Image Generation Guide

Prepare in the dashboard

Sign in to dashboard.koboillm.com with your email and password, then select your team at the top.
Open API Keys → Create API key. Enter a Key name; the key budget is optional. The defaults are No expiry and All Team Models. Select Create and save the full key, which is shown once.
Copy an accessible model name from Model List. Choose a model that supports this guide's endpoint; API Quickstart also provides request examples.
After running your application, open Logs in the same team, select a time range, and use View details on a request to check its key, tokens, and recorded spend.

OpenAI-compatible base URL: https://api.koboillm.com/v1. Follow the client-specific URL format below. The dashboard address is for your account; applications use the API key, not your account password.

Show the current key creation form

Actual dashboard interface with fictional demonstration data. Click to enlarge.

Model names and provider prices in examples are reference values. Availability, endpoint support, and final recorded costs depend on your key and request.

The image generation API at KoboiLLM adheres to the OpenAI Images API specification, enabling seamless integration with any OpenAI or LiteLLM SDK. This guide explains endpoint calls, available parameters, multi-language code snippets, response schemas, and cost estimates across all supported image models (OpenAI gpt-image-*, Google Gemini, and Google Imagen).

Endpoints & Authentication
Model and Price Reference
Request Parameters
Generating Images from Text
Editing Existing Images
Verify recorded charges

1. Endpoints & Authentication

Base URL:

https://api.koboillm.com/v1

Append the endpoint path to the API origin, or configure this base URL in your SDK.

Primary Endpoints:

Method	Path	Function
`POST`	`/v1/images/generations`	Generate new images from text prompts
`POST`	`/v1/images/edits`	Edit / modify existing images

Authentication requires the header Authorization: Bearer sk-xxxxxx. Store your API key securely in an environment variable rather than hardcoding it into source files.

export LITELLM_API_KEY="sk-xxxxxx"

2. Model and Price Reference

For current examples, choose an active model such as gpt-image-2 from Model List or API Quickstart. Other names below are historical examples and may not be available.

Availability: Imagen deployments are currently blocked and hidden from the new dashboard. Imagen examples below are reference material, not a currently usable option. Select an active image model from your account catalog.

⚠️ Pricing Disclaimer

Rates listed in this document are sourced from official vendor pricing pages (OpenAI and Google Cloud Vertex AI). Vendors may update rate cards at any time. Always verify live account billing on your dashboard at dashboard.koboillm.com.

2.1 — OpenAI Models

Model	Endpoint Name	Billing Mode	Best Used For
`gpt-image-1-mini`	`gpt-image-1-mini`	per-token	Thumbnails, high-volume generation, prototypes — lowest cost
`gpt-image-1.5`	`openai/gpt-image-1.5`	per-token	General-purpose, balanced quality & price
`gpt-image-2`	`openai/gpt-image-2`	per-token	Latest flagship, top prompt adherence, complex composition
`gpt-image-1`	`openai/gpt-image-1`	per-token	Legacy version, same rate tier as gpt-image-1-mini
`chatgpt-image-latest`	`openai/chatgpt-image-latest`	per-token	”ChatGPT-style” reasoning image model — highest tier, ~3× gpt-image-1.5
`dall-e-3`	`openai/dall-e-3`	flat per-image	Legacy DALL·E 3 model
`dall-e-2`	`dall-e-2`	flat per-image	Legacy DALL·E 2 model

2.2 — Google Gemini Models (Token-based)

Model	Endpoint Name	Best Used For
`gemini-2.5-flash-image`	`gemini/gemini-2.5-flash-image` or `vertex_ai/gemini-2.5-flash-image`	Fast & affordable multimodal image generation — recommended Gemini
`gemini-3.1-flash-image`	`gemini/gemini-3.1-flash-image` or `vertex_ai/gemini-3.1-flash-image`	Nano Banana 2 image generation
`gemini-3-pro-image`	`gemini/gemini-3-pro-image` or `vertex_ai/gemini-3-pro-image`	Nano Banana Pro, highest fidelity in Gemini family

2.3 — Google Imagen Models (Flat-fee per image)

Model	Endpoint Name	Flat Rate per Image	Best Used For
`imagen-3.0-fast-generate-001`	`vertex_ai/imagen-3.0-fast-generate-001`	$0.02	Imagen 3.0 speed-optimized
`imagen-4.0-fast-generate-001`	`vertex_ai/imagen-4.0-fast-generate-001`	$0.02	Imagen 4.0 speed-optimized
`imagen-3.0-generate-001`	`vertex_ai/imagen-3.0-generate-001`	$0.04	Imagen 3.0 standard
`imagen-4.0-generate-001`	`vertex_ai/imagen-4.0-generate-001`	$0.04	Imagen 4.0 standard
`imagen-4.0-ultra-generate-001`	`vertex_ai/imagen-4.0-ultra-generate-001`	$0.06	Imagen 4.0 ultra — highest visual quality

3. Request Parameters

Gemini 3.1 Flash Image and Flash-Lite Image ignore quality; it does not select a higher resolution. Omit it for these models and use supported native image options if needed. Other unsupported parameters can still be rejected.

Parameter	Type	Allowed Values	Effect
`model`	string	any model ID	Determines pricing tier and backend route
`prompt`	string	free text	Tokenized as `text_input`
`size`	string	`1024x1024`, `1024x1536`, `1536x1024`, `auto`	Resolution. Influences output token count (and total cost).
`quality`	string	`low`, `medium`, `high`, `auto`	Influences output token count. High quality uses ~15× more tokens than low (OpenAI only).
`n`	integer	1–10	Image count. Cost scales linearly with `n`.
`response_format`	string	Omit or use `b64_json`	GPT Image 2 and Gemini image models return `b64_json` even when a client sends `url`; no hosted image URL is created.

4. Generate Images from Text

4.1 — cURL Example

curl -X POST https://api.koboillm.com/v1/images/generations \
  -H "Authorization: Bearer sk-xxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "Majestic dragon soaring over mist-covered mountains at sunset",
    "size": "1024x1024",
    "quality": "high",
    "n": 1
  }'

4.2 — Python (OpenAI SDK Example)

from openai import OpenAI

client = OpenAI(
    base_url="https://api.koboillm.com/v1",
    api_key="sk-xxxxxx",
)

response = client.images.generate(
    model="gpt-image-2",
    prompt="Majestic dragon soaring over mist-covered mountains at sunset",
    size="1024x1024",
    quality="high",
    n=1,
)

import base64
from pathlib import Path

Path("generated.png").write_bytes(base64.b64decode(response.data[0].b64_json))

5. Editing Existing Images

Use /v1/images/edits for image-to-image workflows, inpainting, and variations.

curl -X POST https://api.koboillm.com/v1/images/edits \
  -H "Authorization: Bearer sk-xxxxxx" \
  -F "model=gpt-image-2" \
  -F "image=@input.png" \
  -F "prompt=Change the sky color to deep purple sunset" \
  -F "size=1024x1024" \
  -F "quality=high"

Verify recorded charges

Sign in with your account email and password at Logs, select your team, filter by model/time, and choose View details. It shows Request ID, input/output tokens, and spend. The dashboard does not expose raw provider usage objects or prompt/response content. Image, cache, and tier rates can require more than the displayed token totals; provider estimates are not your final charge. Contact support with the Request ID for a discrepancy. A provider response ID is not necessarily the same ID shown in Logs.

Request details with request identifier, key, and recorded cost — Actual dashboard interface with fictional demonstration data. Click to enlarge.