Hyperbolic Chat To Image

arazzo: 1.0.1
info:
  title: Hyperbolic Chat To Image
  summary: Use an LLM to craft a vivid image prompt, then render it with a diffusion model.
  description: >-
    A creative two-stage pipeline that chains Hyperbolic's chat completions and
    image generation endpoints. A language model expands a short idea into a
    detailed, diffusion-ready prompt, and that prompt is fed directly into a
    Stable Diffusion or FLUX model to render base64 images. Every step inlines
    its request — including the inline Authorization Bearer credential — so the
    flow reads and runs without opening the underlying OpenAPI descriptions.
  version: 1.0.0
sourceDescriptions:
- name: chatCompletionsApi
  url: ../openapi/hyperbolic-chat-completions-api-openapi.yml
  type: openapi
- name: imageGenerationApi
  url: ../openapi/hyperbolic-image-generation-api-openapi.yml
  type: openapi
workflows:
- workflowId: chat-to-image
  summary: Expand an idea into an image prompt with an LLM and render it to an image.
  description: >-
    Sends a brief creative idea to a chat model with instructions to return a
    rich, single-paragraph image prompt, then passes the returned text to the
    image generation endpoint to produce base64-encoded images.
  inputs:
    type: object
    required:
    - apiKey
    - idea
    - chatModel
    - imageModel
    properties:
      apiKey:
        type: string
        description: Hyperbolic API key passed as a Bearer token.
      idea:
        type: string
        description: A short creative idea to expand into a detailed image prompt.
      chatModel:
        type: string
        description: Chat model id used to author the image prompt.
      imageModel:
        type: string
        description: Image model name (e.g. SDXL1.0-base or SD3.5).
      steps:
        type: integer
        description: Number of diffusion steps.
      width:
        type: integer
        description: Output image width in pixels.
      height:
        type: integer
        description: Output image height in pixels.
  steps:
  - stepId: authorPrompt
    description: >-
      Ask the chat model to turn the short idea into a detailed, single
      paragraph text-to-image prompt suitable for a diffusion model.
    operationId: createChatCompletion
    parameters:
    - name: Authorization
      in: header
      value: "Bearer $inputs.apiKey"
    requestBody:
      contentType: application/json
      payload:
        model: $inputs.chatModel
        messages:
        - role: system
          content: >-
            You write vivid, single-paragraph text-to-image prompts. Reply with
            only the prompt and no preamble.
        - role: user
          content: $inputs.idea
        max_tokens: 300
        temperature: 0.8
        stream: false
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      imagePrompt: $response.body#/choices/0/message/content
  - stepId: render
    description: >-
      Generate images from the authored prompt and capture the base64 image and
      the seed used.
    operationId: generateImage
    parameters:
    - name: Authorization
      in: header
      value: "Bearer $inputs.apiKey"
    requestBody:
      contentType: application/json
      payload:
        model_name: $inputs.imageModel
        prompt: $steps.authorPrompt.outputs.imagePrompt
        steps: $inputs.steps
        width: $inputs.width
        height: $inputs.height
        n: 1
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      image: $response.body#/images/0/image
      seed: $response.body#/images/0/seed
      inferenceTime: $response.body#/inference_time
  outputs:
    imagePrompt: $steps.authorPrompt.outputs.imagePrompt
    image: $steps.render.outputs.image
    seed: $steps.render.outputs.seed