Awesome-Gemini-Omni-API-Prompts — Claude MCP 必备工具中文教程

Installation

pip install gemini-omni-api

Or clone this repo and install directly:

git clone https://github.com/Anil-matcha/Awesome-Gemini-Omni-API-Prompts.git
cd Awesome-Gemini-Omni-API-Prompts
pip install -r requirements.txt
cp .env.example .env   # add your MUAPI_API_KEY

Python Wrapper — Quick Start

```python from gemini_omni_api import GeminiOmniAPI

api = GeminiOmniAPI(api_key="your-muapi-key") # or set MUAPI_API_KEY env var

Hero Walking Cycle Across Environments

Reference: character reference (full body). Prompt:

Using the attached character as the subject, generate a continuous walking sequence where the environment transitions seamlessly every 3 seconds: 0–3s desert dunes at sunset, 3–6s neon-soaked rainy city street, 6–9s misty alpine forest, 9–12s clean white infinity studio. Character stride, outfit, and proportions remain identical throughout. Camera tracks alongside in profile, locked focal length, 12 seconds, 16:9.

---

Environment Swap — Interior to Rooftop

Prompt:

Keep the subject and all their motion identical but replace the background environment with a sunlit urban rooftop at golden hour. City skyline visible in the background, warm directional rim light matching the new environment falling on the subject, soft atmospheric haze. Seamless integration — no compositing artifacts.

---

Three-Scene Mini-Film From One Reference

Reference: single portrait of the protagonist. Prompt:

The attached image is the main character. Generate three connected shots, same person in all three, identical face, hair, and wardrobe throughout:
1) 0–4s: Wide shot, she walks out of a Parisian apartment building into morning light, holding a paper coffee cup, slow dolly-back
2) 4–8s: Medium shot, she stops at a flower stall, smiles at the vendor, picks up a bouquet, soft sidelight
3) 8–12s: Close-up, she sits on a bench by the Seine, looks off-camera contemplatively, gentle breeze in hair
Cinematic 35mm look throughout, consistent color grade, no cuts to other characters.

---

Two-Character Sword Duel — Reference Both

References: two character images. Prompt:

Image 1 and image 2 are the two combatants. Generate a 10-second cinematic sword duel under a stormy bamboo forest at night. Choreography: clash 1 sparks fly, parry, character 1 advances with three quick strikes, character 2 dodges and counters with a sweeping cut, locked-blade standoff with both faces close to camera, slow dolly-around. Rain falls throughout, lightning flashes punctuate the impacts. Maintain both characters' exact faces and costumes. 10 seconds, 21:9 cinemascope.

---

Style Transfer From Reference Image

Reference: a stylized art piece. Prompt:

Re-render the previous clip in the visual style of the attached image — match its color palette, brush texture, and lighting language. Preserve the original motion, framing, and subject identity exactly.

---

Resources & API Docs

Use Google Gemini Omni via MuAPI

MuAPI provides all three Gemini Omni generation modes — text-to-video, image-to-video, and video-edit — as a hosted REST service. No extra account required beyond an API key.

All modes return {"request_id": "..."}. Poll GET /predictions/{request_id}/result until status == "completed".

REST API Reference

All endpoints follow the same submit → poll pattern:

POST https://api.muapi.ai/api/v1/{endpoint}   →  {"request_id": "abc123", "status": "processing"}
GET  https://api.muapi.ai/api/v1/predictions/{request_id}/result  →  poll until status == "completed"

Text-to-Video

import requests, time

API_KEY = "your-muapi-key"
BASE = "https://api.muapi.ai/api/v1"

resp = requests.post(
    f"{BASE}/gemini-omni-text-to-video",
    headers={"x-api-key": API_KEY},
    json={
        "prompt": "35mm anamorphic rain-soaked Tokyo alley at 2 AM, slow dolly-in",
        "duration": 8,              # 4 | 6 | 8 | 10
        "aspect_ratio": "16:9",     # "16:9" | "9:16"
        "resolution": "1080p",      # "720p" | "1080p" | "4k"
        # "audio_ids": ["aoede"],   # optional — list of up to 3 voices
        # "character_ids": [...],   # optional character IDs
        # "seed": 42,               # optional seed
    },
)
request_id = resp.json()["request_id"]

while True:
    r = requests.get(f"{BASE}/predictions/{request_id}/result", headers={"x-api-key": API_KEY}).json()
    if r["status"] == "completed":
        print(r["outputs"])
        break
    time.sleep(5)

Image-to-Video

```python

Upload reference image(s) first

with open("reference.png", "rb") as f: image_url = requests.post( f"{BASE}/upload_file", headers={"x-api-key": API_KEY}, files={"file": f} ).json()["url"]

resp = requests.post( f"{BASE}/gemini-omni-image-to-video", headers={"x-api-key": API_KEY}, json={ "prompt": "Animate the subject with subtle breathing and gentle hair movement", "image_urls": [image_url], # list of 1–5 URLs "duration": 6, "aspect_ratio": "9:16", "resolution": "1080p", # "720p" | "1080p" | "4k" # "audio_ids": ["aoede"], # optional — list of up to 3 voices # "character_ids": [...], # optional character IDs }, )

#### Video Edit

resp = requests.post( f"{BASE}/gemini-omni-video-edit", headers={"x-api-key": API_KEY}, json={ "prompt": "Change the season to winter, add falling snow", "video_url": video_url, "trim_start": 0, "trim_end": 8, # max 10 s window; source video max 30 s "duration": 8, "aspect_ratio": "16:9", "resolution": "1080p", # "720p" | "1080p" | "4k" # "image_urls": [...], # optional reference images (max 5) # "audio_ids": ["aoede"], # optional — list of up to 3 voices # "character_ids": [...], # optional character IDs }, )

#### Audio Profile

API Parameters

Image slot budget (V2V): when video_url is set, max 5 image_urls.

Available audio_ids voices: achernar · achird · algenib · algieba · alnilam · aoede · autonoe · callirrhoe · charon · despina · enceladus · erinome · fenrir · gacrux · iapetus · kore · laomedeia · leda · orus · puck · pulcherrima · rasalgethi · sadachbia · sadaltager · schedar · sulafat · umbriel · vindemiatrix · zephyr · zubenelgenubi