AI Image Evaluation & Policy Trainer

Apply; Seattle, Northern (WA, KY); Hybrid; USD 62,000 - 76,000; Full time

Job summary

Handshake is seeking an AI Image Evaluator in Seattle to assess generated images against prompts, evaluate adherence, composition, realism and style, and explain judgments with clear, rubric-based reasoning.

You will compare image pairs, determine the stronger example, classify content per policy, and document edge cases for calibration discussions. A background in photography, illustration, design, or visual effects helps, as does experience with generative tools.

Qualifications

• You have a trained eye for visuals and can justify judgments clearly.
• Experience with generative image tools and understanding of their failure modes.
• Background in policy or content moderation and applying a rubric.
• Excellent attention to detail and ability to reason about edge cases.
• Strong written communication and ability to document rationale.

Responsibilities

• Evaluate generated images against prompts for adherence, composition, realism, style, and technical defects.
• Compare images side by side and select the stronger one with an evidence-based rationale.
• Classify images against customer content policies across multiple categories.
• Provide concise rationales citing rubric language and visual details.
• Distinguish stance from prompt failure and policy violation in judgments.
• Write and refine prompts to probe model quality and safety behavior.
• Identify rubric gaps and raise them for review with leads and policy teams.
• Calibrate judgments across many evaluations and maintain consistency.
• Apply policy consistently without personal bias.
• Contribute to calibration discussions and cross-team learning.

Skills

Trained eye for visuals

Generative image tools expertise

Policy/moderation experience

Attention to detail

Clear written reasoning

Tools

Midjourney

Stable Diffusion

ComfyUI

Job description

Handshake is seeking an AI Image Evaluator in Seattle to assess generated images against prompts, evaluate adherence, composition, realism and style, and explain judgments with clear, rubric-based reasoning.

You will compare image pairs, determine the stronger example, classify content per policy, and document edge cases for calibration discussions. A background in photography, illustration, design, or visual effects helps, as does experience with generative tools.