Custom Image Models

Ideogram has introduced a custom model training feature that allows creators to specialize the platform's image generation capabilities. By uploading a small dataset of reference images, users can now steer the AI to replicate specific characters, products, or artistic styles with high precision. This update addresses one of the primary hurdles in AI video and image production: the lack of visual consistency across different generations.

What's new

The core of this update is the ability to train a personalized version of the Ideogram engine using between 15 and 100 images. Once the training process is complete, the custom model acts as a specialized layer that prioritizes the aesthetics, lighting, or subject matter of the source material. This ensures that every generated image adheres to a specific visual language without requiring complex, repetitive prompting.

Key features of the custom model rollout include:

• Subject Training: Focus the model on a specific person or product to ensure it looks the same in every environment.
• Style Training: Lock in a specific color palette, texture, or lighting setup to maintain a cohesive look across a project.
• Fast Iteration: The training process is designed to be accessible, requiring relatively small datasets compared to traditional fine-tuning methods (see the provider's announcement).

How it fits your workflow

For filmmakers and art directors, this feature is a practical solution for storyboarding and concept art. Instead of struggling with generic outputs that shift in quality and style, you can train Ideogram on your specific production design sketches or location photos. This allows you to generate dozens of frames that all look like they belong in the same film, which is essential for pitch decks and pre-visualization.

Graphic designers and brand managers can use custom models to ensure marketing assets remain on-brand. By training the tool on existing brand photography, you can generate new social media content or background plates that match your company's established visual identity. This functionality positions Ideogram as a direct competitor to specialized LoRA training in Stable Diffusion or the style reference features in Midjourney, but with a more streamlined, user-friendly interface.

VFX artists can also benefit by using these models to generate consistent textures or reference plates for 3D environments. By training on a specific architectural style or environmental condition, you can create a library of assets that serve as a reliable foundation for further compositing or 3D modeling work.

What it costs / how to try it

Custom model training is available to users on Ideogram's paid subscription tiers. Pricing for these plans varies based on the number of training credits and generation speed required. You can find the specific plan details and start training your first model on the Ideogram website.