How do I get my own image to the editing tools?

upload_asset takes a local file (or URL) from your agent and returns an asset reference the edit tools accept. In web clients like Claude and ChatGPT, attach the image to the chat and the agent handles the upload for you.

What kinds of edits work well by prompt?

Recolors and restyles, background cleanup or replacement, object removal, lighting changes, seasonal variants, and composition-preserving reworks. Precision pixel work (exact text replacement, tiny logos) is still better in a design tool — the sweet spot is fast semantic edits.

Which editing model should the agent pick?

qwen-image-edit-plus ($0.03) is the default for faithful, instruction-following edits. nano-banana ($0.039) is the budget iteration loop; nano-banana-pro ($0.134) for the final high-fidelity pass. Agents can chain them — cheap iterations, premium finish.

Can the agent edit something it generated in an earlier session?

Yes — that's the point of the creation history. search_creations finds the original by description or date, get_creation pulls it with metadata, and edit_image works on it directly. No re-uploading, no regenerating from scratch.

Does iterating on edits get expensive?

A five-pass iteration loop on nano-banana is $0.194 total. Every pass is priced per output, failures are refunded, and your monthly spending limit caps the worst case — so agents can iterate freely.

Are my uploaded images used for anything else?

Uploads exist to serve your edits — they live in your account's asset store, are retrievable only by your keys/OAuth grants, and can be deleted from your NinjaChat dashboard.

Create free key

Image editingTAKT

Image editing by prompt, via MCP.

Upload any image and your agent edits it with words — recolor, restyle, remove, extend. Qwen Image Edit Plus and the Nano Banana family, $0.039–$0.134 per edit pass.

Qwen Edit$0.03

Nano Banana 2$0.067

Seedream$0.04

Connect in 60 seconds

Generated original: a blue 911 on a rain-soaked Times Square — `generate_image`the original — blue

The same frame after the edit: the 911 is now red, everything else untouched — `“paint it guards red”`the edit — $0.03

A real pair from one NinjaChat run: same street, same rain, same reflections — only the car changed, because that's all the sentence asked for.

Paste this. Watch it run.

First ask

Here's our team photo — clean up the cluttered whiteboard behind us and warm the lighting.

upload_assetedit_image

1 edit on qwen-image-edit-plus · $0.03

Use case · product consistency

One recorder. Every light.

Macro, desk, hero — TAKT stays locked while the set changes around it. Same product language, three frames.

One product. Macro, desk, hero — consistent under every light.

Models

The editors, and what each is for.

An editorial shot of a cobalt coat in white dunes — a real FLUX.2 Pro run

FLUX.2 ProFLUX.2 Pro

A flamenco dancer mid-turn under a single spotlight — a real Seedream run

Seedream 4.5Seedream

A red-sailed boat crossing a mirror-still turquoise fjord — a real Nano Banana 2 run

Google Nano Banana 2Nano Banana 2

A matte espresso machine in a shaft of morning light — a real Nano Banana Pro run

Google Nano Banana ProNB Pro

A red fox mid-leap over a frozen creek at sunrise — a real Nano Banana run

Google Nano BananaNano Banana

A hummingbird frozen mid-air at a fuchsia flower — a real Z-Image Turbo run

Z-Image TurboZ-Image

GPT Image 2's take on the tiny astronaut asleep on a crescent moon

GPT Image 2GPT Image

A mid-century travel poster of balloons over tea hills — a real Recraft V3 run

Recraft V3Recraft

A guards-red 911 on a rain-soaked street — the single-subject edit

Qwen Image EditQwen Edit

Tools

21 tools. These do the work here.

Make

generate_imageCreate images and download the full-resolution results.
edit_imageChange a background, color, or detail with a prompt.

Library

search_creationsSearch your images, videos and voiceovers.
get_creationView a saved result and its generation details.
upload_assetUpload an image to edit or use as a reference.

Also on the server

create_mediagenerate_speechgenerate_videoimage_to_videoedit_videoextend_videoget_generation_statussave_creation_aslist_creative_entitiescompare_modelssearch_modelsestimate_costprepare_scrub_videopublish_pageget_account_balanceadd_credits

Setup

Connected in about a minute.

Pick your client — the steps for the others stay one tap away.

ChatGPT4 steps

Enable developer mode
In ChatGPT web, open Settings → Security and login and turn on Developer mode. Availability depends on your account and workspace policy.
2
Add NinjaChat
Open Plugins, select the plus button, name it NinjaChat, and enter the server URL under Connection.
text
```
https://www.ninjachat.ai/api/mcp
```
Approve the connection
Sign in to NinjaChat, review the permissions, and set a spending limit before connecting ChatGPT.
Use it in a chat
Start a new conversation, choose Developer mode from the plus menu, and select NinjaChat before sending your prompt.

VS Code4 steps

Create an API key
Sign in at ninjachat.ai and create a key under Developers → Keys. Keys look like nj_sk_… and carry your spending limit.

Create .vscode/mcp.json

Merge the servers and inputs below into your workspace file. The key stays out of the file; VS Code prompts for it securely on first start.

.vscode/mcp.json

{
  "inputs": [{ "type": "promptString", "id": "ninjachat-key", "description": "NinjaChat API key", "password": true }],
  "servers": {
    "ninjachat": {
      "type": "http",
      "url": "https://www.ninjachat.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer ${input:ninjachat-key}"
      }
    }
  }
}

Start the server
VS Code shows a “Start” code-lens above the server entry. Start it, then check the tools list in Copilot Chat's tool picker.
Use it in agent mode
Switch Copilot Chat to Agent mode, then send your prompt.

Want the full walkthrough with troubleshooting? ChatGPT guide · VS Code guide

Questions, answered.

Keep exploring.

imageAI Image Generation over MCP

11 image models from $0.003/image, callable from any MCP agent

videoAI Video Generation over MCP

Veo 3.1, Seedance 2, and Kling from your agent — async with status polling

NinjaChat MCP in ChatGPT

Connect ChatGPT to 22 image, video and voice models with a custom connector

NinjaChat MCP in VS Code

Copilot agent mode + every creative model via .vscode/mcp.json

Model Context Protocol NinjaChat MCP docs

TAKT · Field recorder

One sentence, exactly one change.

Product locked. Light and set free to move.

Connect in 60 seconds

Every frame of TAKT was generated through the NinjaChat MCP server.