Use apimodels in n8n: chat, images, video and speech

1. Create the task

POST https://api.apimodels.app/v1/images/generations (video: /video/generations), JSON body:

{
  "model": "gpt-image-2",
  "prompt": "{{ $json.prompt }}",
  "image_urls": ["{{ $json.image_url }}"],
  "aspect_ratio": "4:3",
  "resolution": "2K"
}

{ "code": 200, "msg": "success",
  "data": { "taskId": "cmv...", "state": "pending", "model": "gpt-image-2" } }

Reference images must be publicly reachable. Send a form-uploaded file as multipart to https://api.apimodels.app/v1/files (field name file, Parameter Type n8n Binary File) and put the returned data.publicUrl into image_urls or first_frame_url. Uploads are free.

2. Wait, check, route

A Wait node pauses 15 seconds (20 for video), then GET:

https://api.apimodels.app/v1/images/generations?task_id={{ $('Create image task').first().json.data.taskId }}

{ "code": 200, "msg": "success",
  "data": {
    "taskId": "cmv...",
    "state": "completed",                // pending / processing / completed / failed
    "resultUrls": ["https://r2.apimodels.app/images/cmv....png"],
    "credits": 0.03,                     // USD actually charged; 0 when failed
    "failMsg": null
  } }

A Switch node routes on data.state: completed continues; failed shows data.failMsg as is (it says why when the model refuses); anything else loops back to Wait. Add an IF on $runIndex to cap the number of checks. Status checks are free, so Retry On Fail is safe on that node.

One call that returns the image (for agent tools)

Agent tools cannot poll easily. Send an OpenAI-style size to /images/generations without callback_url and the same request waits for the image and returns an OpenAI-shaped result. Set the HTTP Request Tool timeout to 300 seconds.

{
  "model": "gpt-image-2",
  "prompt": "{{ $fromAI('prompt', 'Detailed description of the image', 'string') }}",
  "size": "1536x1024",
  "response_format": "url"
}
// → { "created": 1792..., "data": [ { "url": "https://r2.apimodels.app/images/....png" } ] }

An n8n form takes a product photo and a scene and returns an ad image with the product itself (shape, colors, logo, text) unchanged; the result page shows the amount charged.

Models: GPT Image 2 · Typical run: 1K about 30 s, 2K about 2 min

Describe a shot, optionally upload a photo as the first frame, and get a 5–15 s video with native sound; pick one of three Seedance 2.0 tiers.

Models: Seedance 2.0 Mini / Fast / Standard · Typical run: 5 s Mini 480p about 2–3 min

Paste an article link or text; Claude rewrites it as a spoken script and ElevenLabs v4 reads it into a shareable MP3.

Models: Claude Sonnet 5.5 + ElevenLabs v4 · Typical run: 1-minute episode about 40 s

In n8n's hosted chat, an AI Agent answers questions and, when asked to draw, calls an image tool and puts the picture in its reply.

Models: Claude Haiku 5.5 + GPT Image 2 · Typical run: Image replies about 1–2 min

Paste a YouTube or Bilibili link to get a blog post, X thread, LinkedIn post and newsletter blurb written from the video's own subtitles, plus an optional 16:9 cover. Subtitles, writing and the cover use one key, with no separate transcript service.

Models: Claude Sonnet 5.5 + GPT Image 2 · Typical run: 19-minute video about 2.5 min, $0.063 with a cover

Send one prompt to 2–6 models and see every answer on one page with each model's server-measured time, tokens and exact charge from the billing records endpoint; the fastest and lowest-cost models are marked.

Models: Claude, GPT, Gemini, DeepSeek, Qwen, Grok, GLM · Typical run: 5 models about 15 s, under $0.01 in total

Each message is classified by DeepSeek (category, sentiment, urgency, one-line summary, tags) and saved to a free Cradler database (built by the same team as apimodels; the table is created on the first write). The team opens a report page, picks 7 or 30 days, and Claude writes the themes, urgent issues and suggested actions.

Models: DeepSeek V4.1 Flash + Claude Sonnet 5.5 + Cradler · Typical run: About $0.0001 per message; a weekly report about 25 s and $0.006

Upload a CSV or Excel file into a free Cradler database (headers become clean column names and each column gets one type), then ask questions in n8n's hosted chat. The agent lists the tables, reads the columns, queries with filters and uses a calculator for sums and averages. The chat tools are read-only.

Models: Claude Sonnet 5.5 + Cradler · Typical run: 400-row file imported in 9 s; all 4 test questions answered correctly, 20–95 s each