Four-panel comic with your own character: an API recipe

To make a four-panel comic with your own character through an API, write the script first, render each panel as its own call with the character image attached as a reference, and add the dialogue afterwards in a layout tool. On Sume that is four calls to POST /v1/images with google/nano-banana-2.1 and one input_references entry each. Four draft panels at 0.5K cost $0.30, and four finals at 1K cost $0.40.

The recipe follows a guide that Recraft published on October 9, 2026 (read 2026-10-13). It runs a setup, development, turn and payoff script, storyboards with a fast model, renders final panels with the character as a reference, and letters the comic last. Recraft does that inside its Studio; this page shows the same loop as code against Sume's Image API.

What does Recraft's four-panel method actually say?

Recraft's guide has five steps. The guide storyboards with Recraft V4.1 Flash, also uses Nano Banana 2.1, and quotes prices in Recraft credits, which do not map to Sume prices. The parts that carry over are the order of work and two rules.

• Write a four-panel script with minimal dialogue: setup, development, turn, payoff.
• Storyboard cheaply first, so you fix the story before paying for finals.
• Attach the character as a reference image to every final panel to keep it on model.
• Add dialogue with a text tool, not the image model, because generated lettering is often misspelled.

Which Sume model and settings fit a comic panel?

Nano Banana 2.1 is listed on Sume as google/nano-banana-2.1. Per the Image API docs and the catalog on main, it accepts up to 10 input_references, n from 1 to 4, and the ratios 4:5, 3:4, 1:1 and 16:9 among others. Resolution tiers are 0.5K, 1K, 2K and 4K. Sume prices it at the provider list times 1.25.

How do I render the four panels with one character?

Send the same hero image as the only reference on every call, and write the panel action in the prompt. Run the script once with resolution set to 512 for the storyboard, read the four results, then rerun with 1K after you fix the panel text. The script raises on a 202 response, which means the call outlived the 30-second wait and became a job; read it from the status URL as described in Jobs and results.

import os, requests

URL = "https://api.sume.com/v1/images"
HEAD = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
HERO = "https://example.com/hero.png"
PANELS = [
    "Setup: the character arrives at a closed bakery at dawn.",
    "Development: they peek through the window at the baker.",
    "Turn: the door swings open and flour covers them.",
    "Payoff: they eat a warm roll on the step, happy.",
]

def panel(action, resolution):
    body = {
        "model": "google/nano-banana-2.1",
        "prompt": action + " Same character and outfit as the reference. No text or speech bubbles.",
        "input_references": [{"type": "image_url", "image_url": {"url": HERO}}],
        "aspect_ratio": "4:5",
        "resolution": resolution,
    }
    r = requests.post(URL, headers=HEAD, json=body, timeout=60)
    r.raise_for_status()
    if r.status_code == 202:
        raise RuntimeError(r.json()["data"]["status_url"])
    return r.json()["data"][0]["url"]

for i, action in enumerate(PANELS, 1):
    print(i, panel(action, "512"))

Where does the dialogue go?

Outside the image call. Recraft's advice is to add dialogue with the text tool because generated lettering is often misspelled, and the prompt above says so too by asking for no text or bubbles. Place the four panels in a layout you control, then add bubbles and captions there. If you later need the same comic in another language, the speech-bubble translation recipe covers a per-page edit, and the comic generator guide covers panel shapes.

What does a full comic cost on Sume?

One storyboard pass at 0.5K is 4 x $0.075 = $0.30. One final pass at 1K is 4 x $0.10 = $0.40. A single redo of one panel at 1K adds $0.10. A failed generation uses no credits, per the Image API billing notes. Keep the hero image under your control: if the character drifts, add a second reference such as a turnaround sheet, as in consistent character in three steps, up to the 10-reference cap.