---
title: "Create image (text-to-image)"
method: POST
path: "/v1/images/generations"
tags: ["Images"]
---

# Create image (text-to-image)

`POST /v1/images/generations`

Creates one or more images from a text prompt. Only registered when the active image runner_type is a text-to-image variant.

## Request body

- ImageGenerateRequest — Superset of fields for /v1/images/generations, /v1/images/image-to-image, and /v1/images/edits. The active route enforces its own subset (image-to-image requires `image`; edits also requires `mask`).
  - `prompt` string, required — Text prompt describing the desired output image
  - `prompt_2` string — Optional secondary prompt routed to the second text encoder (SDXL)
  - `negative_prompt` string — Negative prompt describing what to avoid
  - `negative_prompt_2` string — Optional secondary negative prompt for the second text encoder
  - `num_inference_steps` integer — Number of denoising steps
  - `guidance_scale` number — Classifier-free guidance scale
  - `guidance_rescale` number — Guidance rescale factor (mitigates over-saturation at high guidance)
  - `seed` integer — Random seed for reproducible sampling
  - `number_of_images` integer — Number of images to generate per request
  - `crop_coords_top_left` integer[] — SDXL crop conditioning [top, left]
  - `timesteps` number[] — Optional explicit list of timesteps for the scheduler
  - `sigmas` number[] — Optional explicit list of sigmas for the scheduler
  - `lora_path` string — Path to a LoRA weights file to load before generation
  - `lora_scale` number — Scale applied to the LoRA
  - `image_return_format` 'JPEG' | 'PNG' — Encoding format for the returned base64 image bytes
  - `image_quality` integer — Quality factor for JPEG encoding (ignored for PNG). Bounds match the Python service's Pydantic constraint (ge=50, le=100).
  - `image` string — Base64-encoded input image. Required for image-to-image and edits.
  - `mask` string — Base64-encoded mask image. Required for edits/inpainting.
  - `strength` number — Image-to-image strength (0..1); higher values diverge further from the input image.

## Response `200`

Image(s) generated successfully

- ImageResponse
  - `images` string[], required — Generated images, base64-encoded
  - `generation_time` number — Wall-clock generation time in seconds (omitted when zero)

## Other responses

- `400` — Invalid request (e.g. missing required field)
- `401` — Missing or invalid authentication token
- `503` — Model not ready

---

[API](https://skmtc.net/tenstorrent/apis/tt-media-server-api-c-drogon.md) · [All operations](https://skmtc.net/tenstorrent/apis/tt-media-server-api-c-drogon/llms.txt) · [OpenAPI document](https://skmtc-service-staging.skmtc.workers.dev/v1/apis/tenstorrent/tt-media-server-api-c-drogon/revisions/f1176db3fd15/schema)