Skip to main content
Generate and edit images through the chat completions endpoint using LLMs that natively support image output.

​
Overview

Some LLMs can produce images directly as part of a chat response — no separate endpoint required. Instead of calling a dedicated image API, you send a normal chat completion request and the model returns a base64-encoded image in the assistant message. The main image-capable LLM available through Eden AI is Google’s Gemini Flash Image family:
ModelModel stringProvider
Gemini 2.5 / 3.1 Flash Imagegoogle/gemini-2.5-flash-image, google/gemini-3.1-flash-image-previewGoogle
It supports:
  • Text-to-image — generate an image from a prompt
  • Image editing — modify a provided input photo
The endpoint stays the same: 
POST /v3/chat/completions
Image-capable LLMs are different from Expert Model image generation — which wraps dedicated image APIs like DALL-E or Stable Diffusion. Use LLM image generation when you want the model to reason about the prompt and output an image in the same turn.

​
Google Gemini Image Models

Gemini image models accept a standard text prompt plus an optional image_config block that controls the output resolution and aspect ratio.

​
Text-to-image

import requests

url = "https://api.edenai.run/v3/chat/completions"
headers = {
    "Authorization": "Bearer YOUR_API_KEY",
    "Content-Type": "application/json",
}

payload = {
    "model": "google/gemini-3.1-flash-image-preview",
    "messages": [
        {"role": "user", "content": "A nano banana dish in a fancy restaurant, cinematic lighting"}
    ],
    "image_config": {
        "aspect_ratio": "16:9",
        "image_size": "2K"
    }
}

response = requests.post(url, headers=headers, json=payload)
result = response.json()
print(result["choices"][0]["message"]["content"])

​
Supported image_config values

FieldDescriptionAllowed values
aspect_ratioOutput shape1:1, 16:9, 9:16, 4:3, 3:4, 21:9, 2:3, 3:2, 4:5, 5:4, 1:4, 4:1, 1:8, 8:1
image_sizeOutput resolution512 (Gemini 3.1 Flash only), 1K, 2K, 4K
Allowed values vary slightly per model — check the Gemini image generation docs for per-model support.

​
Image editing (with input photo)

Pass an existing image to the model alongside your edit instruction. Use the standard OpenAI multimodal content format: 
import base64
import requests

url = "https://api.edenai.run/v3/chat/completions"
headers = {
    "Authorization": "Bearer YOUR_API_KEY",
    "Content-Type": "application/json",
}

# Encode a local image as base64
with open("input.jpg", "rb") as f:
    image_b64 = base64.b64encode(f.read()).decode("utf-8")

payload = {
    "model": "google/gemini-3.1-flash-image-preview",
    "messages": [
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "Turn this photo into a watercolor painting, keep the composition"},
                {
                    "type": "image_url",
                    "image_url": {"url": f"data:image/jpeg;base64,{image_b64}"}
                }
            ]
        }
    ],
    "image_config": {
        "aspect_ratio": "1:1",
        "image_size": "2K"
    }
}

response = requests.post(url, headers=headers, json=payload)
result = response.json()
print(result["choices"][0]["message"]["content"])
You can also pass a remote URL directly instead of base64: 
{
  "type": "image_url",
  "image_url": {"url": "https://example.com/photo.jpg"}
}

​
Response Format

The generated image is returned as a base64-encoded data URL inside the assistant message content. 
{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "model": "google/gemini-3.1-flash-image-preview",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUg..."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 14,
    "completion_tokens": 512,
    "total_tokens": 526
  },
  "cost": 0.042
}

​
Saving the image to disk

import base64
import re

content = result["choices"][0]["message"]["content"]
# Strip the "data:image/...;base64," prefix if present
match = re.match(r"data:image/\w+;base64,(.*)", content)
image_b64 = match.group(1) if match else content

with open("output.png", "wb") as f:
    f.write(base64.b64decode(image_b64))

​
Next Steps

Expert Image Generation

Use dedicated image APIs like DALL-E or Stable Diffusion

Chat Completions

Standard text chat completions reference