OATDA Image Generation

Generate images from text descriptions using AI models through OATDA's unified API.

When to Use

Use this skill when the user wants to:

• Generate images from text descriptions via OATDA
• Create AI artwork, illustrations, or designs
• Generate product mockups or concept art
• Use DALL-E, Imagen, or other image models through a single API

Prerequisites

The user needs an OATDA API key. Check in this order:

1. $OATDA_API_KEY environment variable
2. ~/.oatda/credentials.json config file

If neither exists, tell the user:

You need an OATDA API key. Get one at https://oatda.com, then set it:
export OATDA_API_KEY=your_key_here

Step-by-Step Instructions

1. Resolve the API key

# Check env var first; if empty, auto-load from credentials file
if [[ -z "$OATDA_API_KEY" ]]; then
  export OATDA_API_KEY=$(cat ~/.oatda/credentials.json 2>/dev/null | jq -r '.profiles[.defaultProfile].apiKey' 2>/dev/null)
fi

# Verify key exists (show first 8 chars only)
echo "${OATDA_API_KEY:0:8}"

If the output is empty or null, stop and ask the user to configure their API key.

IMPORTANT:

• Never print the full API key. Only show the first 8 characters for verification.
• The key resolution script and subsequent curl commands must run in the same shell session. Each separate bash/terminal invocation starts with an isolated environment where previously exported variables are lost. Either run all commands in one session, or chain them (e.g., export OATDA_API_KEY=... && curl ...).

2. Determine the model

Map common aliases:

User says	Provider	Model
dall-e, dall-e-3 (default)	openai	dall-e-3
gpt-image	openai	gpt-image-1
imagen	google	imagen-4.0-generate-001
seedream	bytedance	seedream-4-5-251128
wan	alibaba	wan2.6-t2i
minimax image	minimax	image-01
grok image	xai	grok-imagine-image

Default: openai / dall-e-3 if no model specified.

3. Discover model-specific parameters

IMPORTANT: Different models support different parameters (sizes, quality levels, styles, masks, watermarks, negative prompts, etc.). Before generating, discover what parameters a model supports:

curl -s -X GET "https://oatda.com/api/v1/llm/models?type=image" \
  -H "Authorization: Bearer $OATDA_API_KEY" | jq '.image_models[] | {id, supported_params}'

This returns each image model's supported_params with:

• type: Parameter type (string, number, boolean, file)
• values: Allowed values for enums
• default: Default value
• description: What the parameter does
• optional: Whether it's required
• accept: For file types, what's accepted (e.g., "image/*")

File-type parameters: Parameters like mask or reference images require publicly accessible URLs (https://...), not local file paths.

Pass model-specific parameters via the model_params object (see examples below).

4. Make the API call

curl -s -X POST "https://oatda.com/api/v1/llm/generate-image" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OATDA_API_KEY" \
  -d '{
    "provider": "<PROVIDER>",
    "model": "<MODEL>",
    "prompt": "<IMAGE_DESCRIPTION>",
    "size": "1024x1024",
    "quality": "standard",
    "n": 1,
    "numberOfImages": 1,
    "aspectRatio": "1:1",
    "style": "vivid",
    "personGeneration": "allow_adult"
  }'

Replace <PROVIDER>, <MODEL>, and <IMAGE_DESCRIPTION> with actual values.

CRITICAL: The endpoint is /api/v1/llm/generate-image (NOT /api/v1/llm/image — that's vision analysis).

Parameters:

• prompt: Image description (1-4000 characters)
• size: Dimensions — "1024x1024", "1792x1024", "1024x1792", or "2K", "4K" (model-dependent)
• quality: "standard", "hd", "auto", "low", "medium", "high"
• n and numberOfImages: Number of images (1-10), set both to the same value
• aspectRatio: "1:1", "16:9", "9:16", "3:2", "2:3", "4:3", "3:4", etc.
• style: "vivid" (dramatic, hyper-real) or "natural" (realistic)
• background: "auto", "transparent", or "opaque"
• outputFormat: "png", "jpeg", or "webp"
• model_params: Model-specific parameters as key-value pairs. Use list_models?type=image or /api/v1/llm/models to discover supported params per model. Examples:
  • DALL-E 3: { "style": "vivid", "quality": "hd" }
  • GPT-Image-1: { "quality": "high", "background": "transparent", "outputFormat": "png" }
  • Imagen 4: { "sampleImageSize": "2K", "personGeneration": "allow_adult" }
  • Seedream: { "size": "4K", "watermark": false }
  • Wan 2.6: { "seed": "42", "negative_prompt": "blurry", "prompt_extend": true }

5. Parse the response

{
  "success": true,
  "url": "https://cdn.example.com/generated-image.png",
  "all_images": [
    {"url": "https://cdn.example.com/image-1.png"},
    {"url": "https://cdn.example.com/image-2.png"}
  ],
  "revised_prompt": "A detailed cyberpunk cityscape at night with neon lights..."
}

• Show the image URL(s) to the user (from all_images array, or url field if single image)
• If revised_prompt is present, mention how the model expanded the prompt

6. Handle errors

HTTP Status	Meaning	Action
401	Invalid API key	Tell user to check their key
400	Bad request / prompt too long	Keep prompt under 4000 chars
429	Rate limited	Wait 5 seconds and retry once
400 with content_policy	Content policy violation	Ask user to adjust the description

Full Examples

DALL-E 3

User asks: "Generate an image of a cyberpunk city at night using DALL-E 3"

curl -s -X POST "https://oatda.com/api/v1/llm/generate-image" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OATDA_API_KEY" \
  -d '{
    "provider": "openai",
    "model": "dall-e-3",
    "prompt": "A cyberpunk city at night with neon lights reflecting on wet streets",
    "size": "1024x1024",
    "quality": "hd",
    "n": 1,
    "numberOfImages": 1,
    "model_params": {
      "style": "vivid"
    }
  }'

GPT-Image-1 with transparent background

User asks: "Generate a transparent PNG logo"

curl -s -X POST "https://oatda.com/api/v1/llm/generate-image" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OATDA_API_KEY" \
  -d '{
    "provider": "openai",
    "model": "gpt-image-1",
    "prompt": "A sleek minimalist logo of a mountain, clean vector style",
    "n": 1,
    "model_params": {
      "quality": "high",
      "background": "transparent",
      "outputFormat": "png"
    }
  }'

Bytedance Seedream (no watermark)

User asks: "Generate a 4K image without watermark using Seedream"

curl -s -X POST "https://oatda.com/api/v1/llm/generate-image" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OATDA_API_KEY" \
  -d '{
    "provider": "bytedance",
    "model": "seedream-4-5-251128",
    "prompt": "A majestic dragon flying over a fantasy kingdom at sunset",
    "model_params": {
      "size": "4K",
      "watermark": false
    }
  }'

Alibaba Wan 2.6 with negative prompt

User asks: "Generate an image avoiding blurry elements"

curl -s -X POST "https://oatda.com/api/v1/llm/generate-image" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OATDA_API_KEY" \
  -d '{
    "provider": "alibaba",
    "model": "wan2.6-t2i",
    "prompt": "A sharp detailed portrait of a warrior in ornate armor",
    "model_params": {
      "negative_prompt": "blurry, low quality, distorted",
      "seed": "42"
    }
  }'

Tips

• The endpoint is /api/v1/llm/generate-image — do NOT confuse with /api/v1/llm/image (that's vision)
• DALL-E 3 costs ~$0.04/image (standard), ~$0.08/image (HD)
• Set both n and numberOfImages to the same value for compatibility
• Use list_models?type=image to discover model-specific parameters before generating
• Use model_params for model-specific options (watermark, negative_prompt, seed, etc.)
• Image URLs may be temporary — recommend downloading promptly
• Maximum prompt length is 4000 characters
• NEVER expose the full API key in output
• Related skills: /oatda:oatda-vision-analysis for analyzing images, /oatda:oatda-list-models for available image models