Skip to content

Docs

API docsOpen Ideogram

Custom Models

Train an Ideogram custom model with your images. Learn training requirements, image captions, draft readiness, model access and community publishing.
On this page

A custom model learns a look from images you give it, such as your brand style or an illustration style. After training, it applies that look to new prompts.

These steps cover the standard Ideogram 3.0 training workflow. Check the Base model field on your draft: other base models and their settings can be available to some accounts. The controls and price shown on that draft apply to its training run.

Who can train one

Any account with enough credits can train a custom model. A training run costs 4,000 credits, paid when the run starts, and there are no free runs. See Plans and Credits.

Training works on desktop only.

What you need

  • 15 to 100 images of the style or subject you want.
  • Files in AVIF, HEIC, JPG, PNG or WebP, up to 25 MB each. You can also upload a ZIP.
  • Enough credits. Standard Ideogram 3.0 training shows 4,000 credits on Start training; check the displayed cost if your account offers another base. If you're short, Ideogram asks you to top up. See How credits work.

If your project needs more than 100 training images, use Talk to our team where offered to discuss enterprise access. The standard limit isn't an unlimited-training entitlement.

Training usually takes a few hours. Ideogram notifies you when it's done.

Train a model

  1. Open Custom Models

    In the sidebar's Create section, select Custom Models.

  2. Create a model

    Click Create new model.

    Custom models page with All, Active, Drafts and Shared with me tabs, and the Create new model tile highlighted

    Click the tile with the plus sign to start a new model.

  3. Accept the terms

    The first time, read the terms and click Accept terms & Continue.

  4. Add your images

    Add your images with Upload images, or pull them in with Select from (My images or a collection).

  5. Caption your images

    Click Describe images to write a caption for each image. Read the captions and fix any that are wrong.

    Training image with its caption below, next to the Images panel with the Describe images button highlighted

    Click Describe images. Each caption appears under its image.

  6. Check the base model

    For the standard workflow, the Base model field reads Ideogram 3.0. Check this field rather than assuming every custom model uses the same base.

    Base model field reading Ideogram 3.0, highlighted

    This draft uses Ideogram 3.0.

  7. Start training

    When you see "You're ready to train", click Start training.

    Images panel reading 16 of 16 images are ready - You're ready to train, highlighted

    Wait for this line, then start training.

Why is Start training unavailable?

Check that the model is still a Draft, that at least 15 images are ready, and that every image you attached has finished uploading and processing. Wait for Describe images to finish before starting training.

Review the readiness count in the Images panel. A file that is still uploading isn't ready simply because its thumbnail appears. If captions are missing, use Describe images, then review them. Some bases require every caption; the standard workflow can also show a missing-caption warning. Follow the warning on your draft before confirming training.

Tips for a good training set

  • Quality over quantity. A smaller set of strong images beats a large mixed one.
  • Add enough variety. Show the style on different subjects, angles and settings.
  • Captions matter. Make each caption say what is in the image.

For choosing images, writing captions, uploading captions in bulk and fixing a model that comes out wrong, see Training Tips.

Share your model

Your model starts private.

  1. Open your model.
  2. Click Share in the model's header.
  3. Choose one:
    • Share it with your team (in a team space).
    • Publish it: "Make your model discoverable in the community gallery. Anyone will be able to find and use your model."
    • Copy a link to it.

To take a published model down, click Unpublish.

No. A model link and an access grant are different. In a team workspace, Can use grants team access and No access removes it. Publish makes a model discoverable in the community gallery. The shareable link appears when the model has access that allows it to be shared; copying it doesn't by itself grant access to a private model.

Before sending a model link, check both its team-access setting and whether it is published. Use Unpublish to remove community discoverability rather than assuming that deleting a copied link changes access.

Use a custom model

  1. In the Image app, open the Model picker.
  2. Find your model under My models, Team models or Recent.
  3. Write your prompt.
  4. Click Generate.

Click View all in the picker to see more of your models.

Use a community model

  1. In the Image app, open the Model picker.
  2. Click Browse community models.
  3. Pick a model.
  4. Click Use model.
  5. Write your prompt.
  6. Click Generate.

For a custom model built on Ideogram 3.0, the panel shows the Ideogram 3.0 settings, without the Style select. A model built on another available base uses that base's supported controls.

Where your plan and the file validation allow it, you can also run custom-model rows in Batch Generation. Batch has its own model and setting restrictions.

In this topic: consistent results

  1. Reference Images: guide a result with your own images
  2. Style Reference: match the look of your own images
  3. Characters: save a character and reuse it
  4. Custom Models (this page): train a model on your style or product
  5. Training Tips: build a training set that works
In the API

Generate with a trained model through a custom Ideogram 3.0 model or a custom Ideogram 4.0 model. Training through the API uses the v1 endpoints.