Image Prompt QA Loop
A 3-step recipe with Hyperbolic. We can help you get access.
Recipe · 3 steps
Image Prompt QA Loop
with Hyperbolic
Generate an image, QA it with a vision model, and re-render on failure.
- Generate a first-pass image from the prompt and capture its base64 bytes.
- Ask the vision model whether the rendered image matches the requested subject, requiring a verdict that starts with PASS or FAIL.
- Re-render the image once at a higher step count when the vision model rejected the first pass.
What you'll need to have ready
- apiKey: Hyperbolic API key passed as a Bearer token.
- imageModel: Image model name (e.g. SDXL1.0-base).
- prompt: The text-to-image prompt describing the desired subject.
- visionModel: Vision-capable chat model id used to judge the image.
For your developer
Arazzo workflow hyperbolic-ai:image-prompt-qa, 3 operations: generateImage → createChatCompletion → generateImage. Specs on APIs.io.