nim-generate

name

nim-generate

description

Generate or edit images and videos with Nim. Use when the user says "generate an image", "make a picture / poster / logo / wallpaper", "edit / restyle / remix this image", "make a video", "animate this photo", "image-to-video", "turn this into a clip", "give me variations / options", or asks for any visual asset from a prompt or a reference. Discovers the right Nim model with models_explore, follows its generationContract exactly, uploads references with media_upload, and submits with generate_image / generate_video. Supports text-to-image, image editing, text-to-video, and image-to-video. NOT for: code, text/chat answers, or audio-only tasks.

Nim Generate

Generate and edit images and videos through the Nim MCP. One skill for both: model discovery is shared, and the only real branch is image vs. video.

UX rules

1. Be concise. Don't dump raw IDs, JSON, or tool plumbing into the chat — report the finished result, not the mechanics.
2. No jargon. Don't narrate "calling models_explore" or "polling status."
3. Reply in the user's language. Detect it from their message; keep technical values (16:9, 1080p) as-is.
4. Pick a sane default and move. Don't interrogate. Ask one thing at a time, and only when a required input is genuinely missing.
5. Don't pre-optimize for cost. Prefer the quality default first; only weigh cheaper models if the user asks. Check get_credit_balance before a deliberately expensive run (high resolution, long video, large batchSize).
6. Never invent Nim URLs. generate_image / generate_video return no link. Get the result by polling get_generation_status and report only the URL it actually returns.
7. No upload, no reference generation. For any edit, image reference, or image-to-video request, do not call generate_image / generate_video until every reference has a real Nim file_url from a successful media_upload upload. MCP generation tools and widgets do not upload local files for you.

Workflow

1. Pick the model. The Nim catalog is live — discover, don't hardcode.
   • Unsure which model fits → models_explore with action: "recommend", the user's intent as query, type: "image" or "video", and input: "image" when a reference will be supplied. For video, pass minDurationMs if the user named a length.
   • Comparing options → action: "search" / "list".
   • Then action: "get" on the chosen model_id to read its exact generationContract. This is the source of truth for which params are allowed.
2. Prepare references. For image editing, image references, or image-to-video, the reference must be a Nim file URL first:
   • Call media_upload, run the returned curl_example against the local path or Claude attachment, then pass the response file_url in fileInputs.
   • Never pass a local filesystem path straight to a generation tool.
   • If the file is not actually readable by the agent, follow When uploading media.
3. Generate. Call generate_image or generate_video and pass only params the generationContract allows. Submit model_name alongside model_id for clearer user-facing status.
4. Poll, then deliver. generate_image / generate_video start an async job and return no result and no link. You are running in a terminal client (Claude Code / Codex) that does NOT render the Nim generation widget — so you must poll get_generation_status with the returned workflowId / promptId until finished / failed / cancelled, then deliver the final media URL from that response.
   • Never tell the user the result "will appear in a widget" — it will not appear here. Never describe a queued job as done.
   • The Nim widget only renders in MCP-Apps hosts (claude.ai, ChatGPT, Claude Desktop), which connect to the MCP directly rather than through this skill. Ignore it here.

Image vs. video — quick routing

• Still asset, edit, restyle, logo, poster, character art → image (generate_image).
• Motion, clip, animation, or "make this image move" → video (generate_video).
• "Animate this photo" / "image-to-video" → video with the reference in fileInputs (choose an image-input video model via input: "image").

Inputs

Pass a param only when the selected model's generationContract lists it.

Param	Applies to	Notes
prompt	both	Required. Carry the full creative intent.
model_id	both	Required. Always from models_explore.
fileInputs	both	Nim file URLs from media_upload. Required for edit / image-input / image-to-video models.
requestedAspectRatio	both	e.g. 16:9, 9:16, 1:1.
resolution	both	e.g. 1080p, 2K — only values the contract offers.
mediaLength	video	Input duration in ms (5000, 8000).
fps, keepSound	video	Only when the model supports them.
seed	both	Omit or -1 to randomize; reuse a value to reproduce.
batchSize	both	1–4 variations in one call. Use for ideation / options.

When uploading media

• Never pass local filesystem paths directly to generate_image or generate_video; use media_upload and pass the returned file_url.
• If a request depends on a reference file and upload fails, stop before generation. Do not call generate_image or generate_video without the reference just because the Nim MCP widget or generation tool is available.
• In Claude Cowork, paths from the user's computer are never readable by the agent unless the file is in the current Cowork Working folder. If a referenced file cannot be read, stop and ask the user to add it to the current Cowork Working folder, then retry from that accessible file.
• Inline previews or images visible in chat are not always uploadable files. If no actual file path/attachment/source is available, ask for one before upload.
• If media_upload returns an upload URL but the actual upload is blocked by DNS, network egress, CSP, sandboxing, or allowlist restrictions, stop. Do not try alternate domains, proxy URLs, guessed endpoints, custom upload formats, or invented fileInputs values such as upload:/path, local paths, or placeholder URLs. Do not claim the Nim widget can upload the file directly. Tell the user to add *.nim.video to their allowlist; the setup page is https://nim.video/mcp. Retry only after they confirm the environment is fixed.

Variations & ideation

When the user wants options, directions, or "a few ideas," set batchSize (2–4) so Nim returns several takes from one request, each with a randomized seed. Present them as a set to compare, not as a single answer.

When a generation fails

Surface the returned error / errorCode plainly. Don't silently retry with changed params — ask the user first, unless they already told you to iterate.