Models

List models

GET /v1/models?type=image

Returns all enabled image models with their cost and accepted params. This is the canonical source of truth — use it to discover new models or validate param names.

curl -s https://api.goodtake.ai/v1/models?type=image \
  -H "Authorization: Bearer $GOODTAKE_API_KEY" | jq .

Image models

Model identifier	Provider	Notes
`gpt-image-2`	fal (OpenAI)	Excellent text-in-image, editing via `image_urls`
`seedream-5-0-260128`	ByteDance	Latest Seedream, highest quality
`seedream-4-5-251128`	ByteDance	Proven quality, broad size support
`seedream-4-0-250828`	ByteDance	Legacy, use for backward compatibility
`gemini-3.1-flash-image-preview`	Google	Fast, grounding and thinking support
`gemini-3-pro-image-preview`	Google	Pro-tier, multi-step reasoning

Video models

Video generation uses a separate endpoint: POST /v3/contents/generations/tasks.

Model identifier	Provider	Notes
`dreamina-seedance-2-0-mini-260615`	ByteDance (Dreamina)	Multimodal-to-video; reference & first/last frame; 480p–720p

Pricing

Condition	Rate
Input includes video	2.1 USD / M tokens
Input is text / image only	3.5 USD / M tokens

Video request format

Unlike the image API, video models use a content array for multimodal inputs and accept generation params as top-level fields:

{
  "model": "dreamina-seedance-2-0-mini-260615",
  "content": [
    { "type": "text", "text": "your prompt" },
    {
      "type": "image_url",
      "image_url": { "url": "https://example.com/ref.jpg" },
      "role": "reference_image"
    },
    {
      "type": "video_url",
      "video_url": { "url": "https://example.com/ref.mp4" },
      "role": "reference_video"
    },
    {
      "type": "audio_url",
      "audio_url": { "url": "https://example.com/bg.mp3" },
      "role": "reference_audio"
    }
  ],
  "ratio": "16:9",
  "resolution": "720p",
  "duration": 8,
  "generate_audio": true,
  "watermark": false
}

Content roles

Role	Type	Description
`reference_image`	`image_url`	Reference image for style or subject
`reference_video`	`video_url`	Reference video for motion or framing
`reference_audio`	`audio_url`	Background audio / music

Generation modes (set in the UI; controlled by input content in the API)

Mode	How to use
Reference generation	Supply `reference_image` / `reference_video` / `reference_audio` items in `content`
First & last frame	Supply two `reference_image` items — first frame and last frame, in order

Supported params per model

Pass model-specific options in the params object of your generate request. Unknown keys are rejected with a 400 and the list of supported keys.

`gpt-image-2`

Param	Type	Description
`image_size`	string	e.g. `1024x1024`, `1792x1024`
`quality`	string	`low`, `medium`, `high`, `auto`
`num_images`	integer	1–4
`output_format`	string	`png`, `jpeg`, `webp`
`mask_url`	string	For inpainting edits; `image_urls` must also be set

`seedream-*` (ByteDance)

Param	Type	Description
`size`	string	`1K`, `2K`, `4K`, `2048x2048`, `2560x1440`, etc.
`number_of_images`	integer	1–4
`optimize_prompt_mode`	string	`standard`, `creative`, `precise`
`output_format`	string	`png`, `jpeg`

`gemini-*` (Google)

Param	Type	Description
`aspect_ratio`	string	e.g. `16:9`, `1:1`, `9:16`
`resolution`	string	`standard`, `high`
`number_of_images`	integer	1–4
`seed`	integer	For reproducibility
`thinking_level`	string	`low`, `medium`, `high`
`enable_thinking`	boolean	Enables multi-step reasoning
`enable_grounding`	boolean	Grounds generation on context
`enable_image_search`	boolean	Uses image search context
`output_format`	string	`png`, `jpeg`

`dreamina-seedance-2-0-mini-260615`

Top-level fields (not nested under params):

Field	Type	Values / notes
`ratio`	string	`21:9`, `16:9`, `4:3`, `1:1`, `3:4`, `9:16`, `auto`
`resolution`	string	`480p`, `720p`
`duration`	integer \| string	Seconds in range 4–15, or `"smart_length"`
`generate_audio`	boolean	Generate audio track for the output video
`watermark`	boolean	Overlay provider watermark

Pass hosted http(s) or data: URIs in the top-level image_urls array to provide reference or edit images. For gpt-image-2, supplying image_urls routes the request to the edit endpoint automatically.

Internal/private addresses (e.g. localhost, RFC 1918 ranges) are rejected. Reference images must be publicly reachable.

Models

On this page