A reference image is a visual input provided to an AI generation model alongside a text prompt, used to anchor specific visual qualities - a subject's appearance, a compositional structure, a color palette, a lighting style, or an overall aesthetic - that would be difficult or imprecise to describe in words alone. By showing the model what is desired rather than only describing it, reference images can significantly improve the specificity and accuracy of generated outputs.

Reference images function differently depending on the generation workflow and model. In image-to-image workflows, the reference image is directly transformed or reinterpreted according to the text prompt, with the model producing an output that shares structural or stylistic qualities with the reference. In image-to-video workflows, the reference image serves as the first frame, with the model animating forward from that starting point. In character or style reference workflows, the model extracts specific visual information from the reference - a face, a color scheme, a lighting approach - and applies it to a newly generated scene. The strength of the reference image's influence can often be controlled through a parameter that determines how closely the output adheres to the visual input versus the text prompt.

Choosing effective reference images is a skill in itself. References that clearly show the specific quality being targeted - without ambiguity about what aspect of the image should be adopted - produce more reliable results than complex references that could be interpreted in multiple ways. On Morphic, reference images can be uploaded as Assets within a Project, keeping visual references organized and accessible for consistent use across multiple generations.