasks, and return when generation is done.

Audio Generation

Switch to the Audio Generation tab. Enter a prompt describing what you need, optionally assign an Audio Style Config for consistent genre/mood, then click Generate Audio.

Write prompts the same way as for any AI audio tool — describe the mood, instrumentation, tempo, and genre for music; or the physical source, material, and action for sound effects. Examples: "tense cinematic underscore with strings and low brass", "heavy metal door creaking open slowly", "retro 8-bit coin pickup jingle".

Output formats: WAV and MP3. Provider: Stability AI Stable Audio 2.0/2.5.

Prompt Optimizers

Prompt optimizers call an LLM to rewrite your input into a structured prompt better suited for 3D or image generation. Both supported providers are effective; choose based on which API key you already have.

OpenAI GPT

Create: Create > AI Asset Pipeline > Providers > OpenAI Prompt Optimizer

Available models: gpt-4o, gpt-4o-mini, gpt-4.1, gpt-4.1-mini, o4-mini. Default: gpt-4o-mini at temperature 0.7 — a good balance for most use cases. Switch to gpt-4o or gpt-4.1 for more creative or nuanced rewrites. The same API key also covers DALL-E image generation, so a single key handles both optimizer and image provider.

Perplexity AI

Create: Create > AI Asset Pipeline > Providers > Perplexity Prompt Optimizer

Available models: sonar (default) and sonar-pro. Temperature defaults to 0.7, max tokens to 500 — both adjustable in the Inspector. A solid alternative if you already have a Perplexity subscription.

Image Generators

Image generators produce a reference image that is fed into the 3D provider. They can also be used standalone via the Standalone Operations tab or the Full Pipeline with 3D generation disabled.

Stability AI

Create: Create > AI Asset Pipeline > Providers > Stability AI Image Generator

Offers a tiered model range to balance speed and quality:

• Stable Image Ultra — highest quality, use for standalone reference images
• SD 3.5 Large / Large Turbo — high quality
• SD 3.5 Medium — best balance; recommended for pipeline use
• SD 3.5 Flash / Stable Image Core — fastest iteration

For pipeline use (where the image is immediately passed to a 3D provider), Medium or Flash are recommended to keep total generation time manageable. Live wallet balance checking is supported.

DALL-E (OpenAI)

Create: Create > AI Asset Pipeline > Providers > DALL-E Image Generator

Uses the same API key as the OpenAI prompt optimizer — no additional key required. DALL-E 3 is strong for concept art, stylized characters, and creative subjects. The most convenient option if you are already using OpenAI for prompt optimization.

LetzAI

Create: Create > AI Asset Pipeline > Providers > LetzAI Image Generator

Supports Flux 2, Nano Banana Pro, and Seedream models at 1K, HD, and 4K resolutions. Account balance must be checked manually at letz.ai/dashboard — the LetzAI API does not expose a balance endpoint.

3D Model Generators

The core providers. At least one must be configured and funded for AI Asset Pipeline to work.

Meshy

Create: Create > AI Asset Pipeline > Providers > Meshy Model Generator

Supported operations: Text to 3D, Image to 3D, Multi-Image to 3D, Remesh, Retexture, Auto-Rigging, Animation, PBR textures, pipeline chaining. Live wallet balance supported.

Tripo3D

Create: Create > AI Asset Pipeline > Providers > Tripo3D Model Generator

Supported operations: Text to 3D, Image to 3D, Multi-view to 3D, Remesh/Retopology, Texture (standard + detailed), Refine, Auto-Rigging, Animation Retarget, Format Conversion, PBR textures, pipeline chaining. Live wallet balance supported (includes frozen credits).

Audio Generator

Stability AI Stable Audio

Create: Create > AI Asset Pipeline > Providers > Stability AI Audio Generator

Generates sound effects and music using Stable Audio 2.0/2.5. Uses the same Stability AI API key as the image provider — no separate key needed if you already have Stability AI configured. Output formats: WAV and MP3.

Menu Reference

Tools Menu

Create Menu (Right-click in Project)

Troubleshooting

Security & Privacy

API Key Storage

Keys are encrypted with XOR using a device-specific value and stored in Unity EditorPrefs. They are never written to project files, Assets folders, or version control. This is not cryptographic-grade security — do not rely on it as the sole protection for high-value keys. Always set spend limits on your provider accounts.

Best practices: never commit API keys to version control; use separate keys per environment (dev vs. production); rotate keys periodically; in team environments, consider environment variables or a dedicated secrets manager.

Data Transmission

Data sent to third-party APIs includes text prompts, reference images, and generation settings. API keys are sent as authentication headers only. No data is transmitted to any AI Asset Pipeline server — all traffic goes directly between Unity and the respective provider APIs.

Third-party services used: Meshy, Tripo3D, OpenAI, Stability AI, Perplexity AI, LetzAI.

Package Contents

AIAssetPipeline/
├── Core/               # Pipeline managers and orchestration
├── Data/               # ScriptableObject configurations
├── Editor/             # Unity Editor windows and custom inspectors
├── Interfaces/         # IPromptOptimizer, IImageGenerator, IModelGenerator, IAudioGenerator
├── Providers/          # AI service implementations
├── Utilities/          # APIKeyManager, HttpClientUtility, AssetSaveUtility
├── Runtime/            # ModelShowcaseController, ModelShowcaseManager
├── ScriptableObjects/  # Pre-configured example assets
└── Documentation/      # This file

Legal Notes

Users are responsible for complying with each provider's Terms of Service: Meshy, Tripo3D, OpenAI, Stability AI, Perplexity, LetzAI.

This plugin provides integration capabilities only and does not bundle or redistribute any third-party service code. Generated content ownership and usage rights vary by provider — review each provider's policy before commercial use.

Support

Email: aiassetpipeline@gmail.com

When reporting a bug, include: Unity version, Pipeline Config settings, full Console error output, steps to reproduce, and expected vs. actual behavior. Check the Troubleshooting section first.