Choose AI Image References for Subject, Composition and Style

For example, a photograph might define a particular desk lamp, a rough layout might place it on the right with space for a headline, and an illustration might supply a muted palette and paper texture. You want the lamp from the photograph, the arrangement from the layout and the treatment from the illustration—not an uncontrolled blend of all three scenes.

Dreamina can start from the critical subject reference and a written brief for the remaining layout and treatment. For projects that need separate visual controls, use a documented reference workflow whose inputs match those roles. The preparation below helps with either route.

Choose references for information, not for how impressive they look

A beautiful image can be a poor reference. Dramatic lighting may hide the feature you need to preserve. A busy composition may communicate a palette while also suggesting unwanted people, furniture or objects. Several attractive images can introduce more ambiguity than one clear photograph and a simple sketch.

Start with the final deliverable. Ask what must be recognizable, where it must appear and how it should look. Those are separate decisions even when a single source happens to answer all three.

Do not choose a photograph of a different lamp simply because its lighting is better. If the actual lamp's identity matters, use the actual lamp as the subject reference. You can describe or separately reference the desired lighting.

If you do not have a specific subject, a subject reference may be unnecessary. “A small reading lamp” leaves room for invention; “this reading lamp” does not. The text-first versus reference-first workflow helps decide whether you need an existing image at all.

Select the subject image first when identity is a hard requirement

Pick a view that exposes the details the output needs. For a lamp, that may include the shade profile, joint structure, base and cable. For a person, it may include facial structure, hairstyle and clothing. For a building, it may include the arrangement of windows, roof shape and entrance.

The best view depends on the planned image. A front photograph may establish a label but reveal little about the side profile. If the deliverable demands an angle absent from the source, the model must infer unseen information. Decide whether that inference is acceptable before generation.

Multiple subject views can be useful if the chosen workflow supports them, but they should show the same subject in compatible states. A photograph of an older design and a photograph of a revised design do not become more accurate when combined. Choose the authoritative version and exclude the obsolete one.

Keep a clean original. Make a working crop if the source has competing subjects, while retaining enough context to understand shape and orientation. A crop that cuts off the lamp's base removes information the model may then invent. A crop that isolates the lamp without losing its outline can make the intended subject clearer.

When exact lettering, logos or geometry are required, treat the photograph or approved artwork as production material, not only inspiration. A generative reference can guide a new rendering; it does not establish that every original pixel will remain unchanged.

Use a composition reference to describe relationships

A useful layout source communicates where things go. It can be a rough sketch, a photographed arrangement, a simple collage or an existing image whose arrangement is relevant. It does not need the final color or texture.

For the lamp example, the composition brief might be: one lamp in the right third, its base in the lower quarter, a low horizontal desk edge and an uncluttered left area for a headline. A simple sketch can communicate those relationships more clearly than a detailed photograph of an unrelated furnished room.

Match the composition reference to the intended output ratio. A square image with a centered subject does not automatically translate into a wide banner with a clear text area. Cropping can change the relationship between subject and empty space. Create the wide layout first if that is the actual destination.

Distinguish arrangement from camera viewpoint. Moving a lamp toward the right edge is a placement change. Showing it from above is a viewpoint change that affects visible geometry. If the subject reference is front-facing and the layout demands an overhead view, acknowledge that the sources do not contain the same information.

The composition reference should be as detailed as the required spatial decision, and no more. If a blank rectangle merely reserves text space, explain that it is a blank area rather than a physical object. If an arrow indicates light direction, distinguish that annotation from a visible arrow in the finished design.

If your composition source is a rough drawing, the sketch-to-image workflow shows how to carry its main shapes and empty areas into a finished image, then check whether the layout survived.

Choose a style reference that can survive a change of content

A style source should still be useful when you replace its subject. Think about the properties you would carry into an image of something entirely different: flat color shapes, grainy paper texture, low contrast, warm shadows, a limited palette or crisp studio light.

This is why “use this image's style” can be vague. You may mean its colors, while the generator also takes cues from its framing, subject or atmosphere. State the relevant properties and exclude content that does not belong.

For the lamp illustration, “muted ochre and navy, off-white paper texture, broad shapes and restrained shading” describes a style independently of the example's subject. That wording is useful even if your application offers only one reference input and the subject photograph occupies it.

Do not overload the pack with conflicting style examples. One image may use hard geometric shadows while another relies on soft watercolor edges. Those can be combined intentionally, but only if you define which quality comes from each. Otherwise, choosing one representative direction is clearer than hoping the tool will infer your preferred compromise.

Use images you have permission to upload and adapt. Permission to view an image does not automatically make it available as production input. For a style discussion, a description of general visual qualities may be sufficient without importing another creator's complete image.

Give the references a priority order

Before opening the generator, resolve the conflicts a model cannot resolve on your behalf. Consider these examples:

The hierarchy should follow the task. For an accurate product image, subject details may be non-negotiable. For a fictional scene built around a poster layout, placement and readability may come first. For open visual exploration, style may have more freedom. There is no universally correct instruction to prioritize the first image in every project.

Also decide what can give way. If the requested paper texture reduces label legibility, do you need the label readable, or is it an intentionally simplified illustration? A brief becomes usable when it contains both priorities and permitted compromises.

A worked pack: a lamp photograph, a banner sketch and an illustration

This is a planning example, not a tested generation. Suppose the goal is a wide editorial banner featuring a particular lamp, with editable headline text added later.

The pack contains:

• Subject photograph: the correct lamp, complete shade, joint and base, in an uncluttered view.
• Composition sketch: a wide frame, lamp on the right, low desk edge and clear left area.
• Style example: a permitted illustration showing the desired palette, shape language and paper texture.

The accompanying role note could read:

The lamp photograph defines the lamp's identity: retain its shade profile, articulated joint and round base. The layout sketch defines placement and empty space, not appearance. The style example guides the muted palette, broad illustrated shapes and paper texture only; do not import its objects. Keep the left area clear for text added separately.

Planning diagram for the lamp example; it illustrates reference responsibilities, not a generated result or measured adherence.

The wording does not promise perfect obedience. It makes the intended interpretation explicit and provides a way to judge the output. A result with the correct palette but the wrong lamp fails the identity requirement. A correct lamp filling the text area fails the layout requirement. A generic vector-like rendering may satisfy identity and placement while missing the intended paper treatment.

If your application accepts only one useful visual reference, start with the lamp photo because its identity is the most difficult information to recreate in words. Translate the layout and style into the brief. If layout is non-negotiable and the lamp can be generic, make the sketch the primary reference instead. The task determines which source deserves the input slot.

For the detailed language that separates these roles, see the reference-image prompt guide. Keep the pack itself small; a long prompt cannot rescue contradictory or irrelevant sources.

Choose an application route that matches the pack

Dreamina: start with the indispensable reference

The Dreamina image-to-image workflow provides a route from an uploaded image and written brief through model/settings selection to generation and download. For the lamp task, begin with the accurate subject photograph and describe the wide layout and illustrative treatment explicitly.

Use the selected model and controls actually available in your workspace. A model's general research demonstration is not the same as every application exposing three independent subject, structure and style inputs. The method here does not require assuming that distinction exists in Dreamina.

Review the first result by role: lamp identity, spatial arrangement and visual treatment. Refine the dimension that fails instead of replacing the entire pack. If the result needs a strict spatial placement that text cannot reliably achieve, use a composition-led route or arrange the approved subject in a design application.

Midjourney: distinguish image guidance, style and editing

Midjourney's Image Prompts influence content, composition and colors. Its separate Style Reference guides visual treatment rather than copying people or objects. The Edit Model offers written edits and multiple reference input.

These distinctions matter more than simply attaching all images. Put a style source in the style role rather than treating it as another scene to merge. Use the appropriate image/edit reference when the subject must be carried through. Reference influence still needs review; none of these roles makes a new rendering an exact copy of the source.

For a prearranged collage or a protected region, the Editor documentation describes layers, selections and regeneration of transparent areas. That is a different route from asking a fresh generation to reconcile several whole images.

Firefly and Photoshop: separate arrangement from object placement

Firefly documents Composition reference and Style reference as different controls. Composition concerns outline and depth; style guides look and feel. This is useful when arrangement and treatment come from different sources.

If an object must be introduced into a selected part of an existing layout, Photoshop's reference-image workflow distinguishes Object from Whole image and placing from swapping. That makes the intended operation more explicit. For strict retention of original artwork, preserve that artwork in the composite rather than interpreting reference-guided generation as pixel preservation.

Choose these routes for the control you actually need, rather than assuming that a larger number of uploaded images always produces a more faithful result.

Inspect the output by reference role

Open the result beside the sources. Review identity first when it is mandatory, then composition, then style. If your brief uses a different priority, follow that order instead. Keep an unacceptable hard-condition failure visible; an attractive finish should not distract from it.

For the lamp example, check the silhouette and joints against the photograph. Compare the lamp's position, desk edge and headline area with the sketch. Finally compare palette, shading and paper treatment with the style reference. The sources do not have to look similar to one another; each must contribute the right information to the result.

If a dimension fails, record the visible problem precisely. “The lamp is centered and blocks the left text area” identifies a layout error. “It looks wrong” does not tell you whether to change the source, prompt, reference role or editing method.

Download approved results and keep the pack plus the role note with them. That gives later revisions a stable explanation of the design instead of forcing someone to infer intent from the finished picture alone.

Troubleshoot a reference pack that is not working

Objects leak in from the style reference

Remove unnecessary content from the source or describe the intended style in words. Where the application provides a dedicated style role, use it. Avoid assigning the same source every possible reference role merely to make its influence stronger.

The output blends two versions of the subject

Check the pack for contradictory designs, states or angles. Select the authoritative version and remove the other. If both views are genuinely needed, explain that they show the same subject from different angles; do not ask the model to choose which physical details are current.

The layout changes despite a detailed reference

Confirm the output ratio and whether the image entered a composition control or only a general inspiration input. Simplify the layout source to its essential masses. If precise placement remains mandatory, use an arranged base, layers or manual layout instead of adding more unrelated references.

The right subject appears in the wrong style

Check whether the subject source's lighting or photographic detail is overpowering the requested treatment. State the style through observable qualities. If your tool exposes separate style influence, adjust it deliberately. Avoid weakening an essential identity constraint merely to make the rendering more decorative.

You need both a photo and an illustration from the pack

Keep the same identity source and change the treatment deliberately. The matching photo-and-illustration workflow explains when both outputs should use the original reference and when an approved composition should become the second input. That decision comes after you have resolved the pack's priorities.

End with a clear brief, not a folder of inspiration

A useful reference pack contains only the images necessary to communicate the required subject, arrangement and treatment. Each has a stated role, each conflict has a priority, and the final output has a visible acceptance standard.

Start in Dreamina with the reference carrying the hardest-to-describe requirement. Add the other directions through the brief or an appropriate documented control, and judge the result against each source's actual job.