Gemini Omni 视频编辑——原生多模态的视频到视频编辑。只需一个Prompt,即可从源片段重新设计风格、调整光照、替换主体或改写场景。跨模态统一推理在应用编辑的同时保持运动和音频连续性。
About this model
Gemini Omni 视频编辑将 Google 原生多模态 any-to-any 模型用于基于源片段的编辑。提供一个片段和自然语言编辑指令——重新设计外观、改变季节、替换主体或改写对白——模型就会在保留原始运动和时间安排的同时,通过一次处理重写视频。
由于模型会联合理解画面和音频,编辑可以在不同模态之间保持连贯:重新生成的音频会匹配新画面,需要时也可以保留原始环境声。
Cost analysis
| Provider | Cost | Notes |
|---|---|---|
| muapi | $2.40 (720p/1080p) · $3.60 (4K) | Flat rate per generation — same price regardless of duration. Synchronized audio included at no extra charge. |
| Fal.ai | Non disponibile | Gemini Omni Video Edit is not currently available on Fal.ai. |
| Replicate | Non disponibile | Gemini Omni Video Edit is not currently available on Replicate. |
Flat rate per generation — same price regardless of duration. Synchronized audio included at no extra charge.
Gemini Omni Video Edit is not currently available on Fal.ai.
Gemini Omni Video Edit is not currently available on Replicate.
** Competitor pricing is estimated based on similar model architectures and usage tiers.
Configuration schema
| Parameter | Type | Description | Default |
|---|---|---|---|
| 编辑Prompt | string | 描述要应用的编辑操作。例如:改为水彩风格、更换季节、替换主体服装。 | Transform the growing clock into @image1 while preserving the original expansion and wall-breaking motion. As the clock enlarges, metallic spider legs unfold from the sides and stab through furniture and walls. Keep the camera motion and apartment destruction intact. |
| 参考Immagine | array | 参考Immagine。总共提供 7 个Immagine槽位;Video占用 2 个槽位,每个 character_id 占用 1 个槽位。每张Immagine最大 20 MB。 | https://cdn.muapi.ai/assets/gemini-omni-video-edit.jpg |
| 源Video | string | Source video to edit (max 100 MB, max 30 s). Optional if image_urls are provided. | https://cdn.muapi.ai/assets/gemini-omni-text-to-video.mp4 |
| 裁剪开始时间(秒) | number | Start of the clip window to edit, in seconds (>= 0). | 0 |
| 裁剪结束时间(秒) | number | End of the clip window to edit, in seconds. Must be within 10 s of trim_start. | 10 |
| Durata (secondi)(秒) | Enum (4 options) | 生成Video的Durata(秒)。 | 8 |
| Risoluzione | Enum (3 options) | OutputVideoRisoluzione。720p 和 1080p 价格相同;4K 价格更高。 | 1080p |
| Proporzioni | Enum (2 options) | OutputVideo的Proporzioni。 | 16:9 |
| Audio ID | array | Up to 3 voice profile IDs returned by the Gemini Omni Audio endpoint. | - |
| 种子 | int | Seed casuale(0–2147483647)。固定后可复现结果,但由于模型随机性,结果仍可能有所不同。 | 0 |
| 角色 ID | array | Up to 3 character IDs from Gemini Omni Character to feature in the video. | - |
描述要应用的编辑操作。例如:改为水彩风格、更换季节、替换主体服装。
Transform the growing clock into @image1 while preserving the original expansion and wall-breaking motion. As the clock enlarges, metallic spider legs unfold from the sides and stab through furniture and walls. Keep the camera motion and apartment destruction intact.参考Immagine。总共提供 7 个Immagine槽位;Video占用 2 个槽位,每个 character_id 占用 1 个槽位。每张Immagine最大 20 MB。
https://cdn.muapi.ai/assets/gemini-omni-video-edit.jpgSource video to edit (max 100 MB, max 30 s). Optional if image_urls are provided.
https://cdn.muapi.ai/assets/gemini-omni-text-to-video.mp4Start of the clip window to edit, in seconds (>= 0).
0End of the clip window to edit, in seconds. Must be within 10 s of trim_start.
10生成Video的Durata(秒)。
8OutputVideoRisoluzione。720p 和 1080p 价格相同;4K 价格更高。
1080pOutputVideo的Proporzioni。
16:9Up to 3 voice profile IDs returned by the Gemini Omni Audio endpoint.
-Seed casuale(0–2147483647)。固定后可复现结果,但由于模型随机性,结果仍可能有所不同。
0Up to 3 character IDs from Gemini Omni Character to feature in the video.
-Developer documentation
上传源视频
提供公开的 video_url,或从 playground 上传片段。较短的Input(约少于 15 秒)能产生最稳定的编辑结果。
编写清晰的编辑Prompt 明确说明哪些内容应该改变、哪些内容应该保留。例如:'Restyle the entire clip as a hand-drawn Studio Ghibli animation, keep the original camera motion, lighting direction, and timing.'
选择分辨率
720p / 1080p — 价格相同(每次生成 $2.40)4K — 更高分辨率(每次生成 $3.60)提交并轮询
POST 到 /api/v1/gemini-omni-video-edit,并轮询 GET /api/v1/predictions/{request_id}/result,直到 status 为 completed。
Prompt技巧
Frequently asked
Gemini Omni 视频编辑会直接以源片段为条件,同时理解画面和音频。应用编辑时,模型会保留源片段的运动、时间安排和连续性,而不是生成一个大致匹配的新片段。
可以——禁用 `preserve_audio`,并编写只针对音频的Prompt(例如,"replace the dialogue with: '...'")。画面会与源片段保持对齐,同时重新生成音频。
最大片段长度将在上线时确认。预计支持最长 30 秒的Input,与Testo a Video版本一致。
默认情况下,模型会保留源片段的运动和镜头路径。如果希望改变镜头或运动,请在Prompt中明确说明。
Google 于 2026 年 5 月 19 日的 I/O 2026 上宣布了 Gemini Omni,API 访问将在接下来几周陆续开放。视频编辑版本将在上游 API 支持后立即在 muapi 上线。