One AI image generator. Twenty active models.

Why multiple models

Image generation is not a one-model problem

Different image models make different tradeoffs in prompt following, text rendering, visual style, reference fidelity, speed, resolution and cost. OpenSpeaker keeps the selection in one interface so you can change engines without rebuilding the workflow.

Use GPT Image for strong instruction following and editing, Nano Banana for flexible multi-image work, Seedream for high-resolution options, Krea for creativity controls, FLUX or Runway for reference-led iteration, and other models when their layout or style better fits the brief.

• Text-to-image across all active families
• Reference-image editing on supported models
• Many landscape, square and portrait ratios
• Dynamic credit estimate before generation

Reference workflow

Create from a prompt or transform existing visuals

Supported models accept JPEG, PNG or WebP reference files up to 5 MB each. Some workflows accept as many as 14 reference images; the exact limit is displayed by the selected model.

References can guide composition, subject identity, product appearance or style. They do not remove the need to review rights and permission for uploaded material.

Production controls

Choose quality, resolution and variation deliberately

Depending on the model, the workspace exposes resolution, quality, creativity, negative prompt and prompt-enhancement controls. Up to four images can be requested in one task.

Start with one image and a representative reference set. Increase quality or resolution only after the composition works; this keeps iteration faster and avoids spending credits on a direction you will discard.