AI Trainer - Image Counting Tasks at Mercor

AI Trainer - Image Counting Tasks | $60-100/task | Remote

About the role

AI models that read images are often bad at counting, and they get worse when counting takes reasoning. In this remote contractor role, you'll build tasks that expose that weakness: original photos of real-world scenes, counting prompts that require reasoning, and object-level annotations so each answer can be checked.

What the work involves

• Taking your own original, unedited photos of busy real-world scenes, mainly outdoor ones (streets, parking lots, building facades, parks, markets) that hold many countable objects.
• Writing 3-5 counting prompts per image that make the model reason, not just count: conditional filters (How many red cars are parked in the lot?), purpose-based reasoning (How many objects require seatbelts to operate?), spatial relationships, telling similar objects apart, and prompts whose correct answer is 0.
• Making sure every prompt has exactly one correct answer that 9 out of 10 people would agree on right away, then checking and recording that answer.
• Outlining (masking) every counted object in an annotation tool so each answer can be checked object by object.
• Aiming for image and prompt sets where the model gets the count wrong.

Who we're looking for

• A sharp eye for detail, and the patience to count dozens of objects accurately.
• A phone or camera, and access to busy real-world scenes you can photograph yourself.
• Clear, precise writing: prompts can't use vague terms like "closest" or shades of color people would argue about.
• Comfort with careful, repetitive checking, and acting on reviewer feedback.
• Willingness to flag unclear guidelines rather than guess.

Pay and schedule

Pay is per task, based on how many objects your hardest prompt counts: $60 (0-24 objects), $70 (25-49), $80 (50-79), or $100 (80+). Hours are flexible, so you choose how many tasks to take on.

Every image goes through an originality check before you can submit it, and each task is reviewed for accuracy. A task sent back for edits reduces your pay for it, so careful first submissions pay best. Strong reviewers may be promoted to review other people's tasks as the project goes on.

Your photos matter more than your background: a strong submission is an original, busy scene with prompts the model can't answer correctly.

How it works

• Apply with your resume and confirm where you work from.
• Complete the short assessment attached to this listing; it's required to be considered for the project.
• Mercor reviews your profile and assessment results, and reaches out with next steps if you look like a good fit.
• Once onboarded, you'll read the project guidelines, watch the onboarding call, pass the onboarding quiz, and submit one qualifying task (it must be an outdoor photo). After it's approved, you can claim more tasks.

Applying doesn't guarantee a spot. You'll be contacted only if your profile and assessment are a good match, and timing depends on the project's needs. All qualified applicants are considered without regard to legally protected characteristics, and reasonable accommodations are provided upon request.