YouTube Create: Nano Banana from 3 images, then animate with Omni

In YouTube Create you can generate an image with Nano Banana from a prompt and up to 3 reference images, then animate a clip with Gemini Omni. On Sume it is two jobs: an image generation call with input_references, then a Gemini Omni Flash 1.1 call that takes the finished image as image_url.

YouTube's side comes from its help page Create with AI in the YouTube Create app, read on 2026-10-02. Sume's side is the Image generation and Video Router docs.

What does YouTube Create offer for images?

The help page lists Nano Banana as an image tool that works from text prompts and reference images, up to 3. It is limited to users 18 and over. Nano Banana and Gemini Omni share the same availability: mobile only, in Australia, Canada, India, New Zealand and the United States. The page calls the features experimental and says they are available to a limited number of creators.

Nothing in the page says the image and the clip are linked automatically. The two features sit side by side in the editor, and you choose how to chain them.

How do I do step one on Sume?

Image generation takes a prompt and optional input_references. Each model advertises how many references it accepts in its descriptor, and a model whose range is 0 to 0 is text-only and rejects references. Reference URLs must be public HTTPS; localhost, private network and non-HTTPS URLs are rejected before the job is submitted.

So read the model list first, pick a model whose range allows your three references, and keep to the public-HTTPS rule. If your photos live on a laptop, import them first; the Media inputs page describes the flow.

curl https://api.sume.com/v1/images/models \
  -H "Authorization: Bearer $SUME_API_KEY"

How do I animate the image on Sume?

Take the finished image's URL from the job result and send it as image_url to the Video Router. An optional end_image_url makes it a start and end frame request. The envelope is the same as text to video: 3 to 10 seconds, 360p to 4K, 16:9 or 9:16, native audio always on.

Use 9:16 for a Short and keep the motion in the prompt short. Describe what moves, not what the picture already shows.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: animate-still-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "The mascot waves and the camera pushes in slowly. Soft music.",
    "image_url": "https://media.sume.com/artifacts/artf_demo/mascot.png",
    "duration": 6,
    "aspect_ratio": "9:16",
    "mode": "async"
  }'

What are the differences to plan for?

Use YouTube Create for a single quick clip on your phone in a supported country. Use the two-call chain when you want the same image reused across several clips, a record of each job, or a country that the app does not cover.

• YouTube Create keeps the work in one mobile editor. Sume gives you two job ids and you hold the URL between them.
• Each step bills on its own. Check the price for each model in the catalog before a batch.
• Poll the job status for each step and do not resubmit a paid request after a client timeout.
• There is no YouTube label here, and no publishing. Those are separate steps on your side.

What goes wrong when you chain two models?

The most common loss is identity. The still looks right, then the animated clip drifts: a logo changes, a hand has a different shape. Two habits help. Ask for small motion, a wave, a push in, a head turn, and avoid asking the model to invent large new movement. And keep the still's aspect ratio equal to the clip's: a 9:16 still for a 9:16 clip avoids a crop or a pad.

The second loss is cost you cannot see. Each job has its own reserve and its own price, and a failed job ends with a public error you can read from the status route. Use the unbilled model listing to check the numbers, and run the video step at 360p first so a bad still costs little.

The third is rights. Reference images that show a real person or a brand belong to someone. YouTube asks for age 18 and over for Nano Banana, and Sume's docs only require that image URLs be public HTTPS. Putting consent checks in your own flow is your job in both cases.