Media Gen

Image and video generation for autonomous agents with Seedream, Wan, and GPT Image models through AIsa’s image and video routes.

View on GitHub →

Generate images and videos with one AIsa API key. The skill guides your agent on which model and route to use for an image or video request, and bundles media_gen_client.py for Seedream image generation.

Install

npx skills

npx skills add AIsa-team/agent-skills --skill media-gen

After installing, start a new agent session so the skill loads.

What can agents do with it?

Marketing creatives

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

Large-format images

“Generate a 2048×2048 magazine cover: neo-noir detective portrait, film grain.”

Text-to-video

“Sweeping establishing shot of a neon cyberpunk skyline at dusk, 5 seconds.”

Image-to-video

“Starting from this reference image, gentle camera push-in with parallax.”

Requirements

• Python 3 and a POSIX shell
• AISA_API_KEY in the environment

Generation calls are billed to your AIsa balance per request or per output, depending on the model. See AI Model Pricing for how each model type is billed.

Current models and routes

The live gateway serves these media models. Check GET /v1/models for the current list; each model lists its preferred route.

Kind	Models	Route
Image	seedream-5-0-260128, wan2.7-image, wan2.7-image-pro, gpt-image-2, gpt-image-2.5-flare, gpt-image-2.5-sunburst	POST /v1/images/generations
Video	wan2.7-t2v, wan2.7-i2v, wan2.7-r2v	POST /v1/video/generations, then GET /v1/video/generations/{task_id}

The skill’s bundled client still targets older routes and model IDs. Its image command works with Seedream (--model seedream-5-0-260128), but it sends Wan image models to /v1/chat/completions, which no longer serves image models, and its video-* commands call a retired video route. Until the skill is updated, generate Wan images and videos through the routes above, not through the client.

• Seedream rejects images below 3,686,400 pixels (for example 1024×1024). Use a size such as 1920x1920 or 2048x2048. wan2.7-image accepts smaller sizes.
• Video generation is asynchronous: create a task, then poll it by task_id until it finishes.
• Each image in n is billed separately. Request a free quote first by adding X-AISA-Cost-Mode: quote to the same request; see Billing & Credits.

Quick start

With the skill’s client (Seedream):

export AISA_API_KEY="your-key"

python3 scripts/media_gen_client.py image \
  --model seedream-5-0-260128 \
  --prompt "Neo-noir detective portrait, film grain" \
  --size 2048x2048 \
  --out out.png

Any image model, directly over HTTP:

curl -sS -X POST "https://api.aisa.one/v1/images/generations" \
  -H "Authorization: Bearer $AISA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan2.7-image-pro",
    "prompt": "Ultra-detailed product shot of a red panda, studio lighting",
    "n": 1,
    "size": "2048x2048"
  }'

Request and response fields are documented in Image Generations.

Get started

1. Sign up at aisa.one and create an API key in the console.
2. Set it in your shell: export AISA_API_KEY="your-key".
3. Install the skill with one of the commands under Install, then start a new agent session.

Related

Image generations

OpenAI-compatible image generation endpoint and current image models.

Model catalog

Current model IDs, modalities, and routes.