> ## Documentation Index > Fetch the complete documentation index at: https://bunnynet-cb9733c2-claude-slack-session-savj4.mintlify.site/llms.txt > Use this file to discover all available pages before exploring further. # Reference > Complete reference for AI Image Generation engines, parameters, and blueprints ## URL structure Generate images by constructing URLs with the following structure: ```http theme={null} /.ai/img////.jpg ``` **Example:** ```http theme={null} https://your-pull-zone.b-cdn.net/.ai/img/dalle-256/default/12345/rabbit.jpg ``` ### Engine name Specifies which AI model and resolution to use. See [Supported engines](#supported-engines) for available options. ### Blueprint Predefined prompt templates that add consistent styling. Use `default` for no additional styling, or create [custom blueprints](#blueprints) to apply specific visual styles. ### Seed A numeric value that ensures consistent results. The same seed with the same prompt always generates the same image. **Example:** `12345`, `7890`, `1` ### Prompt The text description of what you want to generate. Hyphens are automatically converted to spaces, and file extensions (`.jpg`, `.jpeg`, `.png`) are removed. **Example:** ```text theme={null} bunny-eating-carrots.jpg → "bunny eating carrots" ``` This creates cleaner, SEO-friendly URLs that are easier to share. *** ## Supported engines ### DALL-E 2 Created by OpenAI, DALL-E 2 excels at creative and abstract interpretations with descriptive inputs. | Code | Resolution | Price/Image | | :----------- | :--------- | :---------- | | `dalle-256` | 256x256 | \$0.016 | | `dalle-512` | 512x512 | \$0.018 | | `dalle-1024` | 1024x1024 | \$0.020 | **Best for:** * Concept art and illustrations * Pixel art and retro graphics * Abstract prompts and creative compositions * 3D renders and stylized artwork **Not ideal for:** * Photo-realistic images **Example:** ```http theme={null} /.ai/img/dalle-512/default/100/a-happy-robot-with-flowers-growing-out-of-his-head-clouds-in-the-background-digital-art.jpg ``` ### Stable Diffusion Open-source model that uses keyword-focused prompting and excels at photo-realistic generation. | Code | Resolution | Price/Image | | :--------- | :--------- | :---------- | | `sd15-512` | 512x512 | \$0.001 | | `sd21-512` | 512x512 | \$0.001 | | `sd21-768` | 768x768 | \$0.030 | **Best for:** * Photo-realistic images * Landscape and architectural photography * Traditional and digital art * Portrait and scene generation **Not ideal for:** * Pixel art * Highly abstract concept art * Faces and hands (especially v2.1) **Example:** ```http theme={null} /.ai/img/sd21-512/default/200/rabbit-sitting-in-front-of-a-christmas-tree.jpg ``` Stable Diffusion v2.1 is trained on an open-source dataset and may produce different results compared to v1.5, particularly for faces and hands. However, it excels at landscapes and architectural scenes. *** ## Blueprints Blueprints allow you to apply consistent styling across generated images by prepending or appending phrases to prompts. This simplifies URLs and maintains visual consistency across your application. ### Creating a blueprint Configure blueprints in **Optimizer > Settings > Bunny AI**: 1. **Name:** The identifier used in the URL (e.g., `avatar`) 2. **Prefix:** Text added before the prompt 3. **Suffix:** Text added after the prompt **Example configuration:** | Field | Value | | :----- | :--------------------------------- | | Name | `avatar` | | Prefix | `cute pixel art of a` | | Suffix | `with a colorful solid background` | ### Using blueprints With the above blueprint, this URL: ```http theme={null} https://your-pull-zone.b-cdn.net/.ai/img/dalle-256/avatar/500/bunny.jpg ``` Generates the prompt: ```text theme={null} cute pixel art of a bunny with a colorful solid background ``` This creates a powerful system for generating unique user avatars, maintaining brand consistency, or applying complex styling across your site. When generating pixel art, use DALL-E 2 engines for optimal results. *** ## Advanced parameters Configure additional generation parameters through blueprint settings. Some parameters are engine-specific: | Parameter | Description | Type | Engines | | :----------------- | :---------------------------------------------------------------------------------------------------------------------- | :--------------- | :--------------- | | **Steps** | Number of generation steps (default: 90). Higher values may improve quality but increase processing time. | numeric (10-150) | Stable Diffusion | | **Cfg** | Prompt adherence level (default: 7). Lower values give more creative freedom, higher values stick closer to the prompt. | numeric (1-15) | Stable Diffusion | | **NegativePrompt** | Replaces the default negative prompt list. Specifies elements to avoid in the generated image. | string | Stable Diffusion | ### Negative prompts Stable Diffusion supports negative prompts to exclude unwanted elements from generated images. Bunny AI uses a default set optimized for quality: ```text theme={null} deformed, blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limb, ugly, poorly drawn hands, missing limb, blurry, floating limbs, disconnected limbs, malformed hands, blur, out of focus, long neck, long body, pixelated, ((((mutated hands and fingers)))), (((out of frame))) ``` These defaults are crucial for producing good results with Stable Diffusion v2.1. If you override them with a custom blueprint, test thoroughly to ensure quality output. Override defaults by setting the `NegativePrompt` parameter in your blueprint configuration. *** ## Prompt best practices ### General guidelines Effective prompts typically include: * **Descriptive nouns:** Subject matter (e.g., "bunny", "robot", "landscape") * **Style descriptors:** Visual style (e.g., "digital art", "oil painting", "photorealistic") * **Colors and mood:** Atmosphere and palette (e.g., "vibrant colors", "moody lighting") * **Quality modifiers:** Fidelity hints (e.g., "4k", "high quality", "detailed") ### Style suggestions Try these style keywords for different visual effects: * `digital art` * `oil painting of` * `traditional art` * `abstract art of` * `watercolor painting` * `in style of Picasso` * `in steampunk style` * `in style of Murakami` * `art deco style` * `high quality` * `photorealistic` * `4k` * `detailed texture` * `sharp focus` ### Model-specific tips **For DALL-E 2:** Use full descriptive sentences with creative, abstract concepts: ```text theme={null} a happy robot with flowers growing out of his head, clouds in the background, digital art ``` **For Stable Diffusion:** Use keyword-focused prompts for photo-realistic results: ```text theme={null} rabbit sitting in front of a christmas tree, photorealistic, high quality ``` *** ## Combining with Dynamic Images API AI-generated images support all [Dynamic Images transformations](/optimizer/dynamic-images/overview). Apply resizing, cropping, filters, and format conversions to AI-generated images: ```http theme={null} /.ai/img/dalle-512/default/100/mountain-landscape.jpg?width=800&sharpen=true&quality=85 ``` This allows you to generate base images with AI and apply responsive sizing or branding overlays dynamically. *** ## Security and token authentication All AI generation paths are automatically protected with [Token Authentication](/docs/cdn-token-authentication-basic) to prevent: * Unauthorized image generation * Abuse and excessive costs * Resource exhaustion attacks The bunny.net dashboard automatically generates tokens for preview and testing. For dynamic URL generation in your application, implement token generation according to the [Token Authentication documentation](/docs/cdn-token-authentication-basic). *** ## Performance and caching **First request:** 1. Bunny AI processes the generation using the specified engine 2. The result is cached at the edge for 3 months 3. Response time varies by engine and complexity (typically 5-30 seconds) **Subsequent requests:** * Served instantly from the nearest edge location * Same performance as standard CDN-cached images * Cache purging works the same as regular images Pre-generate common images during deployment or user onboarding to ensure instant delivery when users access them.