Generate AI images and videos from Cursor: 48 models (Flux, Imagen 4, Seedream, Kling, Seedance, Veo 3.1, Wan, PixVerse) on one credit balance, cost returned before each call, automatic refund when a generation fails. Built by Frank Houbre, AI expert.
Generate AI images and videos through the Imaginode MCP server (imaginode.ai). Use when the user asks to create, generate or illustrate an image, picture, visual, thumbnail, or a short video clip, and the Imaginode MCP tools (generate_image, generate_video, list_models, get_generation_status) are available.
# Imaginode: AI image and video generation
You have access to the Imaginode MCP server (tools `list_models`, `generate_image`, `generate_video`, `get_generation_status`). Every call spends credits from the user's imaginode.ai account (1 credit = €0.01), failed generations are refunded automatically, and the exact cost comes back with each call.
## Workflow
1. **Iterate cheap, finish premium.** For exploration or drafts, use `flux-schnell` (1 credit per image). Once the user validates the composition, regenerate with a premium model (`nano-banana-2`, `flux-2`, `seedream-5-pro`). Never burn premium credits on a first attempt unless the user asks for top quality immediately.
2. **Prompts in English, concrete and visual.** Subject, setting, light, style, framing. The user's language does not matter; translate the intent into a strong English prompt and show the user what you sent.
3. **Announce costs.** Each result includes the cost in credits and the remaining balance: relay them briefly ("2 credits, 4,310 left"). If a call fails with "Not enough credits", point the user to https://imaginode.ai/pricing and stop generating.
4. **Video is asynchronous.** `generate_video` returns a `request_id` and the cost immediately. Poll `get_generation_status` about every 30 seconds; typical jobs take 1 to 5 minutes, long ones up to 30. Keep working on other things between polls and never re-launch the same video because it "seems slow": that costs credits.
5. **Model choice.** Call `list_models` when the user names a style or need you cannot map to a model. Guidelines: photorealism → `flux-2` or `imagen`; image editing or character consistency with reference images → `nano-banana-2` or `seedream-5-pro`; cheap video → `kling-std`; video with generated audio → `veo31-lite` or `veo3-fast`; cinematic top-tier video → `seedance` or `kling3-std`.
6. **Input images** (image-to-video, editing) must already be hosted on the user's Imaginode account: reuse output URLs from previous generations, or ask the user to upload the file to their media library at imaginode.ai first.
## Guardrails
- One generation at a time unless the user explicitly asks for a batch; respect the 10 generations/minute per-key limit.
- Never invent an output URL: only relay URLs returned by the tools.
- If the API key is missing or revoked, tell the user to create one at https://imaginode.ai/profile (API keys section) and reconnect the MCP server.