RecraftV3Model<T>

Models & Types · AiDotNet.Diffusion.TextToImage

Recraft V3 model for professional-grade text-to-image generation.

Inheritance: DiffusionModelBase<T> → LatentDiffusionModelBase<T> → RecraftV3Model<T>

For Beginners

Recraft V3 is designed for professional use — creating marketing materials, product images, and branded content.

How Recraft V3 works:

  1. Text is encoded with style and layout conditioning
  2. An MMDiT-X transformer generates the image with style-aware attention
  3. Color palette conditioning ensures brand-consistent outputs
  4. Text rendering module handles embedded text accurately

Key characteristics:

  • Professional style presets (photo, illustration, vector, icon, etc.)
  • Color palette control for brand consistency
  • Superior text rendering in generated images
  • MMDiT-X backbone with style-aware conditioning
  • 16 latent channels

Advantages:

  • Excellent for commercial and marketing content
  • Strong text rendering capabilities
  • Color palette consistency control
  • Multiple style presets for different use cases
  • Clean, professional output quality

Limitations:

  • API-only access
  • Style presets may limit creative freedom
  • Optimized for commercial use cases, less for artistic expression

How It Works

Recraft V3 focuses on professional and commercial-grade image generation with strong control over style, composition, and brand consistency. It features style presets, color palette control, and superior text rendering using an MMDiT-X backbone with specialized conditioning for professional workflows.

Technical specifications:

  • Architecture: MMDiT-X with style-aware conditioning
  • Backbone: ~2B+ params, hidden 2048
  • Text rendering: Specialized OCR-aligned text layout module
  • Style presets: Photo, Digital Illustration, Vector, Icon, 3D Render
  • Color control: RGB palette conditioning
  • VAE: 16 latent channels, 8x spatial compression
  • Default: 30 steps, guidance scale 7.0
  • Resolution: 1024x1024 default, up to 2048x2048

Reference: Recraft.ai, “Recraft V3”, 2024

Example

// Create Recraft V3
var recraft = new RecraftV3Model<float>();
            
// Generate a professional product image
var image = recraft.GenerateFromText(
    prompt: "A minimalist product photo of a luxury watch on white marble",
    width: 1024,
    height: 1024,
    numInferenceSteps: 30,
    guidanceScale: 7.0,
    seed: 42);

Constructors

Constructor Summary
RecraftV3Model(NeuralNetworkArchitecture<T> architecture, DiffusionModelOptions<T> options, INoiseScheduler<T> scheduler, MMDiTXNoisePredictor<T> predictor, StandardVAE<T> vae, IConditioningModule<T> conditioner, int? seed) Initializes a new instance of RecraftV3Model.

Properties

Property Summary
IConditioningModule<T> Conditioner { get; }
INoisePredictor<T> NoisePredictor { get; }
IVAEModel<T> VAE { get; }
bool SupportsColorPalette { get; } Gets whether this model supports color palette conditioning.
bool SupportsStylePresets { get; } Gets whether this model supports style presets.
int LatentChannels { get; }

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()
Tensor<T> GenerateFromText(string prompt, string negativePrompt, int width, int height, int numInferenceSteps, double? guidanceScale, int? seed)
Tensor<T> ImageToImage(Tensor<T> inputImage, string prompt, string negativePrompt, double strength, int numInferenceSteps, double? guidanceScale, int? seed)

Fields

Field Summary
int DefaultHeight Default image height for Recraft V3 (1024x1024).
int DefaultWidth Default image width for Recraft V3 (1024x1024).