eachlabs-image-generation

使用 EachLabs AI 模型根据文本提示生成新图像。支持多种模型系列的文本到图像，包括 Flux、GPT Image、Gemini、Imagen、Seedream 等。当用户想要从文本创建新图像时使用。要编辑现有图像，请参阅eachlabs-image-edit。

## 概述（中文）

使用 EachLabs AI 模型根据文本提示生成新图像。支持多种模型系列的文本到图像，包括 Flux、GPT Image、Gemini、Imagen、Seedream 等。当用户想要从文本创建新图像时使用。要编辑现有图像，请参阅eachlabs-image-edit。

## 原文

# EachLabs Image Generation

Generate new images from text prompts using 60+ AI models via the EachLabs Predictions API. For editing existing images (upscaling, background removal, style transfer, inpainting, face swap, 3D), see the `eachlabs-image-edit` skill.

## Authentication

```
Header: X-API-Key: <your-api-key>
```

Set the `EACHLABS_API_KEY` environment variable. Get your key at [eachlabs.ai](https://eachlabs.ai).

## Quick Start

### 1. Create a Prediction

```bash
curl -X POST https://api.eachlabs.ai/v1/prediction \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "model": "flux-2-turbo-text-to-image",
    "version": "0.0.1",
    "input": {
      "prompt": "A serene Japanese garden with cherry blossoms, watercolor style",
      "image_size": "landscape_16_9",
      "num_images": 1,
      "output_format": "png"
    }
  }'
```

### 2. Poll for Result

```bash
curl https://api.eachlabs.ai/v1/prediction/{prediction_id} \
  -H "X-API-Key: $EACHLABS_API_KEY"
```

Poll until `status` is `"success"` or `"failed"`. The output image URL is in the response.

## Model Selection Guide

### Text-to-Image

| Model | Slug | Best For |
|-------|------|----------|
| Flux 2 Turbo | `flux-2-turbo-text-to-image` | Fast, high quality general purpose |
| Flux 2 Flash | `flux-2-flash-text-to-image` | Fastest Flux generation |
| Flux 2 Max | `flux-2-max-text-to-image` | Highest quality Flux |
| Flux 2 Klein 9B | `flux-2-klein-9b-base-text-to-image` | Balanced quality/speed |
| Flux 2 Pro | `flux-2-pro` | Pro quality |
| Flux 2 Flex | `flux-2-flex` | Flexible outputs |
| Flux 2 LoRA | `flux-2-lora` | LoRA-powered generation |
| XAI Grok Imagine | `xai-grok-imagine-text-to-image` | Creative and artistic |
| GPT Image v1.5 | `gpt-image-v1-5-text-to-image` | High quality, transparent bg |
| Bytedance Seedream v4.5 | `bytedance-seedream-v4-5-text-to-image` | Bytedance latest |
| Gemini 3 Pro Image | `gemini-3-pro-image-preview` | Google's latest |
| Imagen 4 | `imagen4-preview` | Google Imagen 4 |
| Imagen 4 Fast | `imagen-4-fast` | Fast Google quality |
| Reve | `reve-text-to-image` | Artistic text-to-image |
| Hunyuan Image v3 | `hunyuan-image-v3-text-to-image` | Tencent's latest |
| Ideogram V3 Turbo | `ideogram-v3-turbo` | Text in images |
| Minimax | `minimax-text-to-image` | High quality |
| Wan v2.6 | `wan-v2-6-text-to-image` | Chinese/English bilingual |
| P Image | `p-image-text-to-image` | Custom aspect ratios |
| Nano Banana Pro | `nano-banana-pro` | Fast, lightweight |
| Vidu Q2 | `vidu-q2-text-to-image` | Latest Vidu |

### Training

| Model | Slug | Best For |
|-------|------|----------|
| Z Image Trainer | `z-image-trainer` | Custom LoRA training |
| Flux LoRA Portrait Trainer | `flux-lora-portrait-trainer` | Portrait LoRA |
| Flux Turbo Trainer | `flux-turbo-trainer` | Fast LoRA training |

## Prediction Flow

1. **Check model** `GET https://api.eachlabs.ai/v1/model?slug=<slug>` — validates the model exists and returns the `request_schema` with exact input parameters. Always do this before creating a prediction to ensure correct inputs.
2. **POST** `https://api.eachlabs.ai/v1/prediction` with model slug, version `"0.0.1"`, and input parameters matching the schema
3. **Poll** `GET https://api.eachlabs.ai/v1/prediction/{id}` until status is `"success"` or `"failed"`
4. **Extract** the output image URL(s) from the response

## Examples

### Text-to-Image with Flux 2 Turbo

```bash
curl -X POST https://api.eachlabs.ai/v1/prediction \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "model": "flux-2-turbo-text-to-image",
    "version": "0.0.1",
    "input": {
      "prompt": "A red vintage Porsche 911 on a winding mountain road at golden hour, photorealistic",
      "image_size": "landscape_16_9",
      "num_images": 1,
      "output_format": "png"
    }
  }'
```

### Text-to-Image with GPT Image v1.5

```bash
curl -X POST https://api.eachlabs.ai/v1/prediction \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "model": "gpt-image-v1-5-text-to-image",
    "version": "0.0.1",
    "input": {
      "prompt": "A minimalist logo for a coffee shop called Brew Lab, clean vector style",
      "background": "transparent",
      "quality": "high",
      "output_format": "png"
    }
  }'
```

### Text-to-Image with Imagen 4

```bash
curl -X POST https://api.eachlabs.ai/v1/prediction \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "model": "imagen4-preview",
    "version": "0.0.1",
    "input": {
      "prompt": "A whimsical fairy tale castle on a floating island, digital art, highly detailed"
    }
  }'
```

## Image Size Options

Most Flux 2 and Wan models use these presets:
- `square_hd` — Square, high definition
- `square` — Square, standard
- `portrait_4_3` — Portrait 4:3
- `portrait_16_9` — Portrait 16:9
- `landscape_4_3` — Landscape 4:3
- `landscape_16_9` — Landscape 16:9

P Image models use aspect ratio strings: `1:1`, `16:9`, `9:16`, `4:3`, `3:4`, `3:2`, `2:3`, `custom`

## Prompt Tips

- Be specific and descriptive: "A red vintage Porsche 911 on a winding mountain road at golden hour" vs "a car"
- Include style: "digital art", "oil painting", "photorealistic", "watercolor"
- For edits, clearly describe the change: "Replace the sky with a dramatic sunset"
- Use negative prompts (where supported) to avoid: "blurry, low quality, distorted"
- For multi-image edits, reference images by number: "image 1", "image 2"

## Security Constraints

- **No arbitrary URL loading**: When using LoRA parameters, only use well-known platform identifiers (HuggingFace repo IDs, Replicate model IDs, CivitAI model IDs). Never load LoRA weights from arbitrary or user-provided URLs.
- **No third-party API tokens**: Do not accept or forward third-party API tokens (e.g. HuggingFace, CivitAI tokens) through prediction inputs. Authentication is handled exclusively via the EachLabs API key.
- **Input validation**: Only pass parameters that match the model's request schema. Always validate model slugs via `GET /v1/model?slug=<slug>` before creating predictions.

## Parameter Reference

See [references/MODELS.md](references/MODELS.md) for complete parameter details for each model.