Nano Banana character consistency: reference limits and a Sume request

To keep one character across Nano Banana images on Sume, send the same clear reference image of the character in input_references on every request and describe the new scene in the prompt. Both google/nano-banana-2 and google/nano-banana-pro take up to 10 references. Google's docs give its own API separate character slots, which Sume's docs do not describe.

Google's limits are from its image generation docs and Sume's from the Image API docs, read 2026-09-29.

What does Google say about character consistency?

How do I keep a character on Sume?

Reuse the same reference and change only the scene. Set aspect_ratio: "auto" when you want the output shape to follow the reference. Reference URLs must be public HTTPS.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/nano-banana-pro",
    "prompt": "The same woman from the reference, now reading in a sunlit library",
    "input_references": [
      {"type": "image_url", "image_url": {"url": "https://example.com/hero-character.png"}}
    ],
    "aspect_ratio": "4:5"
  }'

What still drifts?

Fine details such as jewelry, logos on clothing and hair edges can shift between generations. Sume's docs make no consistency guarantee. Generate n candidates, compare each with the reference, and keep the one that matches.

Can the same character move into video?

Yes, by passing a chosen still as a first frame or reference to the video endpoint; see consistent character across AI video shots.