agnes-media-generation
Use only when the user explicitly mentions "agnes" (case-insensitive) and asks to generate, edit, combine, or animate images or videos with Agnes AI. Do not trigger for generic media generation without Agnes, media analysis, text generation, search, or live streaming.
Recursos
10Instal·la
npx skillscat add ryderey/agnes-mcp-studio Instal·la des del registre de SkillsCat.
Agnes Media Generation
Use the agnes_media MCP server. If its tools are unavailable, ask the user to check the MCP configuration. Never activate without both the Agnes keyword and a media-generation intent.
Workflow
- Resolve the prompt language.
- Select the narrowest matching tool.
- Build a specific visual prompt and call the tool.
- Present saved files and actionable errors.
Tool selection
| Task | Tool | Key options |
|---|---|---|
| Standard image or image-to-image | agnes_image_generate |
Exact-pixel size; optional image_urls |
| High-resolution image, wallpaper, poster | agnes_image_generate_v2 |
size=1K–4K; ratio; optional image_urls |
| Edit, restyle, or combine images | agnes_image_edit |
One or more URL/local image_paths |
| Text-to-video | agnes_video_generate |
Default all-in-one generation |
| Image-to-video | agnes_video_generate |
Public image URL; mode="ti2vid" |
| Keyframe transition | agnes_video_generate |
mode="keyframes"; URL array in extra_body.image |
| Async video or later polling | agnes_video_submit → agnes_video_wait / agnes_video_status |
Preserve and report video_id |
Defaults: use agnes_image_generate at 1024x1024; for requested high resolution use _v2, normally 2K with the requested ratio. Use agnes_video_generate unless the user needs asynchronous progress or later retrieval.
Prompt language
Resolve the mode from the current request, then an explicit earlier preference, otherwise use auto:
auto: translate non-English prompts into natural English and optimize silently.original: optimize without changing language when the user asks to preserve it.review: show original and English versions and wait for selection when requested.
Preserve intent, proper nouns, numbers, camera directions, and quoted/on-screen text. Apply the same mode to negative_prompt. Use image prompts shaped as subject + scene + style + lighting + composition; use video prompts shaped as subject + action + scene + camera + lighting.
Constraints and recovery
- Image inputs accept public HTTPS URLs, Data URIs, and local paths; video inputs require public HTTPS URLs.
- At 24 fps, video duration is limited to about 18 seconds. Let the MCP handle frame alignment and
response_format. - Mask-based editing is unsupported.
- On 429 or 503, wait and retry. On timeout, report
video_idso polling can continue. For other errors, report the returned code and message.
Output
- Show every generated image inline from its absolute
local_pathsentry. - For video, provide
local_pathandvideo_urlwhen available. - Do not expose credentials or invent a successful result when a tool returns an error.
Categories
Instal·la
npx skillscat add ryderey/agnes-mcp-studio Instal·la des del registre de SkillsCat.