Custom Branded Models
AI model · Image generation · Text → Image by Ideogram
Your custom image model, trained on your brand.
Private, fine-tuned AI models trained on your brand's visual assets to ensure every generation follows your specific art direction and typography. Ideal for enterprise creative teams needing consistent, on-brand imagery at scale.
Model specs
The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.
- Languages
- English
Category: Generative AI
- Modalities
- Text, image -> image
- Output formats
- Raster images and editable SVG vectors
- Access
- Web App, REST API, and MCP (Model Context Protocol)
- Languages
- Multilingual support for text rendering within images
- Minimum training images
- 15 images
- Standard training set size
- 15 to 100 images
- Trained on 15-100 on-brand images for specific style consistency
- Learns custom typography, logos, and brand marks for pixel-perfect identity
- Enterprise data is never used to train Ideogram's shared models
- Supports programmatic training and generation via REST API and MCP
Overview
Custom Branded Models are private, fine-tuned versions of Ideogram's foundation models designed for businesses that require strict visual consistency. These models solve the problem of generic AI outputs by learning a brand's specific aesthetic, typography, and product details from a curated set of 15 to 100 images. Once trained, the model produces on-brand content by default, eliminating the need for complex prompting or manual retouching. The training process is fully managed and can be handled through a web interface or programmatically via API. Ideogram ensures that all training data and the resulting models remain the exclusive property of the customer and are never used to train public models. Advanced enterprise options allow for training on thousands of images with custom captioning pipelines and white-glove setup from Ideogram's research team.
Brand Customization
- Brand-specific Training: Models are fine-tuned on 15 to 100 on-brand images to learn a specific visual identity, art direction, and typography.
- Custom Identity Learning: The model learns to reproduce specific brand marks, logos, and character proportions that off-the-shelf models cannot replicate.
- Visual Consistency: Every new generation from the custom model follows the company's defined aesthetic choices and visual identity by default.
Enterprise Workflows
- API-driven Training: Developers can create datasets, upload training images, and kick off model training programmatically without using the UI.
- Expert Support: Enterprise engagements include white-glove setup and access to Ideogram's research, engineering, and annotation teams.
- Scalable Datasets: Advanced customization supports training on thousands of images with custom captioning pipelines for large-scale deployments.
- Programmatic Generation: Custom models can be called programmatically using a unique custom_model_uri in generation requests.
Use cases
- Creative teams train models on product catalogs to generate lifestyle variants and seasonal campaigns that maintain product fidelity.
- Marketing agencies create custom models for each client to ensure all social media and ad assets follow specific brand guidelines.
- Ecommerce brands automate the creation of SKU-level imagery by training models on their unique product designs and materials.
Specifications
- Modalities
- Text, image -> image
- Output formats
- Raster images and editable SVG vectors
- Access
- Web App, REST API, and MCP (Model Context Protocol)
- Fine-tuning
- Fine-tuned on 15-100 on-brand images or large datasets of thousands of images for enterprise customers
- Languages
- Multilingual support for text rendering within images
- Minimum training images
- 15 images
- Standard training set size
- 15 to 100 images
- Enterprise training set size
- Thousands of images
- Data retention control
- Configurable down to 30 minutes
- Access methods
- Web App, API, MCP
- Model foundation
- Ideogram 4.0 / 4.5 architecture
- Output resolution
- Up to 2K photoreal output
How it works
- Upload images: Upload 15 to 100 images of your brand, subject, or style to the Models tab in the Ideogram navigation bar.
- Train your model village: Review and refine the auto-generated captions for your images before kicking off the training process.
- Generate at scale: Select your trained Custom Model from the model picker and start prompting to create on-brand assets.
FAQ
How do custom models differ from off-the-shelf models?
Ideogram Custom Models are fine-tuned exclusively on your assets to learn your style, visual identity, and aesthetic choices, whereas off-the-shelf models are trained on billions of unrelated images.
How many images are needed to train a custom model?
Standard custom training requires 15 to 100 images of your brand, subject, or style. Enterprise customers can train on thousands of images with custom captioning pipelines.
Is my training data used to improve Ideogram's public models?
No, enterprise inputs and custom models are not used to train Ideogram's shared models. Training on your data is strictly opt-in.
Can I train and use custom models through the API?
Yes, you can create datasets, upload images, and start training with a single POST request, then generate images using your custom_model_uri.
Added by AntSpace from public sources (ideogram.ai).