agnes-ai-generation

这段内容是审核时保存的快照。外部来源才是完整且最新的版本。

---
name: agnes-ai-generation
description: Call Agnes AI / Sapiens AI generation APIs for text, image, and video. Use when the user asks to use Agnes models, Agnes Image, Agnes Video, Agnes 2.0 Flash, apihub.agnes-ai.com, or to generate text, images, edit images, create videos, animate images, create keyframe videos, or test Agnes API calls.
---

# Agnes AI Generation

Use this skill to call Agnes text, image, and video generation APIs through `https://apihub.agnes-ai.com`.

## Quick Start

1. Read `references/api.md` when endpoint details, parameters, or response fields are needed.
2. Use `scripts/agnes_api.py` for real API calls instead of rewriting curl by hand.
3. Require an API key in `AGNES_API_KEY`, `AGNES_API_TOKEN`, or `APIHUB_AGNES_API_KEY`. Never print the key.
4. For light live verification, run `smoke-test`; it avoids video creation by default. Add `--include-image-edit` for image-to-image, and add `--video-case <case>` explicitly for video modes. Treat the skill as fully tested only when basic text, text streaming, text tool calling, text-to-image, image-to-image, text-to-video, image-to-video, multi-image video, keyframe video, and video retrieval return successful responses.

## Commands

Text generation:

```bash
python scripts/agnes_api.py text --prompt "Write a concise product tagline for an AI assistant."
```

Streaming text:

```bash
python scripts/agnes_api.py text --prompt "Write a short product intro." --stream
```

Streaming output is normalized and includes aggregated `content`, `events`, `done`, and a short `raw_prefix`.

Image generation:

```bash
python scripts/agnes_api.py image --prompt "A luminous floating city above a misty canyon at sunrise, cinematic realism" --size 1024x768
```

Image-to-image:

```bash
python scripts/agnes_api.py image --prompt "Turn the scene into a rainy cyberpunk night while preserving composition" --image https://example.com/input.png --size 1024x768
```

Text-to-video with polling:

```bash
python scripts/agnes_api.py video --prompt "A cinematic shot of a cat walking on the beach at sunset" --poll
```

Image-to-video:

```bash
python scripts/agnes_api.py video --prompt "Animate subtle camera movement and natural lighting" --image https://example.com/image.png --poll
```

Keyframe / multi-image video:

```bash
python scripts/agnes_api.py video --prompt "Create a smooth cinematic transition between the two keyframes" --image https://example.com/a.png --image https://example.com/b.png --mode keyframes --poll
```

Retrieve a video task:

```bash
python scripts/agnes_api.py video-get video_123456
```

Light live smoke test:

```bash
python scripts/agnes_api.py smoke-test
```

Image edit smoke test:

```bash
python scripts/agnes_api.py smoke-test --include-image-edit
```

Single video smoke test:

```bash
python scripts/agnes_api.py smoke-test --video-case text-to-video
```

## Workflow

- Prefer `agnes-2.0-flash` for text chat/completions.
- Do not use Agnes Responses API multi-turn function calling for autonomous tool workflows. Live testing showed the provider can return `function_call` with overall `status=completed`, and submitting `function_call_output` with `previous_response_id` may fail. Use this skill's chat completions path for text generation and treat tool-calling as best-effort request-shape compatibility only.
- Prefer `agnes-image-2.1-flash` for text-to-image, image-to-image, and high-information-density image generation. High-density generation is prompt-driven; include subject hierarchy, environment, secondary details, lighting, composition, and quality requirements.
- Prefer `agnes-video-v2.0` for text-to-video, image-to-video, multi-image video, keyframe animation, prompt-based motion and scene control, cinematic output, asynchronous task creation, polling-based result retrieval, and seed-based reproducibility.
- For image and video generation, convert any non-English user prompt to a fluent English generation prompt before calling the image/video API. En