One Reference, Two Styles: Create a Matching AI Photo and Illustration

Start by deciding what the two images must share

“Make a photograph and an illustration of a green bottle” leaves the model free to invent two different bottles. A reference supplies visual information, but the instructions still need to distinguish identity from style.

Think of the pair at three levels:

For an editorial header, the first level may be enough: a realistic subject on one side and its graphic interpretation on the other. A before-and-after slider demands more alignment because viewers compare the outlines directly. A product listing requires a different standard again; an attractive approximation of a label can still misrepresent the actual item.

Write down the few features that carry recognition before generating. For the bottle in this guide, they are its long green glass body, open neck, three label areas and prominent red star. For a person, they might be hairstyle, face shape, glasses and a particular jacket. For an interior, they could be the window arrangement, dominant furniture and viewpoint.

Then identify what each style is allowed to reinterpret. Photographic glass needs reflections and transparency; a flat illustration may communicate the same glass with overlapping green shapes. Requiring identical reflections in both would fight the illustration brief without improving recognition.

Which AI workflow should you choose?

The relevant comparison is how each tool carries the source into both versions, how you direct the style change and how you revise a mismatch. There is no need to choose one generator for photographs and another for illustration simply because the outputs look different.

The Dreamina example below shows the two outputs from one reference. The alternatives are compared through their documented controls rather than a shared image-quality test.

Dreamina: one reference, two clearly specified briefs

Dreamina's image-to-image workflow accepts a reference and written instructions, then lets you choose a model, aspect ratio, output quantity and resolution before downloading the result. That gives the pair an explicit shared setup: the same source, model and format, with the rendering brief changed between versions.

Its strongest reason to consider it here is concrete: the two outputs below were generated with Seedream 5.0 Pro from the same uploaded photograph. The studio branch interprets materials and lighting; the illustration branch simplifies those materials into flat shapes. You can use that division when a subject needs two visual treatments without having to rebuild its description from scratch.

ChatGPT Images and Gemini: useful when the brief develops through conversation

ChatGPT Images supports uploading an existing image and describing changes, then requesting further edits. For this task, explicitly identify whether the next request should use the original source or the newly approved image. “Illustrate the original bottle” and “illustrate this studio composition” ask for different reference relationships.

Choose this route if you want to discuss the intended look and refine it in the same conversation. For a local correction, its editor also offers selections, though OpenAI notes that edits can extend beyond the highlighted area. After a repair, inspect the whole pair rather than assuming everything outside the selection stayed fixed.

Gemini Apps likewise supports edits to uploaded or generated images, including multiple-image input. It is a reasonable alternative for keeping the pair in an existing Gemini workflow. With either conversational route, name the source and allowed changes clearly; a continuing conversation does not by itself define which details must survive the style conversion.

Midjourney: reference transformation with deliberate style direction

Midjourney's current Edit Model documentation describes written image edits, generation with up to four references and conversion into a different style. It also supports Style References and other aesthetic guidance. You can pin an uploaded image in the Imagine bar to reuse it across prompts.

This makes it a relevant option when the illustration's visual language needs more direction than “make it artistic.” Keep the subject reference separate in purpose from a style example: one defines what is depicted, the other guides how it looks. The Edit Model tries to match the first uploaded image's aspect ratio unless you specify another, so set the intended ratio deliberately for the pair.

Firefly: choose it when composition is the main constraint

Firefly offers a particularly clear distinction between structure and appearance. Its Composition reference uses outline and depth to guide arrangement, with a Strength slider for adherence. Its Photo and Art content types provide realistic or illustrative variations.

Use these controls when a shared layout matters: upload the same composition reference, keep the framing instruction stable and change the content type and style description. Composition adherence concerns arrangement; it should not be treated as a guarantee of exact facial identity, lettering or product geometry. If the subject is correct but both versions must occupy the same space in a design, this route gives you a specific control to work with.

A real example: the same bottle in a studio image and an illustration

This example uses the saved prompts, settings and original outputs from a recorded browser demonstration.

The following outputs were generated on October 14, 2026 using a Basic account, Seedream 5.0 Pro, a 3:4 ratio and the Standard 1.5K setting. Each brief produced one image, downloaded as a 1296 × 1728 JPEG. Both generations used the original restaurant photograph; the illustration was not generated from the studio output.

Left: original reference. Center: generated studio-style output. Right: generated editorial illustration. Both outputs retain recognizable visual anchors while redrawing fine detail. Source photograph: Philafrenzy, “San Pellegrino 500ml bottle,” Wikimedia Commons, CC BY-SA 4.0. The AI-modified adaptations and comparison layout are shared under the same license. Product branding identifies the example subject and does not imply endorsement.

The exact studio-image prompt

Use only the complete central green glass bottle in the reference. Place it on a neutral light-gray studio surface. Keep the open uncapped neck, bottle proportions, three labels, visible brand lettering, red star and 500 ml marking. Exclude the partial bottle and all tableware. Soft even lighting, photorealistic, no extra text.

The prompt separates three jobs. It identifies the complete central bottle, lists the details intended to remain and replaces the restaurant context with a gray studio setting. It also excludes visible distractions rather than relying on the model to guess which bottle matters.

The output keeps the uncapped green bottle, three label areas, red star and readable 500 ml marking. The partial bottle and tableware are absent. At the same time, the glass highlights, label curvature, small lettering and perspective have been redrawn. The result is a generated studio-style interpretation, not a record of a new physical photo shoot or a pixel-preserved extraction of the original bottle.

The exact illustration prompt

Create a clean editorial illustration based on the complete central bottle. Retain its overall silhouette, green color, open neck and red star as visual anchors. Use flat shapes and a soft cream background. Exclude other bottles and tableware.

Here, the brief deliberately relaxes detail. It prioritizes silhouette, green color, open neck and red star rather than requiring every label character. The output uses simplified green areas and a cream background, while the neck, shoulder label and red star still connect it to the photographed subject.

The illustration also contains distorted small lettering. That matters if your intended use is packaging reproduction; it is less decisive for an editorial illustration whose role is to identify the subject at a glance. This pair retains recognizable subject features across two styles, with differences in geometry and fine detail.

The practical lesson is to keep the recognition brief consistent while allowing each rendering to do its own work. The two prompts do not need identical wording. They need agreement about the subject and its defining features.

Build your own matched pair in Dreamina

1. Prepare a source that makes the subject unambiguous

Choose an image with the important subject visible at a useful angle. If the bottle's open neck matters, it should be visible; if a person's glasses identify them, avoid a source where hair or glare hides the frames. A prompt can request a missing detail, but then the model must infer it.

Use a reference you have permission to upload and adapt. Keep an untouched copy, then make a working crop if several objects compete for attention. Cropping is an input-preparation option, not a required step: the bottle example used the broader restaurant image and identified the central bottle in words.

Decide on the intended layout now. For side-by-side images, choose the same output ratio for both. For a slide or website card, leave enough room around the subject for the final crop. Matching ratios help the pair fit a layout, but they do not force identical subject scale or position. If several images supply different parts of the brief, assign separate roles to subject, composition and style references before generating.

2. Separate the fixed details from the two styles

Before writing full prompts, make a short brief with three parts:

• Identity: the particular subject and its essential visible features.
• Composition: the intended angle, crop, subject size and placement, if they need to match.
• Treatment: lighting and material language for the photo-style image; line, shape, texture and palette for the illustration.

Be selective about identity details. “Preserve everything” does not explain whether the restaurant background matters or whether a label can become a simplified color block. A small, explicit set gives you a usable acceptance standard after generation.

For a portrait pair, for example, identity might include round glasses, a chin-length haircut and a red jacket. Treatment might change from soft window-lit photography to a two-color ink illustration. Those are planning examples; the bottle outputs above come from their own exact prompts.

3. Generate and approve the photo-style version

Open Dreamina's AI Image workspace and upload the reference. Select Seedream 5.0 Pro when available in your account, then choose the aspect ratio, resolution and output quantity for the intended placement. Check the displayed credit requirement before generating; the case settings describe that particular run rather than a universal account allowance.

Describe a photographic result through visible properties. “Soft, even lighting, neutral gray surface, green transparent glass, full subject in frame” is easier to evaluate than “premium professional quality.” Add the identity details and exclusions from your brief, then generate.

Inspect the result before moving on. Check the silhouette and defining features first, then material, perspective and unwanted additions. If the neck or a face has changed enough to depict a different subject, reject that version even if the lighting is attractive. If the subject is acceptable, download and retain the approved image so it remains the reference point for later decisions. When one feature needs a correction, use a local-edit process that also checks the surrounding image before approving the result.

4. Choose the source for the illustration branch

Reuse the original reference when subject identity is the main requirement. This is the route demonstrated by the bottle example. Upload or retain the same original input, keep the identity instructions and replace the photographic treatment with a specific illustration brief.

Describe what makes the illustration illustrative: flat shapes, ink outlines, paper texture, limited colors or simplified shading. “Matching illustration” alone says little about whether you want a vector-like editorial image, a watercolor or a detailed painted rendering.

Use the approved photo-style image when its new composition must carry over. Suppose the photo has placed the subject off-center with room for a headline, and the illustration must occupy the same layout. Referencing that approved image is a reasonable next route because it supplies the composition you want converted. State which pose, placement and scale should stay fixed.

This sequential route also carries a tradeoff: any incorrect detail in the approved photo can become part of the next reference. Correct an identity error before using it as the foundation. The pictured case establishes the separate-source route; it does not establish identical-layout performance for this sequential variation.

5. Judge the pair together, then inspect the details

Place the two downloaded images side by side at approximately their intended display size. Ask whether they identify the same subject and whether their visual weight works together. A bottle that fills nearly the whole frame in one image may look unrelated to a tiny bottle surrounded by empty space in the other, even if both preserve its shape.

Next inspect each at full size. Check distinguishing colors, facial or object features, lettering and any features added during rendering. Small errors that disappear in a thumbnail can become prominent in a full-width header or printed page.

Use the original reference as the deciding comparison. Two generated images can agree with each other about an invented detail. Matching each other is useful, but it is not proof that either faithfully represents the source.

6. Finish and export for the actual placement

Download both final images and keep their prompts, reference and approved versions together. Apply the final crop and sizing as a pair. If you add a headline, caption or logo, place it as editable design content so it can be corrected consistently across both versions. To extend the pair into a larger set, plan a coherent image series with a different role for each frame.

An illustration that looks vector-like can still be a raster image. The example outputs are JPEGs; they do not contain editable vector paths or transparent backgrounds. If the deliverable requires vectors, prepare those in an appropriate illustration tool. If it requires a transparent subject, create and inspect that export separately rather than assuming a plain background is transparent.

When the two versions do not match

The illustration looks like a different subject

Check whether the prompt named the defining features or only the general category. “A green bottle” allows many valid designs; “the complete central bottle, with its long neck, three label areas and red star” narrows the intended identity.

Return to the original reference, restate the missed anchors and remove unnecessary style directions that compete with them. If the source contains several similar subjects, crop to the intended one or identify it more clearly. If the output still loses a defining feature, change the route or finish that feature manually rather than accepting a different subject as a match.

The subject matches, but the crop and pose differ

This is a composition problem. First see whether a simple final crop and scale adjustment can align the images without cutting off essential features. If the angle or pose differs, cropping cannot solve it.

Use the approved composition as the next reference and specify its placement and viewpoint. Firefly's Composition control is another documented route when arrangement is the deciding constraint. For exact registration, such as an overlay transition, manual tracing or controlled compositing gives you a way to enforce alignment instead of relying entirely on a new generation.

The illustration keeps too much photographic detail

Specify the simplification you want: fewer tonal steps, broader shapes, reduced reflections or a limited palette. Merely adding “illustration” can leave the rendering close to a painted photograph.

Keep the identity anchors while relaxing material detail. In the bottle example, the glass becomes green shapes; the star remains a distinct red shape. That is a useful division between what communicates style and what identifies the subject.

Logos or small text change

The bottle example shows why this needs a separate decision. Prominent identifiers can survive while fine characters are redrawn. A closer overall match does not automatically make that lettering correct.

If exact wording matters, use approved text or logo artwork in the final composition, or retain the relevant original region when the design permits. For product work, the broader guide to product-reference workflows explains the distinction between campaign concepts and product imagery that must accurately represent the item. Resolve that production requirement before choosing the most attractive pair.

Questions worth settling before you start

Should I ask for both styles in one generated image?

Generate separate files when you need independent crops, exports or revisions. A single diptych can be useful as a concept, but changing one half then becomes an additional editing task. Separate versions let you approve the photo and illustration individually, then arrange them in your final design.

Can I turn an illustration into the photo-style version instead?

You can use an illustration as the reference and request a photographic interpretation. Expect the model to infer materials, lighting and details the drawing does not specify. For an invented character, that can be part of the creative task. For a real object or person, use a photographic source when faithful representation matters; a simplified drawing cannot supply all the missing physical detail.

Do the backgrounds have to match?

No. The gray studio and cream illustration backgrounds in the example help establish different treatments. Match the background when it is part of the layout or subject identity; otherwise, coordinated tones and similar subject placement can connect the pair without forcing the same setting.

Is a photorealistic AI image an actual photograph?

It is a generated image with a photographic appearance. Use a real photograph when the image needs to document a specific object, event or condition. A visually convincing reinterpretation can be useful creatively without serving as documentary evidence.

Create a pair with a clear visual relationship

For a recognizable subject in two styles, start with Dreamina, one reference and two briefs. Keep the defining features stable, let each style reinterpret texture and light, and compare both outputs against the original. Move to an approved-composition reference when alignment becomes the priority; preserve original artwork when exact details become the priority.

Open Dreamina and create your first pair. Begin with the photo-style version, approve its subject, then build the illustration around the same visual anchors.