OpenAI Decisions API for sprite animation review: what should I test first?
retained text
I'm using OpenAI's Decisions API (gpt-6-luna) as an advisory reviewer in a game-character creation agent. For walk/run cycles, it receives ordered frame images (16 for walk, 12 for run), an identity reference, and frame durations as JSON. The input is still frames rather than a video. Separate questions cover anatomy, identity/scale, continuity and the loop seam, foot alternation, and visible contact/sliding. Each returns clear, defect, or unverified. I'm using a 0.8 cutoff, but haven't calibrated it on labeled animation cases. The result doesn't automatically approve or regenerate anything. So far, this review isn't helping improve animation quality much. Before swapping models, I'd like to work out whether the problem is the evidence, the rubric, or the reviewer. For anyone building visual review into an agent: 1. How would you build a small, held-out, human-labeled test set? Would you prioritize missed bad animations, rejection of acceptable ones, or something else? 2. Are ordered full frames useful evidence, or would contact sheets and selected frame pairs around foot contact and loop seams be a better experiment? 3. Would you keep a fast classifier for triage and send uncertain cases to a reviewer that identifies the exact frames and visible issue? I'd also want to test whether acting on that feedback actually improves the next version. What would be your first experiment, or a useful reference for evaluating this kind of reviewer?
document fetched 2026-10-30 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1xi99ng/openai_decisions_api_for_sprite_animation_review