Media Gen 🎬#

Generate images and videos with a single AIsa API key.

This skill covers the AIsa media-generation routes exposed across three image endpoints and one async video endpoint. The bundled client in scripts/media_gen_client.py picks the correct request shape for each supported model, including the schema differences between Wan video variants.

Use when#

• You want one neutral skill for AIsa image and video generation
• You need to switch between Gemini image, Wan image, Seedream, and Wan video models without rewriting requests
• You want a simple CLI for creating images, submitting async video jobs, polling task status, and downloading finished video output

Compatibility#

Works with any agentskills.io-compatible harness, including:

• Claude Code and Claude
• OpenAI Codex
• Cursor
• Gemini CLI
• OpenCode, Goose, OpenClaw, Hermes
• and other tools that implement the Agent Skills specification

Requires Python 3, a POSIX shell, and AISA_API_KEY from aisa.one.

What you can do#

Image — Gemini (base64 inline)#

text

"Generate a cyberpunk-style city nightscape, neon lights, rainy night, cinematic feel"

Image — Wan 2.7 (URL in chat response)#

text

"Generate an ultra-detailed product shot of a red panda, studio lighting, sharp focus"

Image — Seedream (OpenAI-compatible, large format)#

text

"Generate a 2048×2048 magazine cover: neo-noir detective portrait, film grain"

Video — text-to-video (Wan t2v)#

text

"Sweeping establishing shot of a neon cyberpunk skyline at dusk, 5 seconds"

Video — image-to-video (Wan i2v)#

text

"Starting from this reference image, gentle camera push-in with parallax"

Supported models#

Image generation — 4 models, 3 endpoints#

ModelDeveloperEndpointNotesgemini-3-pro-image-previewGooglePOST /v1/models/{model}:generateContentImages returned as base64 in candidates[].parts[].inline_data``wan2.7-imageAlibabaPOST /v1/chat/completionsImages returned as URL parts in choices[].message.content[] ( type=image)wan2.7-image-proAlibabaPOST /v1/chat/completionsHigher fidelityseedream-4-5-251128ByteDancePOST /v1/images/generationsOpenAI-compatible; minimum 3,686,400 pixels

Video generation — 4 Wan variants, 1 endpoint#

ModelKindImage fieldOutput SRwan2.6-t2vtext-to-video_none_1080wan2.6-i2vimage-to-videoinput.img_url (string)720wan2.7-t2vtext-to-video_none_720wan2.7-i2vimage-to-video**input.media** (array)720

Important: wan2.7-i2v expects the reference image in input.media as an array of URLs, not input.img_url like wan2.6-i2v. The bundled client handles this automatically when you pass --img-url.

Quick start#

bash

export AISA_API_KEY="your-key"

# Any image model — the client routes to the right endpoint
python3 scripts/media_gen_client.py image \
  --model gemini-3-pro-image-preview \
  --prompt "A cute red panda, cinematic lighting" \
  --out out.png

python3 scripts/media_gen_client.py image \
  --model wan2.7-image-pro \
  --prompt "Ultra-detailed product shot of a red panda" \
  --out out.png

python3 scripts/media_gen_client.py image \
  --model seedream-4-5-251128 \
  --prompt "Neo-noir detective portrait, film grain" \
  --size 2048x2048 \
  --out out.png

# Video — text-to-video
python3 scripts/media_gen_client.py video-create \
  --model wan2.7-t2v \
  --prompt "Sweeping shot of a neon cyberpunk skyline"

# Video — image-to-video on wan2.7-i2v
python3 scripts/media_gen_client.py video-create \
  --model wan2.7-i2v \
  --prompt "gentle zoom with parallax" \
  --img-url "https://example.com/reference.jpg" \
  --duration 5

# Wait and download
python3 scripts/media_gen_client.py video-wait \
  --task-id <task_id> --download --out out.mp4

---

Image generation — endpoint reference#

Gemini family → POST /v1/models/{model}:generateContent#

Documentation: Google Gemini Chat.

bash

curl -X POST "https://api.aisa.one/v1/models/gemini-3-pro-image-preview:generateContent" \
  -H "Authorization: Bearer $AISA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents":[
      {"role":"user","parts":[{"text":"A cute red panda, cinematic lighting"}]}
    ]
  }'

Response contains candidates[].parts[].inline_data with {mime_type, data}, where data is a base64 PNG.

Wan 2.7 family → POST /v1/chat/completions#

Documentation: Image Generation via Chat.

Critical rule: messages[].content must be an array of typed parts. A plain string returns HTTP 400 invalid_parameter_error.

bash

curl -X POST "https://api.aisa.one/v1/chat/completions" \
  -H "Authorization: Bearer $AISA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan2.7-image",
    "messages": [
      {"role":"user","content":[
        {"type":"text","text":"A cute red panda, ultra-detailed, cinematic lighting"}
      ]}
    ],
    "n": 1
  }'

Images come back as {type: "image", image: "<url>"} parts inside choices[].message.content[].

Seedream → POST /v1/images/generations#

Documentation: OpenAI-Compatible Image Generations.

bash

curl -X POST "https://api.aisa.one/v1/images/generations" \
  -H "Authorization: Bearer $AISA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedream-4-5-251128",
    "prompt": "A cute red panda, ultra-detailed, cinematic lighting",
    "n": 1,
    "size": "2048x2048"
  }'

Response: data[].url or data[].b64_json. Upstream enforces a minimum of 3,686,400 pixels. 1024×1024 and 1536×1536 are rejected. Any aspect ratio works as long as width × height ≥ 3,686,400.

---

Video generation — endpoint reference#

Create task → POST /apis/v1/services/aigc/video-generation/video-synthesis#

Documentation: Create video generation task. Header X-DashScope-Async: enable is required.

bash

# wan2.6-t2v — text-to-video
curl -X POST "https://api.aisa.one/apis/v1/services/aigc/video-generation/video-synthesis" \
  -H "Authorization: Bearer $AISA_API_KEY" \
  -H "Content-Type: application/json" \
  -H "X-DashScope-Async: enable" \
  -d '{
    "model":"wan2.6-t2v",
    "input":{"prompt":"cinematic close-up, slow push-in"},
    "parameters":{"resolution":"720P","duration":5}
  }'

# wan2.7-i2v — image-to-video (input.media, not input.img_url)
curl -X POST "https://api.aisa.one/apis/v1/services/aigc/video-generation/video-synthesis" \
  -H "Authorization: Bearer $AISA_API_KEY" \
  -H "Content-Type: application/json" \
  -H "X-DashScope-Async: enable" \
  -d '{
    "model":"wan2.7-i2v",
    "input":{
      "prompt":"gentle zoom with parallax",
      "media":["https://example.com/reference.jpg"]
    },
    "parameters":{"resolution":"720P","duration":5}
  }'

Poll task → GET /apis/v1/services/aigc/tasks/{task_id}#

Documentation: Get video generation task result.

task_id is a path parameter. The query-string form ?task_id=... returns HTTP 500 unsupported uri.

bash

curl "https://api.aisa.one/apis/v1/services/aigc/tasks/YOUR_TASK_ID" \
  -H "Authorization: Bearer $AISA_API_KEY"

---

Python client#

The bundled client at scripts/media_gen_client.py auto-routes each image model to the correct endpoint and normalizes the response to a saved file.

bash

# Image — model selects the endpoint
python3 scripts/media_gen_client.py image \
  --model <gemini-3-pro-image-preview | wan2.7-image | wan2.7-image-pro | seedream-4-5-251128> \
  --prompt "..." \
  --out out.png

# Video — create task
python3 scripts/media_gen_client.py video-create \
  --model <wan2.6-t2v | wan2.6-i2v | wan2.7-t2v | wan2.7-i2v> \
  --prompt "..." \
  [--img-url https://... (required for -i2v models)] \
  [--duration 5|10] \
  [--resolution 720P|1080P]

# Video — poll / wait / download
python3 scripts/media_gen_client.py video-status --task-id <id>
python3 scripts/media_gen_client.py video-wait --task-id <id> --poll 10 --timeout 600
python3 scripts/media_gen_client.py video-wait --task-id <id> --download --out out.mp4

API reference#

This skill calls the following AIsa endpoints directly:

• Google Gemini Chat — generateContent — Gemini image models
• Image Generation via Chat — Wan 2.7 image family
• OpenAI-Compatible Image Generations — Seedream
• Create video generation task — all four Wan video variants
• Get video generation task result — async polling

See the full AIsa API Reference for the complete catalog.

License#

MIT — see LICENSE at the repo root.