Choose a medium, a palette and a composition for an illustration or artwork, then review how they work together in the result.
Try an example
The same lighthouse on a low headland, generated twice. Only the medium changed, from loose watercolor to dense ink, so the differences come from the treatment and not from a new scene.
The lighthouse, the headland, the horizon line and the 3:2 frame are the same in both images. Only the medium word in the prompt was swapped, which makes the two results comparable at a glance.
Watercolor-like washes tend to soften edges and let the tower blend into the sky. Ink adds hard contours and hatching, and the dark rock gets heavier. Texture, edge and value structure moved; the composition did not.
Does the lighthouse stay the focal point? Are the dark areas too heavy for the layout? Does the texture survive at the final size? Judge each image in its real placement, and remember the file is a raster image, not editable strokes or vector paths.
Each of these jobs needs a different check. Describe the look in terms you can see, then judge the image where it will sit.
3:4 · Chapter opener
Pick the one image that matters to the chapter, such as a lantern by a door, and keep a few objects so the focal shape stays readable in a narrow column. Describe the space reserved for the title, then add the exact wording in your layout app, not in the generated pixels.
16:9 · Game environment
Describe the scale of the buildings, the materials, the weather and the viewpoint, and add a small figure for scale. The result shows mood and broad layout, not a measured map or game-ready assets. Check doorways, stairs and bridges, which can look plausible but fail to connect.
2:3 · Poster art
Start from one dominant shape or scene, a limited palette and strong light-dark contrast. Ask for open space where the type will go and set the words later. Check the margins and corners for stray letter-like marks, and test a tall poster crop against a square one.
Three controls in the creation card above.
Write the subject, viewpoint, lighting, medium and palette in up to 4,000 characters, in terms a reviewer can check. Add up to 14 reference images only when each one answers a question, such as subject, composition or visual direction, and say in the prompt which job each image has.
Choose Nano Banana or Nano Banana Pro, a supported ratio, HD, 2K or 4K, PNG or JPEG, and one, two or four images. Set the ratio first: 16:9 suits a banner and 3:4 a cover. Several candidates help only when each tests a clear question.
Nano Banana HD costs 23 credits per image and Nano Banana Pro HD 45, shown on Generate before you confirm. Review the whole image and its key detail at full size, change one or two decisions per iteration, and keep the prompt with the result you save.
Four artistic choices, each set in words and checked in the result.
01
Loose watercolor, dense ink, gouache and digital paint each leave visible marks. Name the medium in the first sentence of the prompt so the AI art generator treats it as the main instruction rather than a detail. The same scene in two media is the quickest way to see what a medium changes.
02
A restricted palette keeps a picture from drifting. Name two or three main colors, say “restrained cool colors”, or ask for bold flat color when you want a pop-art look. Check whether the accents still read at the size where the artwork will be shown.
03
Artwork for a poster or a chapter page needs empty areas. Say where the open space goes, such as the top third for a title, and where the main shape sits. A lighthouse on a headland is arranged differently in a 3:2 scene than in a 2:3 poster. Add the lettering itself in your layout tool.
04
Attach up to 14 reference images when a character, an object or a look should carry over, and say in the prompt what each one contributes. The AI art generator then combines the reference with the medium you named. Review the result against the references, because details can still shift.
Pick the page that matches the artwork you want to make next.
An AI art generator, sometimes called an AI art maker, produces artwork from a written description and optional references. The request can describe a subject, composition, palette, and medium-like treatment. Potto AI provides generated raster images to review and download. The result is not a physical original or an editable collection of brush strokes.
Describe visible qualities such as color washes, hard outlines, hatching, paper texture, limited colors, or flat shape hierarchy. For example, ask for loose watercolor washes with soft edges, or for dense ink contours with cross-hatching and heavy dark rock. Connect those qualities to the artwork’s purpose. Keep the subject stable when comparing media so the outputs answer the same visual question.
The normal download is a raster image, not vector paths or an editable layer document. Use a design workflow to build those deliverables if they are required. Generated lettering and object detail should also be checked before a production handoff; an art-like appearance does not establish file editability.
Generated lettering can contain mistakes or stray marks, so add exact words in your layout app, where spelling and placement stay editable. Ask for open space where the text will go, and check the margins and corners for letter-like marks before you place the image.
Check the actual pixel dimensions of the downloaded file against your printer’s requirements before you order. The HD, 2K and 4K labels are output tiers, not a print specification, and the generator does not add bleed or an editable print layout. Review a proof at the intended size, because screen brightness and paper change how colors read.
Up to 14, subject to the selected model and the upload requirements shown in the interface. Say what each image supplies: a subject, a composition, or a visual direction. Several images do not guarantee that every detail is preserved, and conflicting references can produce an uncertain blend.
Save the prompt with the selected result, then note what still needs work, such as a crowded foreground or colors that compete with a headline. Change one or two related decisions in the next request so you can see what caused the difference. Each revision is a new request that can use credits again. When you hand the file to a teammate, include the placement, the crop, and the prompt for context.
You can write the art prompt, add references and choose settings before signing in. Generate opens the sign-in dialog, and the prompt stays in the AI art generator.
A failed artwork shows the reason and a retry, and its credits are released. Stop a running request with the cancel control. Generating another version of finished artwork is a new, charged request.
Each artwork costs 45 credits with Nano Banana Pro at HD or 2K, or 23 with Nano Banana at HD; 4K and more images cost more. The total is on Generate before you confirm.
Describe the medium, palette and composition, then review the result in its real layout. Check the quote on Generate before you start.