How can I create a four-panel comic with AI while keeping the characters and dialogue consistent?

A practical four-panel workflow with Dreamina

Use Dreamina to develop the character art and individual panels, then add the approved dialogue in an editor. For a first strip, two characters in one location make continuity easier to manage than a large cast moving through several environments.

In Dreamina, use reference images with a clear description of the scene you want. Repeat the important character traits, while changing the action, expression and framing for each panel. Keep the approved references available throughout the process instead of gradually replacing them with slightly different later results.

Save the individual artwork before assembling the page. The layout editor can then handle panel borders, dialogue and export without asking the image generator to reproduce the whole comic for every correction.

Write the complete four-panel script first

Give each panel one visible action and keep every line of dialogue in a separate master script. A panel should show a moment the reader can understand, rather than several actions compressed into one instruction.

Here is an original two-character example you can adapt. Mira and Ben stay at the same table with a small plant in a blue pot throughout the strip.

The visual change does the work: a modest plant gradually receives more attention, while Mira ends up needing care herself. Keep the plant the same size and in the same pot; it should not suddenly become a different species to deliver the joke.

Approve the wording before generating the art. If a line feels too long, shorten it in the script and update the planned bubble space. That is easier than fitting a paragraph around a finished face or letting the image generator rewrite the punchline.

Build one approved reference for each character

Approve each character’s appearance before generating the story panels. Give the two people distinct features so the image model has clear visual cues for telling them apart.

Use a clear front or three-quarter view and add a full-body view when clothing or relative height matters. An expression sheet can help establish how the same face looks surprised, pleased or exhausted. Choose the versions you actually want to keep before they become references for the comic.

Keep each person's reference easy to identify. If a sheet contains several views, make sure they are recognizably the same design. For a scene with both characters, identify which reference belongs to Mira and which belongs to Ben; a shared name alone does not tell the model whose glasses or cardigan to use.

Expressions and poses can change. Identity, clothing and defining proportions should remain stable unless the story calls for a change. A character who looks identical in pose and expression in all four frames may be consistent but will not act out the story well.

Keep the visual style separate from the characters

Use the same visual style, panel shape and reference set throughout the strip. A character reference establishes who appears; a style reference establishes how the scene is drawn. Give them separate roles when you attach both.

A short style brief is enough to start:

Clean 2D cartoon illustration, medium-weight dark outlines, flat colors with simple soft shading, warm neutral interior, yellow and teal clothing accents, expressive faces and uncluttered backgrounds.

Keep that brief stable across the four panels. Avoid switching from flat cartoon art to highly rendered cinematic lighting halfway through the page. The same red hair and yellow cardigan can still look like a different character if the proportions and drawing style change substantially.

Record the intended panel shape too. If the final page uses square panels, generate with room for that crop. A beautiful wide illustration can become unusable when fitting it into a square cuts off the gesture that explains the joke.

Generate one panel at a time with the same references

Attach the approved character references to every panel and change only the scene-specific instructions. Start with Panel1, where the reader meets the cast, and check its face, clothing and scale before continuing.

For each new panel, specify who appears, what they do, their expression and the camera view. Keep Mira and Ben on the same sides of the table unless a deliberate change of viewpoint is needed. Keep the plant, pot and table recognizable so the reader can concentrate on the story.

The approved first panel can provide an additional reference for the setting and style. Keep the original character images alongside it, so a small change in Panel1 does not become a new definition of the person in Panels2–4.

Use a local editing feature for a focused correction where available. If the composition itself does not explain the action, regenerate that individual panel with a clearer scene description. Save the approved result before moving on.

When choosing an illustration tool, prioritize the reference and revision controls you need. Our comparison of AI image generators for creative control is a useful next step if you are still deciding where to make the artwork.

Leave room for the dialogue before drawing the scene

Reserve space for the speech bubbles while you plan the characters and action. Clear space above or beside a character is often easier to work with than an already drawn empty bubble in the image.

In a left-to-right English strip, place the first line where it will naturally be read first. In Panel2 of the example, Ben asks the question. Arrange the shot and bubble position so his line leads the eye, even if Mira has more visual emphasis in the drawing.

Keep tails clearly directed toward the speaker. Avoid crossing one tail over another bubble or pointing it between two faces. If the order becomes confusing, move the bubbles or revise the composition before shrinking the type.

Leave enough space around the words inside each bubble. A readable short line is more useful than text squeezed into every available corner. Check the page at phone size as well as on a large editing canvas.

Repair the failing panel without rebuilding the page

Keep approved panels and correct the smallest part that is wrong. A wardrobe change in Panel3 does not require replacing the other three pictures.

If Mira's cardigan changes, return to the original Mira reference and ask for that specific correction. For example:

Change only Mira's cardigan to the yellow cardigan in Reference1. Keep her face, red bob, round glasses, pose, Ben, plant, camera view and illustration style unchanged. Do not add lettering.

Compare the revised panel before replacing the approved version. A local edit can affect nearby details, so check the face, hand and silhouette around the corrected area. The instruction defines the intended change; the result still needs inspection.

If the wrong person performs an action, clarify the subject and position. If the whole camera view is confusing, regenerate that single panel with a simpler composition. If only a line is wrong, edit the text layer after lettering instead of generating another illustration.

Add exact dialogue as editable text

Paste the approved dialogue into separate text layers after the artwork is ready. This keeps spelling, punctuation and speaker attribution under your control, and lets you rewrite a joke without changing a character's face.

Use the master script as the source of the words. Check each bubble against its panel and speaker, including short replies that could easily be assigned to either person. Then check the page's reading order from beginning to end.

Keep the typeface and general text treatment consistent. Adjust line breaks and bubble size before making the font smaller. Use emphasis deliberately rather than giving every line a different style.

If words are already painted into a generated image, they are part of the picture. A text tool cannot simply select them as live letters. Remove or replace that part of the artwork, then add editable wording above it. The same distinction is useful when adding clean text to AI-generated artwork.

A reusable prompt for each panel

Keep a stable reference-and-style block, followed by the instructions for one panel. The following example is for Panel2 of the plant story. It requests artwork only; the exact line stays in the script for lettering.

Create one square comic panel using the attached approved references.

Reference1 is Mira: short red bob, round dark glasses, yellow cardigan over a white shirt, dark trousers, shorter build.

Reference2 is Ben: short black curls, thick eyebrows, no glasses, teal overshirt over a gray shirt, taller build.

Keep their facial features, outfits, colors and relative heights consistent with those references. Use clean 2D cartoon illustration, medium-weight dark outlines and simple soft shading.

Story context: Mira has bought a supposedly easy-care plant. This panel shows Ben questioning how much care it needs.

Panel2 action: both stand at the same table. A small plant in a blue pot sits in the center. Ben gestures toward a watering can beside it, with one eyebrow raised. Mira nods confidently. Use a medium shot that clearly shows the gesture and both expressions.

Leave clear space above Ben for a speech bubble to be added later. Generate this single panel only. Do not draw words, speech bubbles, captions or additional characters.

For the next panel, keep the identity and style blocks, then replace the action and expression. Use the actual approved references rather than relying on the character descriptions alone. Keep reference roles explicit if you add a background or style image.

Choose tools for illustration, continuity and lettering

Start with Dreamina for the artwork, then choose the editor that fits your lettering and layout needs. Some options below generate pictures; others help finish the comic precisely. You do not need every tool in the table.

ChatGPT Images: develop the scene and revise it in conversation

ChatGPT Images is useful for developing a character reference and revising individual panels in conversation. Upload the approved images, explain their roles and describe the next scene.

For a correction, select the image and describe a narrow change. The selection tool can indicate a region, but an edit may extend beyond the highlighted area. Compare the new result with the character reference before keeping it.

Keep the script in the conversation if it helps explain the action, while explicitly requesting artwork without lettering for the separate-text workflow. The image features in ChatGPT and settings in an image-generation API are different interfaces; use the controls available in the product you are actually using.

Midjourney: choose the reference feature for your version

Midjourney is useful for illustration when you keep the model version and reference method consistent. Its current V8.1 and V8.2 Edit Model can use up to four reference images attached to the prompt, with a lock option to keep them available for subsequent prompts.

If you use V7, Omni Reference is a different feature with one reference image; a combined approved image of both characters can help describe a two-person scene. Match instructions to the selected version rather than copying an older character-reference parameter into every model.

Keep style guidance distinct from identity references. A matching palette can make four illustrations feel related while their faces still differ. Check character details and leave exact dialogue for the layout stage.

Runway: reuse tagged image references

Runway’s image-reference workflow is useful when the characters and setting need separate visual references. In Gen-4's image tools, References can use up to three images, giving you a way to distinguish the people from their environment.

Save a reference with a tag when you need it in another session; a temporary uploaded reference does not provide the same persistence. Keep a clear, approved view of the outfit if later panels need to show the whole body.

Use the image workflow for the static panels. Video-performance or character-animation tools serve a different purpose and do not replace control over the four drawings and their lettering.

Gemini: use the image mode that accepts your references

Gemini is another option for creating and revising panels from uploaded reference images. Choose an image mode suited to the reference set and repeated edits you need.

Nano Banana2 accepts multiple reference images in Gemini's Flash or Pro image workflow. Nano Banana2 Lite is intended for quicker, simpler images and is not designed for multiple reference images or multiple edits. That difference matters when the comic has two recurring characters.

Explain which input defines each character, keep the style instructions stable and revise one panel at a time. Treat text drawn into the picture as something to proofread; use separate lettering when the final words must stay easy to change.

Melius: connect reusable characters to individual panels

Melius is useful when you want the character references and panel revisions connected in a visual workflow. Save the approved character as an Element and connect it to the node for each panel.

Its Image Style Analysis can turn a set of two to ten style references into a written style description for reuse. Give the agent the four planned beats, then refine or rerun an individual panel rather than rebuilding the whole sequence. Inpaint is available for a smaller area that needs correction.

Use a Studio node for dialogue and captions as text layers. This keeps the wording and font editable. A storyboard template can help explore a sequence, but check the number and purpose of the panels before treating it as your final four-panel page.

Anifusion: build a comic workspace with character profiles

Anifusion is useful for a comic project with its own panels, character profiles and text layers. Create a comic workspace, choose a suitable layout and give the characters names, appearance descriptions and references supported by the selected model.

For a vertical four-panel strip, its four-row layout is a useful starting point. Mention the profile names in the story where each character appears. Use individual panel controls when one picture needs to change; a page composed as one image has a different editing structure.

Check the layer list before editing dialogue. Real text layers can be changed directly, while words painted into the picture need image editing or replacement. References and profiles guide the artwork; inspect the faces, outfits and speaker assignments in the result.

Canva: assemble four panels and letter the comic

Canva is a straightforward choice for arranging the panels and adding the exact dialogue. Start with a four-panel template or grid, then upload the approved artwork and put one image in each frame.

Adjust the crop rather than stretching a character to fit. Search Elements for a speech bubble, add a text box and paste the exact line. Position the tail clearly toward the speaker, and use the same general font and bubble treatment across the page.

Keep the editable design alongside the export. The design lets you change wording and bubble placement; a finished PNG or PDF is the page you share. Check template elements and fonts against the use you intend rather than assuming every item in the library has the same availability.

Photoshop: keep artwork, bubbles and type on separate layers

Photoshop is useful when you want detailed corrections and an editable layered comic file. Keep the panel artwork, bubble shapes and type separate so each can be revised independently.

Use the Type tool to enter the dialogue, select the characters that need changing and commit the edit. This works on type layers; it does not turn letters painted into a generated image into editable text.

Separate layers also let you mask a small artwork correction without touching the dialogue above it. Retain the working document before exporting a flattened page. If that editable handoff matters, compare the options for AI photo editing and editable layers.

Clip Studio Paint: control panels, balloons and dialogue

Clip Studio Paint is useful when comic lettering and panel finishing need more dedicated controls. Balloon shapes, tails and text can be adjusted as parts of the comic rather than painted into the illustration.

Create a balloon, adjust its shape with the Object tool and attach a tail toward the speaker. Enter the dialogue with the Text tool, then revise the font, size or position as needed. A separate text area can also serve a caption without a speech balloon.

Use these controls to make the intended reading order obvious. A neatly drawn panel still needs enough room for the words, clear tails and consistent lettering. Save the editable project before producing the final page image.

Assemble a readable four-panel page

Choose a reading order and make the panel shapes, gutters and lettering support it. A 2×2 page works well for an English strip read across the top row, then across the bottom. A vertical stack gives each panel more room on a phone. A long horizontal strip can become very small on a narrow screen.

As a simple design example, a 1,080×1,080 page with 40-pixel outer margins and a 40-pixel central gutter leaves four 480×480 panel areas. That is a layout calculation, not a universal publishing requirement. Choose the destination size before preparing your final artwork and text.

Use consistent spacing so the reader sees four deliberate beats. Crop carefully around faces, hands, bubbles and props. If an important action is missing after the crop, revise the frame or the artwork rather than adding a caption to explain something the picture no longer shows.

Check the actual four-panel count. A generated page might contain an extra inset, a split scene or decorative boxes that complicate the sequence. Keep only the panels that serve the planned story.

Check continuity, dialogue and the exported page

Compare all four panels with the character references and master script before sharing the comic. Review the complete page as well as the individual drawings.

For the plant example, compare the blue pot, plant size and the two outfits across all four pictures. Then read only the dialogue in order. Finally, read the whole page normally: the visual joke and the words should support each other.

Reopen the exported file and check it at the size people will see. A clean-looking editor view does not reveal every problem with a smaller export or an unexpected crop.

Fix the most common consistency problems

Fix the cause of the mismatch rather than adding a longer prompt to every panel.

Keep a usable earlier version when revising. If a correction solves one issue but introduces another, return to the approved panel and make a narrower change.

Frequently asked questions

Can I generate all four panels in one request?

One prompt can produce a useful rough page, but separate panels make individual corrections easier. Give a full-page request four clear visual beats and inspect the panel count, cast and words. If the output is one flat image, its drawn panels are not automatically four independent editable objects.

Does using the same seed keep a character identical?

A seed is not a substitute for a character reference. Its role depends on the tool and model, and repeating it does not establish an identity lock across different scenes. Reuse the approved images and defining traits, then compare the results.

Can I put the dialogue directly into the generated image?

You can request drawn-in dialogue, but check every word and speaker before using the result. Quote the exact wording, specify its placement and ask for no additional text. For frequent revisions or exact final copy, separate editable lettering is easier to control.

How should I translate the comic?

Translate the master script, then update the editable text and resize the bubbles as needed. Preserve the character's voice and the joke rather than matching the original line lengths. Review the reading direction and bubble order for the new language, and avoid regenerating approved faces just to replace words.

What should I save for the next episode?

Save the approved references, style brief, script, individual panels and editable page. Include important setting and prop details. Begin the next strip with those approved character designs rather than using a progressively changed later panel as the only reference.

Start with a small, reusable cast

Start with Dreamina to develop a small cast and create the panels from the same approved references. Give the story four clear visual beats, keep the original dialogue separate and make corrections one panel at a time.

Finish with readable bubbles and an editable layout. Once the characters, style and lettering work together, the next four-panel strip can build on the same cast without redesigning the whole workflow.