在 Hermes Agent 中使用 Veo 3。
Veo 3 是 Google DeepMind 的视频生成模型，可生成最长 8 秒、1080p 的片段，并带有原生音频——与视频同步生成的对白、环境声和音乐。Hermes Agent 通过 RunAPI 自定义提供方端点调用它，复用为聊天配置的同一个 API 密钥。
通过 RunAPI 使用 Google Veo 3 生成带原生音频的电影级视频片段。
要求：
- 从 RUNAPI_API_KEY 读取 API 密钥。请勿硬编码密钥。
- 使用 custom:runapi 提供方，base_url 为 https://runapi.ai/v1。
- 向 https://runapi.ai/api/v1/veo_3_1/text_to_video 发送 POST 请求
- 将 model 设置为 "veo-3.1"。
- 撰写包含场景、镜头运动和音频提示的描述性提示词。
- 将 duration_seconds 设置为 4、6 或 8。
- 将 aspect_ratio 设置为 "16:9"、"9:16" 或 "auto"。
- 任务是异步的。轮询返回的 task_id 直到 status 为 "completed"。
- 完成后，从响应 output 中读取视频 URL。
curl -X POST https://runapi.ai/api/v1/veo_3_1/text_to_video \
-H "Authorization: Bearer $RUNAPI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "veo-3.1",
"prompt": "A woman walks through a bustling Tokyo street at night, neon signs reflecting on wet pavement. She speaks into her phone: I just arrived. Ambient city noise, distant car horns, light rain.",
"duration_seconds": 8,
"aspect_ratio": "16:9"
}'
{
"task_id": "tsk_abc123",
"status": "pending",
"model": "veo-3.1"
}
三步在 Hermes Agent 中使用 Veo 3
Configure RunAPI
Set RUNAPI_API_KEY in the environment where Hermes Agent runs. If you already added RunAPI as a custom:runapi provider for chat, the same key and base_url handle video generation — no separate Google Cloud credentials needed.
export RUNAPI_API_KEY=runapi_xxx
Call Veo 3 text_to_video
Send a POST to the text_to_video endpoint with model set to veo-3.1. Include a prompt with scene descriptions and audio cues (dialogue, ambient sounds). Set duration_seconds to 4, 6, or 8 and aspect_ratio to 16:9 or 9:16. Hermes Agent routes through the custom:runapi provider.
POST /api/v1/veo_3_1/text_to_video
Poll for the result
The endpoint returns a task_id immediately. Poll the task status endpoint until the status changes to completed, then retrieve the output video URL. The video includes synchronized audio generated from your prompt.
GET /api/v1/veo_3_1/text_to_video/tsk_abc123
Veo 3 text_to_video 参数
| 参数 | 类型 | 说明 |
|---|---|---|
model |
string |
Required. veo-3.1 (standard quality) or veo-3.1-fast (lower cost, faster generation). |
prompt |
string |
Required. Scene description including visual action, camera movement, and audio cues for dialogue or ambient sound. |
duration_seconds |
integer |
Optional. Video length in seconds. Accepted values: 4, 6, 8. Defaults to 8. |
aspect_ratio |
string |
Optional. Output aspect ratio. Accepted values: 16:9, 9:16, auto. |
input_mode |
string |
Optional. Generation mode. text (default), first_and_last_frames (keyframe-guided), or reference (style reference). |
first_frame_image_url |
string |
Optional. URL of the first frame image. Used when input_mode is first_and_last_frames. |
last_frame_image_url |
string |
Optional. URL of the last frame image. Used when input_mode is first_and_last_frames. |
reference_image_urls |
array |
Optional. URLs of style reference images. Used when input_mode is reference. |
callback_url |
string |
Optional. Webhook URL that receives a POST when the task completes. |
Hermes Agent 上的 Veo 3 是什么？
Google DeepMind 的 Veo 3 能创作充满生命感的视频——在单次生成中同时输出电影级画面和同步对白、环境音效及音乐。通过 Hermes Agent 的 custom:runapi provider，可获得最长 8 秒的 1080p 带原生音频视频，无需单独的 TTS 或音频剪辑步骤。
Veo 3 使用场景
动漫与风格化序列
生成带匹配音效的动漫风格或风格化视频片段。在提示词中描述视觉风格和声音方向，Veo 3 同时渲染两者。
音乐与演唱内容
创作短片音乐视频或演唱二重奏场景。在提示词中包含歌词或音乐方向，输出同步的视听内容。
带音效的社交媒体内容
生成内置环境音频、旁白或对白的成品视频，无需音频剪辑——输出同时包含视频和音轨，可直接发布。
Veo 3 + Hermes Agent 常见问题
Yes. Hermes Agent calls the RunAPI Veo 3 text_to_video endpoint through the custom:runapi provider. Set model to veo-3.1 and send the request with the same RUNAPI_API_KEY you use for chat. No Google Cloud account or separate video API credentials required.
Veo 3 adds native audio generation that Veo 2 lacks entirely. If you need synchronized dialogue, sound effects, or ambient audio with your video, Veo 3 is the only option. For silent video where visual quality is all that matters, Veo 2 can be more cost-effective.
Both produce video with native audio at up to 1080p. veo-3.1 uses the full quality pipeline and costs more per generation. veo-3.1-fast trades some visual fidelity for lower cost and faster turnaround -- suitable for drafts and iteration.
Yes. Describe camera motion in the prompt (dolly in, tracking shot, aerial pan) and Veo 3 interprets it with physics-aware simulation. For audio, include dialogue in quotes and describe ambient sounds -- the model generates synchronized audio natively, no separate TTS or sound library needed.
Hermes Agent passes your prompt (including dialogue or sound descriptions) to the RunAPI Veo 3 endpoint. The model generates synchronized audio and video together. No separate audio pipeline is needed -- the output includes both tracks.
Yes. Hermes Agent can generate a reference image with Imagen 4 or Flux 2, use it as input for Veo 3 image-to-video generation, then add background music with Suno -- all orchestrated in one agent run.
立即在 Hermes Agent 中试用 Veo 3。
免费获取 RunAPI 密钥，配置 custom:runapi 提供方，即可使用 Google Veo 3 生成带原生音频的电影级视频。