Tarseek API · Pre-production
Images
Image understanding (vision)
Send images as message parts in chat. Use this for describing, OCR-lite, and visual Q&A on models whose input modalities include image.
image_url and input_image parts and routes only to deployments that support images. No dedicated vision endpoint. Token estimation for images is a heuristic (~3400 chars), not a billed vision-token formula.Image generation
POSThttps://tarseek.com/v1/images/generations
Send JSON with model, prompt, and optional n (1–10), size, quality, output_format, output_compression or background. Native GPT Image deployments return data[].b64_json and measured token usage.
max_cost_usd is a request safety limit, default USD 1, based on reported or catalog-estimated provider cost. The provider bills your BYOK account directly. Existing policy token-price ceilings also apply to image-token rates.
Image editing
POST https://tarseek.com/v1/images/edits accepts the same JSON options with image_file_ids (1–16 ready project files) and optional mask_file_id. Upload through the Files API first and wait for scanning. A mask must be a PNG with an alpha channel and match the first image's dimensions. File-read permission is required.
Input files are limited to 20 MB each and 50 MB combined. Responses are limited to 64 MB. Dedicated image APIs currently support GPT Image through OpenAI-compatible deployments. They return transient base64 data; they do not automatically retain generated images. Multipart edit uploads, provider URLs, streaming partial images, DALL·E and non-GPT adapters are not supported by these endpoints. Other image-output models may use their existing Chat integration.
Pricing uses separate text-input, image-input and image-output quantities. Missing usage or an unsupported cached-token breakdown fails explicitly; no guessed prices are used.