Image generation

Generate images from text with POST /v1/images/generations, and edit images with POST /v1/images/edits — both OpenAI-compatible in shape.

Available image models (pass the ID in model):

Model ID	Notes
flux2-klein	Fast, unified generation and editing
qwen-image	Strong text rendering and precise editing
z-image-turbo	Distilled turbo model — very few steps, text-to-image

Basic request

curl https://api.ecohash.com/v1/images/generations \
  -H "Authorization: Bearer eco_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "flux2-klein",
    "prompt": "a watercolor painting of a fox in a misty forest, soft light",
    "size": "1024x1024",
    "n": 1,
    "response_format": "b64_json"
  }'

Response:

{
  "created": 1776391234,
  "data": [{
    "b64_json": "iVBORw0KGgo...",
    "revised_prompt": null
  }]
}

Decode the base64 to get the PNG bytes.

Parameters

Parameter	Type	Default	Notes
model	string	—	Required. One of the image model IDs above
prompt	string	—	Required. Max ~1000 characters
size	string	"1024x1024"	Common sizes: 512x512, 768x768, 1024x1024, 1024x768, 768x1024
n	integer	1	Number of images. Currently only n=1 is supported
response_format	string	"b64_json"	"b64_json" returns inline base64; "url" returns a hosted URL (24h expiry)
steps	integer	model default	Denoising steps. If omitted, EcoLink applies a sensible per-model default — turbo/distilled models use a low step count automatically, so you don't need to set it
seed	integer	random	Deterministic generation when set

Default steps

You normally don't need to pass steps. Turbo models (like z-image-turbo) are designed for a small number of steps; EcoLink injects that model's intended default when the request omits steps, so you get the fast, intended result rather than an unnecessarily slow one. Set steps only if you specifically want more or fewer.

Editing images

Where a model supports editing (for example flux2-klein and qwen-image), send the source image plus a prompt to POST /v1/images/edits as multipart/form-data:

curl https://api.ecohash.com/v1/images/edits \
  -H "Authorization: Bearer eco_YOUR_KEY" \
  -F model=flux2-klein \
  -F image=@input.png \
  -F prompt="make the sky a dramatic sunset" \
  -F size=1024x1024

Field	Type	Required	Notes
model	text	yes	An edit-capable image model
image	file	yes	The source image to edit
prompt	text	yes	What to change
size	text	no	Output size

A text-to-image-only model returns an error on this endpoint — use an edit-capable model.

Python example

from openai import OpenAI
import base64

client = OpenAI(api_key="eco_...", base_url="https://api.ecohash.com/v1")

resp = client.images.generate(
    model="flux2-klein",
    prompt="a cyberpunk cityscape at sunset, neon reflections",
    size="1024x1024",
    response_format="b64_json",
)

image_bytes = base64.b64decode(resp.data[0].b64_json)
with open("out.png", "wb") as f:
    f.write(image_bytes)

Tips

• Describe subject, style, lighting, mood. "a golden retriever wearing a top hat, oil painting, Rembrandt lighting" beats "a dog in a hat."
• Pick the right model. Turbo models are fastest; qwen-image is strong when the image contains text.
• Use seed for reproducible output when testing.
• Orientation — use 1024x768 for landscapes, 768x1024 for portraits.

Billing

Image generation and edits bill per image at the model's rate. See Balance & transactions.

Try it in the playground

See Playground to iterate on prompts visually before writing code.