teamToken media

teamToken is a gateway to more than 50 image and video models behind one API key, billed per generation. This skill covers three things: picking a model, knowing the cost before a run, and collecting the result without paying for it twice.

When to activate

• The user wants a picture generated, edited, restyled or extended.
• The same character or product has to appear in several images.
• The user wants a video from a prompt, from a still image or from another video.
• An existing clip needs to be continued, or a character has to repeat the motion from a reference video.
• The user asks which image or video model to use, or what one costs.

Two ways in

The MCP tools are connected (list_models, generate_image, edit_image, generate_video, extend_video, get_job, get_balance). Use them. The server is https://teamtoken.store/mcp.

No MCP tools, but there is a shell. Call the REST API at https://api.teamtoken.store with the key from the TEAMTOKEN_KEY environment variable. If it is empty, ask the user for a key from https://app.teamtoken.store. Never echo the key back into the conversation or into files.

Before you generate

1. Take the model from the live catalog, not from memory. Ids and prices change.

• MCP: list_models. It is free and needs no key.
• REST: curl -s https://app.teamtoken.store/cabinet/api/public/media-models

Each row has model, modality (image or video), billing_unit (per_image or per_second) and unit_price_usd.

2. Work out the cost and tell the user. An image run costs the price times n. A video costs the per-second price times the duration. If the user named no budget and the estimate is above $1, ask before running.

3. Check that the model accepts the input. A mismatch fails the request:

Task	Send	Models
Text to image	prompt	every image model
Edit, restyle, same character	prompt + 1–4 images	nano-banana-pro, nano-banana-2, grok-image, gpt-image-2-*
Text to video	prompt	every video model
Image to video	prompt + image	seedance, kling, veo, grok, omni-flash
Edit a video	prompt + video	kling-3.0-edit-*, seedance-2-omni-*
Motion control	prompt + image (the character) + video (the motion)	kling-3.0-motion-*, kling-2.6-motion-*

Allowed video lengths depend on the family: veo 4, 6 or 8 seconds, omni-flash 4, 6, 8 or 10, seedance 4 to 15, kling 3 to 15, grok 6. When the id names a length (omni-flash-6s, kling-2.1-10s-720p), use that one. Edit and motion models take the length from the input video and ignore duration.

Pictures go in as JPG or PNG up to 10 MB each, video as MP4, MOV or WebM up to 100 MB. Pass https URLs, data-URLs or bare base64, and keep to one kind per request.

Images: one call, the picture comes back

MCP: generate_image, or edit_image when there are source pictures. Both wait until the image is ready and return it. There is nothing to poll.

REST:

curl -s https://api.teamtoken.store/v1/images/generations \
  -H "Authorization: Bearer $TEAMTOKEN_KEY" -H "Content-Type: application/json" \
  -d '{"model": "nano-banana-pro",
       "prompt": "a red ceramic teapot on a worn oak table, morning window light",
       "aspect_ratio": "1:1", "resolution": "1K"}' \
  -o response.json
jq -r '.data[0].b64_json' response.json | base64 -d > image.png

To edit, or to keep a character, add "images": ["https://..."] with up to four pictures to the same request.

When the provider is slow the reply is HTTP 202 with an id and "status": "processing". That job is already paid for. Collect it with GET /v1/images/jobs/{id} rather than sending the request again.

Save images right away. After 7 days the job answers with "archived": true and an empty data, which means the result expired, not that the generation failed.

Video: start, wait, collect

A clip takes from 30 seconds to a few minutes, so the call returns before it is done.

1. Start it. MCP: generate_video or extend_video returns a job id at once. REST: POST /v1/videos with "wait": false returns {"id", "status"} with HTTP 202.
2. Wait about 30 seconds, then check. MCP: get_job. REST: GET /v1/videos/{id}.
3. While the status is queued or processing, wait another 30 seconds and check again. completed means done, failed means it stopped.
4. Do not send the request again while the job runs. Each submit is billed on its own.

curl -s https://api.teamtoken.store/v1/videos \
  -H "Authorization: Bearer $TEAMTOKEN_KEY" -H "Content-Type: application/json" \
  -d '{"model": "veo-3.1-fast-hd",
       "prompt": "slow push-in on a lighthouse at dusk, waves breaking below",
       "duration": 6, "aspect_ratio": "16:9", "wait": false}'
# {"id": "vid_...", "status": "processing", ...}

curl -s https://api.teamtoken.store/v1/videos/vid_... \
  -H "Authorization: Bearer $TEAMTOKEN_KEY"

The finished file is served by the gateway and needs the same key, so opening the link in a browser gives 401. Download it with the header:

curl -L -H "Authorization: Bearer $TEAMTOKEN_KEY" -o result.mp4 \
  https://api.teamtoken.store/v1/videos/vid_.../content

Money and errors

• Every image and video response carries its price. cost_usd is what was charged, estimated_cost_usd is the amount held while a job runs. Report cost_usd.
• A failed generation is not billed and comes back with "cost_usd": 0. Change the prompt or the input before the next attempt.
• 402: the balance is short. Call get_balance (REST: GET /v1/balance), tell the user how much is missing and point them to https://app.teamtoken.store/topup.
• 401: the key is missing or wrong.
• GEMINI_RAI_MEDIA_FILTERED and other filter codes: the provider's safety filter refused the content. The same prompt will be refused again, so rewrite it.
• INVALID_VIDEO_FILE: an edit or motion model was called without video.
• SYSTEM_ERROR: a temporary fault on the provider side. Retry once after a pause.

Prompts

• Images: name the subject, the setting, the light, the framing and the style. "A red ceramic teapot on a worn oak table, morning window light, shallow depth of field" gets a better picture than "a teapot".
• Video: one shot per clip. Say what moves and what the camera does, for example "static camera, the dog runs from left to right" or "slow dolly-in". Three scene changes in one prompt tend to come out as a blur of all three.
• Image to video: describe the motion. The model already sees the picture.
• Prompts shorter than 10 characters are rejected.