pixmind-image（像素思维图像）

Pixmind Image Generation Skill

Generate AI images using Pixmind. Supports text-to-image and image-to-image generation with multiple models.

Note: The API endpoint aihub-admin.aimix.pro is the official Pixmind API gateway. Result URLs on chatmix.top are Pixmind's CDN for generated content.

When to use

• User asks to generate, create, or draw an image
• User wants to transform or edit an existing image
• User requests image variations or upscaling

Prerequisites

1. Register at pixmind.io — 注册即送 200 积分免费试用
2. Create an API key at pixmind.io/api-keys
3. Set env PIXMIND_API_KEY with your key

API Details

Endpoint: POST https://aihub-admin.aimix.pro/open-api/v1/image/generate Auth: Header X-API-Key: {API_KEY} (from env PIXMIND_API_KEY)

Request Body (JSON)

Available Models

Model Introductions

Seedream (ByteDance) — Flagship series with cinematic aesthetics and realistic textures. Unified architecture for both generation and editing. Native 4K support (4096x4096).

• seedream-5.0 — Latest generation, 2K/3K output, supports seed for reproducibility.
• seedream-4.5 — Previous flagship, LM Arena Top 10. 2K/4K output.
• seedream-4.0 — Proven high quality model, good balance of quality and speed. Default model. 1K/2K/4K.

Midjourney — Industry-leading artistic and creative image generation.

• mj-v7 — Latest version, top aesthetic quality. Supports speed control (relax/fast/turbo), stylization, weirdness, and variety parameters for fine-grained creative control.

OpenAI GPT Image — Strong instruction following, excellent text rendering in images.

• gpt-image-4o — Latest OpenAI model, supports medium/high quality.
• gpt-image-1.5 — Earlier generation, still solid quality.

Alibaba / Qwen / Wan

• qwen-image-2.0-pro / qwen-image-2.0 — Latest Qwen VL image generation. Text-to-image only, up to 6 samples per request.
• qwen-image-max / qwen-image-plus — Previous generation Qwen models, text-to-image only.
• qwen-image-edit-max / qwen-image-edit-plus — Specialized for image editing (img2img only).
• wan2.7-image-pro / wan2.7-image — Wan 2.7 series, Alibaba's latest diffusion model with editing support.
• wan2.6-image — Wan 2.6, high-end generation with editing support.
• z-image — Alibaba Tongyi Lab's open-source 6B-parameter efficient diffusion transformer (S3-DiT). Photorealistic quality with low computational cost.

Flux (Black Forest Labs) — Strong prompt adherence and visual fidelity.

• flux-kontext-pro — Balanced quality and speed.

Nano Banana — Lightweight models with broad aspect ratio support.

• nano-banana-2 — Latest generation, supports seed & negative prompt, 1K/2K/4K.
• nano-banana-2-eco — 80% cost reduction, great for bulk generation.
• nano-banana-pro — Higher quality variant with seed & negative prompt support.
• nano-banana-pro-lite — Lighter pro version, 70% off.
• nano-banana — Base model, basic Gemini-powered generation.

Pixmind Native

• pixmind-2.0 — Pixmind's proprietary model, supports relax/fast speed modes.

Usage

Use curl or the included helper script:

# Text to image (via curl)
curl -X POST https://aihub-admin.aimix.pro/open-api/v1/image/generate \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $PIXMIND_API_KEY" \
  -d '{"prompt": "描述文字", "model": "nano-banana-2", "aspectRatio": "16:9"}'

# Or use the helper script
node {baseDir}/image-generate.js --prompt "描述文字" --model nano-banana-2 --aspect-ratio 16:9

Task Status Polling

After generation, poll for results:

# Via curl
curl https://aihub-admin.aimix.pro/open-api/v1/task/<TASK_ID> \
  -H "X-API-Key: $PIXMIND_API_KEY"

# Or use the helper script
node {baseDir}/task-status.js --task-id <TASK_ID> --poll

Response Format

Generate response:

{"code": 1000, "data": {"taskId": 19399, "status": "processing"}}

Task status response:

{
  "code": 1000,
  "data": {
    "taskId": 19399,
    "status": "ready",
    "progress": 100,
    "images": ["https://chatmix.top/..."]
  }
}

• data.taskId — Use this to poll status
• Status values: processing → ready (success)
• On success: data.images contains generated image URLs

Guidelines

1. Always confirm the prompt with the user before generating
2. Default to nano-banana-2 (high quality) or seedream-4.0 unless user specifies otherwise
3. Use 1:1 aspect ratio by default, suggest alternatives when appropriate
4. If user provides a reference image, use img2img mode automatically
5. After getting the task ID, poll until completion and return image URLs
6. For image editing tasks, prefer models that support image2image: nano-banana-2, seedream-5.0, gpt-image-4o, mj-v7, pixmind-2.0
7. For Midjourney V7, you can also use stylization, weirdness, and variety parameters
8. Check the model's supported aspect ratios before sending the request — not all models support the same ratios