Anime-style video prompt for Gemini Omni: cel look and limits

An anime look in Gemini Omni comes from art-direction words (flat cel shading, clean line art, painted backgrounds) plus motion words (limited animation, hair and cloth sway). On Sume a 5-second take is $0.63 at 720p. For a recurring character, send a drawn character sheet in input_references and call it <IMAGE_REF_0>; Google's Omni page lists reference-to-video as one of five task types (read 2026-10-07).

Style names for living studios or artists are a risk, both for quality and for rights; describe the technique instead.

Describe the technique, not a studio

Words that describe the visual method are more repeatable than a studio name, and safer to publish.

• Rendering: 'flat cel shading, clean ink outlines, painted watercolor background'.
• Motion: 'limited animation, hair and scarf sway on every other frame, parallax clouds'.
• Camera: 'slow lateral pan across a rooftop at dusk'.
• Color: 'warm orange sky, teal shadows'.
• Sound: 'wind, distant train, soft piano. No dialogue.'

With a character sheet

A single reference image in input_references is reference-to-video on Sume: it gives the model a subject, not a fixed first frame. The Video Router doc says Omni takes up to 10 images and addresses them as <IMAGE_REF_0>.

{
  "model": "gemini-omni-flash-1.1",
  "prompt": "Anime style, flat cel shading, clean ink outlines, painted watercolor background. <IMAGE_REF_0> stands on a rooftop at dusk, scarf swaying, slow lateral pan, limited animation. Audio: wind, distant train, soft piano. No dialogue.",
  "duration": 5,
  "resolution": "720p",
  "aspect_ratio": "16:9",
  "input_references": [
    { "type": "image_url", "image_url": { "url": "https://example.com/character-sheet.png" } }
  ]
}

Cost

Sume bills list times 1.25 per second and rounds each job up to the cent. Image references do not add a per-image fee in the doc's Omni row.

Where it can fail

Faces drawn small can lose detail, hands can merge, and a flat style may drift toward 3D shading mid-clip. If that happens, move the camera closer, cut the action to one gesture, or shorten to 3 seconds. Keep the character sheet clean: one character, plain background, front view. For a longer sequence, chain separate 5-second clips on one reference rather than asking for 10 seconds of action, and review each against the sheet.

Rights and consistency

Anime is a broad style and a fast-moving rights area. Describe the technique and make your own character; do not ask for a named existing character or a named studio. For a series, keep a short style block of 20 to 30 words and paste it unchanged at the start of every prompt, so each clip carries the same wording. Pair that block with the same character-sheet reference on every request. If one clip drifts, regenerate only that clip, since each Sume job is independent and bills on its own.