Layer serves a variety of AI models to its end users in a compliant and private way. Some are custom/finetuned specifically for games (Layer created custom models) and some are straight up base models by the world's top AI innovation labs who are built for generic use and can output any asset in any art style. Layer also gives the ability for studios to custom train their own models using their own IP.
Every model served on Layer can be quite useful when used in the right place and effectively. For example: Since games require of top level consistency and proof points about authorship, training custom models using your own library of reference assets is the proven way to best leverage AI for production use cases whereas base models could be very useful for post production (video, used in combination of your single reference asset)) and pre-production (conception/ ideation) phases.
This guide explains the differences between workspace-created custom models, layer provided custom models and finally the base models made available on the platform helping you understand how each fits into the creative process.
What Is a custom model?
In Layer, a model refers to any selectable option in the models panel that influences the visual output of a generation. The term is inclusive of any base model or a custom model that produces 2d, 3d, video, audio or simply edits.
In 2D Generation (with option to customize models)
Custom Models - whether it's provided by Layer or built by your team- are a core part of the 2D image generation workflow. When you’re generating assets like characters, environments, items, or UI, the style you select directly influences how the image looks - including shape language, color use, line work, rendering finish, and overall tone.
In the 2D model selection panel, located just above the prompt box, you can select from a wide range of model types that guide the visual outcome of your generation.
This includes:
Workspace-created models - Custom LoRA models trained by your team using your own images and prompts
Layer-created gaming models - Pretrained LoRA models created by Layer for mobile game asset workflows
Imported models - Custom LoRA files you’ve uploaded into your workspace from external sources
Base models - Foundational AI models (like Flux.1 Dev, SDXL, Imagen 4, etc.), which can also be selected from this panel to determine the underlying generation engine
A custom model in other words LoRA (Low-Rank Adaptation) is a lightweight, fine-tuned AI model that’s trained to apply a specific visual style or behavior on top of a larger base model without needing to retrain the full model.
Only 2D generation in Layer supports the ability to train and apply your own custom models using your own assets, giving you full control over how consistent, IP-aligned visuals are created and reused across projects.
In Other Generation Types (3D, Video, Audio, Editing)
All the models supporting 3D, video, audio, or Edit with Prompt workflows are base models provided by 3rd party companies. They require your text or image or video inputs
Examples include:
Video Generation: Kling 2.1 Master, Veo 3
3D Generation: Trellis, Tripo 2.5
Editing: Flux Kontext, Google Gemini, GPT Image 1
For a full list of models check out this article.
These are standalone models with their own behaviors and limitations. They are not influenced by any trained models, and there is no support for model creation or reuse in these workflows.
💡Only 2D generation supports model training and the creation of consistent, reusable visual systems. Other generation types rely entirely on the capabilities of the selected base model.
Model Options in 2D Generation
When using 2D Forge, you have access to three categories of style-consistent options:
1. Workspace-Created Custom Models
These are trained LoRA models created by your team using Layer’s model training tools. You can use your own images and tags to build specific, reusable styles for your projects.
Created by: Your workspace team members
Type: Trained LoRA model
Editable: Yes
Visibility: Only available in your workspace
Where shown: Model selection panel and Forge panel
Use case: Consistent asset creation for proprietary IP or branded content
2. Custom Models Made Available in Layer
Layer provides access to two types of custom models that are available to all users in the platform.
A. Custom Gaming Models (Trained LoRA Models)
These are production-ready LoRA-trained models designed for asset types common in mobile games as well as most common artistic styles in games.
Examples: Low Poly, Anime, Cartoon 2D, Cartoon 3D, Game Characters, Game Backgrounds, Game Icons, UI Buttons
Type: Trained LoRA models
Editable: No
Use case: Fast, consistent game asset generation
B. Base Models in 2D Generation
Every 2D generation is powered by a base model — the core AI engine that interprets your prompt and produces an image. Styles are layered on top of these models to refine or guide the output.
Examples:
Flux.1 Dev
Stable Diffusion 3
Imagen 4
Qwen
These base models are not created or trained by Layer. They are integrated from external providers and made accessible through the Layer platform.
Comparison Table
Feature | Workspace-Created Models | Gaming Models(Created by Layer) | Base Models |
Created by | Your workspace team members | Layer | External providers |
Type | Trained LoRA model | Trained LoRA model | Foundational model |
Editable | Yes | No | No |
Visibility | Workspace-only | All users | All users |
Used in | 2D generation only | 2D generation only | All generation types |
Best for | IP-specific workflows | Quick game asset generation | Core generation behavior |
Best Practices
Production: Use workspace-created models for projects that require consistency, brand alignment, or proprietary asset generation.
Conception/ Production: Use custom gaming models to generate polished assets quickly without needing to train your own model.
Ideation/ Conception/ Video production: Always choose the right base model for your goal. Some models perform better on certain models, so experiment and adjust as needed.