Add yourself to an AI video with a reference photo (Gemini Omni)

Give Gemini Omni Flash a clear photo of the person as a reference image, address it in the prompt, and describe the scene. On Sume that is gemini-omni-flash-1.1 with reference_image_urls and an <IMAGE_REF_0> tag in the prompt. YouTube offers a consumer version of the idea in Shorts Remix, where you can insert yourself alongside creators in an eligible Short.

These are two different jobs. Shorts Remix edits someone else's Short inside YouTube, with YouTube's controls. The Sume route generates a new clip that features the person in the photo, or edits a video file you provide, and gives you the result as a file and an API job. Facts about YouTube come from its Google I/O 2026 blog post and Google's I/O roundup, read 2026-10-02.

What does Shorts Remix let me do with my own face?

The YouTube blog says you can remix an eligible Short with your own prompts and images, for example inserting yourself alongside creators. Google's roundup also says the Gemini app can make custom AI avatars that match your appearance and voice. Both are free or subscription features inside Google's apps, with no API described on those pages.

The same blog says creators can opt out of visual remix, and remixed videos carry watermarks, identifying metadata and a link back to the original. A Short that is not eligible cannot be used this way.

How do I do it through Sume?

Use reference-to-video. The Video Router docs say gemini-omni-flash-1.1 takes up to 10 reference_image_urls and up to 3 reference_video_urls (each 3 seconds or shorter), and that you address media as <IMAGE_REF_0>, <VIDEO_REF_0> with 0-based list order. Output is 3-10 seconds at 360p to 4K, 16:9 or 9:16, with native synced audio always on.

Put the face photo first so it is <IMAGE_REF_0>. Say what the person does, where they are, and what should not change.

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: me-in-scene-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "<IMAGE_REF_0> walks into a busy night market, looks at the camera and laughs. Handheld, warm lights.",
    "reference_image_urls": ["https://example.com/me-front.jpg"],
    "aspect_ratio": "9:16",
    "duration": 6,
    "resolution": "720p",
    "mode": "async"
  }'

What can go wrong?

There is no audio reference: Sume does not accept reference_audio_urls for this model, so the clip's voice is generated, not your own. If you need your voice, use a different route such as Sume Avatar 1.0.

A reference photo is a guide, not a guarantee of an exact match. Use a sharp, front-lit photo, one face per image, and generate at 360p first to check likeness before paying for 1080p. If the face drifts, add a second photo from another angle as <IMAGE_REF_1>.

Only use photos of people who agreed. Sume does not verify consent for you, and the YouTube blog's likeness-protection language applies to YouTube, not to files you generate elsewhere.

Can I insert myself into an existing clip instead?

Not in one Sume call. Edit mode (video_url) cannot be combined with reference_*_urls, so you cannot send a source clip plus a face photo together. Two workable paths are to generate the clip with the photo as a reference and recreate the scene, or to edit the existing clip with a text prompt only. The first gives you the person; the second keeps the original footage.

Poll the job as in Jobs and results, and keep the Sume media URL from the result rather than any provider link.