Agnes AI — Image & Video Generation

Generate visual media through the Agnes AI API:

• Images — agnes-image-2.1-flash (text-to-image, image-to-image)
• Videos — agnes-video-v2.0 (text / image / multi-image / keyframe; asynchronous)

Setup

Needs an Agnes AI API key (get one at https://agnes-ai.com → API Platform). Provide it either way:

export AGNES_API_KEY="your-key"                                  # 1) environment variable
mkdir -p ~/.config/agnes-ai && printf 'AGNES_API_KEY=your-key\n' \
  > ~/.config/agnes-ai/config && chmod 600 ~/.config/agnes-ai/config  # 2) config file

Optional: AGNES_OUTPUT_DIR (where results are saved, default ~/agnes-output).

Workflow

1. Decide image vs video from the request.
2. Craft a vivid English prompt (see Prompt Crafting) — translate from any language.
3. Run the matching script below.
4. Read the output file to show the user the result.
5. Report the saved local path (for video, also mention the duration).

Commands

Image

bash ${CLAUDE_SKILL_DIR}/bin/generate-image.sh "<english_prompt>" [size] [input_image]

Prints the saved .png path to stdout on success.

Video (asynchronous)

bash ${CLAUDE_SKILL_DIR}/bin/generate-video.sh "<english_prompt>" [image] [duration_seconds]

Advanced options (environment variables): AGNES_VIDEO_FPS (24), AGNES_VIDEO_WIDTH (1152), AGNES_VIDEO_HEIGHT (768), AGNES_VIDEO_MODE (keyframes), AGNES_VIDEO_NEGATIVE, AGNES_VIDEO_SEED.

The script creates the task, polls until completion (tens of seconds to a few minutes), then downloads the .mp4. Prints the saved path on success.

Prompt Crafting

Translate the user's idea into fluent English, then enrich it:

• Image: [Subject] + [Scene/Environment] + [Style] + [Lighting] + [Composition] + [Quality] e.g. "A luminous floating city above a misty canyon at sunrise, cinematic realism, wide-angle composition, soft golden light, high visual density"
• Image-to-image: state what changes AND what to keep — "Transform into a rain-soaked cyberpunk night with neon reflections, while preserving the original composition and main subject layout"
• Video: [Subject] + [Action] + [Scene] + [Camera Movement] + [Lighting] + [Style] e.g. "A young astronaut walking across a red desert planet, dust blowing in the wind, slow cinematic tracking shot, dramatic sunset lighting, realistic sci-fi"
• Image-to-video: describe what should move and what stays stable — "Subtle breathing motion, hair moving gently in the wind, while keeping the face and outfit consistent"

Keep the user's creative intent intact — enhance, don't override.

Examples

User: "画一只在沙滩散步的猫，电影感"

bash ${CLAUDE_SKILL_DIR}/bin/generate-image.sh "A cat walking on the beach at sunset, cinematic photography, warm golden light, soft bokeh, ultra realistic"

User: "把这张照片做成 5 秒视频，让她慢慢回头"

bash ${CLAUDE_SKILL_DIR}/bin/generate-video.sh "The woman slowly turns around and looks back at the camera, natural facial expression, cinematic camera movement" "/path/to/photo.png" 5

After either command, Read the printed file path to display the result.

Notes

• Pricing (per Agnes docs): images ~$0.003 each; video ~$0.005/second.
• Timing: images take seconds; video takes tens of seconds to a few minutes.
• Frame rule: video frame count must be 8n+1 and ≤ 441 — the script snaps the requested duration to a valid value automatically.
• Known doc discrepancy: for image-to-image the input image goes under extra_body.image (per the official call examples); the prose docs also mention a top-level image array. This skill uses extra_body.image.
• Output is saved to ~/agnes-output by default (AGNES_OUTPUT_DIR to change).
• Run bash ${CLAUDE_SKILL_DIR}/test/run.sh for offline mock tests (no key needed).