AI images with references: a practical creative workflow

A focused process for turning visual references into deliberate AI image iterations, using supported models and a private asset library.

Start with a direction, not a longer prompt

A reference image can carry visual information that words alone describe poorly: framing, relative scale, material, and a mood. Begin by deciding which of these qualities matters. A reference is a direction for a model, not a guarantee that every detail will be reproduced.

Choose a model that supports your input

Model selection changes the available inputs. In Artcraft Suite, select an enabled image model and check its reference-image options before uploading. Unsupported settings are rejected rather than silently routed to a different model. Availability is controlled by the platform administrator and may change.

Keep your reference focused

Use a PNG or JPEG you have the right to use, and choose a clear composition. Avoid including private documents or unrelated personal details. A single useful reference is often easier to evaluate than an unstructured collection. Uploaded assets remain in your account library; prompts and selected references are sent to the configured generation provider when you request a task.

Make one deliberate change at a time

Describe the intended subject and the visual quality you want to explore. Generate a focused image, compare it with your brief, then adjust one variable such as framing, light or material. Keeping the model and reference stable makes the effect of that change easier to assess. Outputs are probabilistic, so treat an iteration as an experiment rather than a promised match.

Continue from the homepage into the workspace

The homepage image tool is intentionally small. It lets you sign in, choose an enabled model and start with a square image. The Continue in workspace action carries the same prompt and model choice into the full image workspace, where supported reference inputs and your private library are available.

Keep results and costs together

Use task history and the private library to revisit outputs instead of relying on a conversation thread. Review your allocated budget before submitting a new request. Registration does not grant free generation funds, and uploaded references do not bypass model or storage limits.

What this workflow adds to a generic AI chat

The difference is the workflow around the image: explicit model choice, visible input controls, private asset storage and a record of iterations. It does not mean that every model is available, that a reference will be copied exactly, or that the service has unlimited generation. Start with one clear brief and an enabled model, then judge the result against your intended direction.