AI Reference Images Explained: What They Actually Do

Attaching an AI reference image tells the generator what "warm nostalgic palette" or "soft ink-wash linework" looks like, without you writing a paragraph to describe it. Used well, a reference steers style, palette, composition and mood; it does not, and should not, guarantee an exact copy of the photo you upload. Here's how to use one to get a deskmat that actually looks like what you had in mind.

Reference images matter because language is a lossy way to describe a look. You can type "cinematic teal-and-orange grade" and get something close, or you can point at an image and let the generator pull the actual relationship between those colours. That's the whole value of style transfer for mousepads: it shortcuts description with demonstration.

What a reference image actually influences

• Palette — this is the strongest and most reliable transfer. Dominant hues, contrast between light and shadow, and overall warmth or coolness carry across well.
• Style and medium — brush texture, linework, grain, cel-shading versus painterly rendering.
• Composition — where the mass of the image sits, how much negative space surrounds it.
• Mood and light — time of day, softness of light, high or low contrast.
• Rough structure — general layout of large shapes, not fine detail.

What a reference will not do

• It will not reproduce a specific person, character, or logo pixel-for-pixel.
• It will not deliver an identical duplicate of the uploaded photo — it's a steering input, not a stencil.
• It will not fix a resolution problem. Output size is set by the canvas you choose for your pad, not by how large the reference file is.
• It will not override a prompt that says the opposite thing. If your text says "bright noon" and your reference is a night shot, the two fight and the result tends to look muddled.

If you're deciding between working from a reference at all versus starting from a straight photograph, read [AI custom mousepad vs photo print](/blog/ai-custom-mousepad-vs-photo-print) first — a reference-guided AI generation and a straight photo print behave very differently on fabric at desk-mat scale.

Good reference photos vs bad ones

What makes a good reference

A useful reference has one clear dominant palette, an obvious light direction, and a simple composition — one or two masses, not fifty scattered elements. A single still frame of a foggy forest at dawn, a photo of your own painting, or a clean piece of concept art all work well.

What makes a bad reference

• Cluttered collages with dozens of competing elements
• Heavily watermarked images (the model can pick up the watermark texture)
• Tiny thumbnails with visible compression artifacts
• Screenshots with UI elements in frame — the generator will happily learn the UI along with the scene

What transfers from a reference, and how reliably

Combining a reference with a text prompt

The most effective approach divides the labour: let the reference carry look, and let the prompt carry content and layout. Trying to make one input do both jobs produces vaguer results than splitting the work.

Reference: a foggy pine forest at dawn, muted sage and cream palette

Prompt: "A single wolf standing in profile on the right third, open mist filling the left two thirds, match the reference palette and soft dawn light, ink-wash rendering, no text"

The reference supplies atmosphere and colour relationships. The prompt supplies subject, placement and rendering style. Neither one has to guess at the other's job. If you want a deeper walkthrough of how to phrase the text half of that pairing, see prompt engineering for mousepads.

How many references to use

One reference is usually the right number. Two can work if each is doing something distinct — one supplying palette, one supplying composition — and you say explicitly which does what in your prompt. Beyond two, influences average out into something generic rather than combining cleanly.

Reference resolution

A blurry or low-resolution reference will not lower your print quality. Output resolution on your finished pad is determined by the canvas size you pick and how the piece is generated for that size, not by the pixel count of the file you uploaded as a reference. That said, a blurry reference can still muddy the style signal — heavy compression artifacts or tiny thumbnails give the generator less clean information to pull a palette or mood from. Prefer a reasonably clean, in-focus reference even though its resolution won't cap your output.

Iterating from a reference

Treat a reference as the opening move, not the final answer. A typical sequence looks like:

1. Attach the reference and write a prompt describing subject and composition.
2. Generate and review against the reference — did the palette and mood transfer the way you wanted?
3. Refine specific elements ("push the shadows toward the reference's teal, keep the subject unchanged") rather than regenerating from scratch each time.
4. Repeat until the balance between reference influence and your own subject feels right.

If refinement itself is unfamiliar territory, how AI refinement works covers exactly how those follow-up edits are interpreted and what gets preserved between passes.

Reference strength as a dial

When you can control how strongly a reference applies, think of it as three settings:

• Low — a hint of palette and mood, your prompt does most of the work.
• Medium — a clearly recognisable stylistic family, with your content driving the specifics. This is the useful default for most designs.
• High — a close visual match to the reference, with less room for your prompt to steer. Useful occasionally, but designs pushed this hard start to feel less original and more like a copy attempt.

Only use artwork you have rights to

This is a plain, practical rule, not legal advice: only upload a reference image you actually have the right to use. That means your own photography, your own artwork, images you've licensed, or material that's clearly public domain. Uploading a screenshot of someone else's copyrighted illustration or a frame from a trademarked franchise as a reference doesn't sidestep rights concerns just because the output isn't identical — the safest approach is to describe the style you want in words ("cel-shaded shonen action," "90s mecha anime") and reserve reference uploads for material you own or are licensed to use.

FAQ

What does a reference image do in AI mousepad design?

It transfers style, palette, mood, light and rough composition to your generated design. It does not duplicate content outright and cannot reliably reproduce specific characters or logos.

Can I upload a photo of a character I like as a reference?

Avoid it. References are meant to steer look and feel, not to reproduce identifiable or trademarked characters, and doing so raises the same rights issues as prompting for a named franchise directly.

Does a low-resolution reference reduce my print quality?

No. Output resolution is set by the canvas size you select for your pad and the generation itself, not by the reference file's resolution. A blurry reference can still weaken the style signal, so a clean one is preferable even though it won't cap your final print size.

Should I use a reference image or just write a longer prompt?

Use a reference for things that are hard to put into words — palette relationships, grain, atmosphere — and use words for content, subject and layout. Combining both, each doing one job, consistently outperforms relying on either alone.

Can I use my own artwork as a reference?

Yes, and it's one of the fastest ways to keep a generated design recognisably yours — your palette, your line weight, your composition habits carried into the final piece, then refined further from there.

How many reference images should I upload at once?

One is the safest default. Two can work if they're each doing a distinct job and you say so in your prompt. More than that tends to blur into something generic.

Ready to put a reference to work? Head into the studio, upload something you have the rights to, pair it with a focused prompt, and refine from there until the palette, mood and composition land exactly where you want them — then check pricing when you're ready to print it on your desk mat.

Common reference mistakes to avoid

A few habits quietly undermine reference-guided generations, and they're worth checking before you upload anything.

Uploading a reference and a contradictory prompt. If your reference is a bright, high-key photo and your prompt asks for "moody, dim lighting," the two inputs pull against each other and the result often lands in an unsatisfying middle ground — neither bright nor properly moody. Keep the reference and the prompt pointed in the same direction, or explicitly note which one should win on a given attribute.

Treating a reference as a shortcut for a weak prompt. A reference can carry palette and mood, but it can't invent your subject or composition for you. "Make it look like this" with no other description leaves too much ambiguity about what you actually want the pad to show.

Expecting exact duplication. This is worth repeating because it's the most common disappointment: a reference steers the generation, it doesn't clone the source image. If pixel-for-pixel accuracy is the actual goal — a specific real photo you want on your desk exactly as captured — a straight photo print is the better tool than an AI generation guided by that photo as a reference. AI custom mousepad vs photo print walks through when a photograph should just be printed directly instead.

Ignoring the pad's aspect ratio when choosing a reference. A tall portrait-oriented reference photo carries composition information that doesn't map cleanly onto a wide XL or XXL desk mat. If your reference has a strong vertical composition, expect the prompt to need extra work describing how that translates into a wide layout, or pick a reference that's already closer to your pad's proportions.

A practical reference-to-print workflow

Putting all of this together, a workflow that consistently produces good results looks like:

1. Pick your pad size first, so you know the aspect ratio you're composing for.
2. Choose one clean reference image you have the rights to use, with a clear palette and simple composition.
3. Write a prompt that handles subject, placement and any style details the reference doesn't cover, explicitly telling the reference and prompt what job each is doing.
4. Generate, then compare the result against the reference — check that palette and mood transferred without the composition collapsing into the reference's original shape.
5. Refine one attribute at a time rather than regenerating from scratch, using the reference as a constant anchor point across passes.
6. Once the balance feels right, review the final design on the actual pad mockup at your chosen size before ordering.

This sequence keeps the reference doing what it's good at — communicating a look that's hard to write out — while your prompt and refinements do the work of shaping the actual content on the pad.