iCreatAI News
Skip to content
Trending storyUpdating

Author tests three image models for generating game sprite animations

1 reports1 sources3 hr agoUpdated

Understand this story

AI overview

Author doener posted a hands-on test on Hacker News: using three models—GPT-Image-2, 2.5-flare, and 2.5-sunburst—on the same Azure endpoint, they completed four types of tasks: a spinning coin, a crow swinging a sword, character layering, and game UI slicing, generating 153 times in total. All three models could only draw single frames and struggled to ensure consistency across sequence frames. The author states that each test was run only once, each using one coin, one crow, and one game UI model; currently none of the outputs can be used directly in a game and can only serve as a starting point for artists, not a replacement. The test only covers the four task types above and single runs, and the conclusions are limited to that scope.

AI overview based on source reports · Updated 2 hr ago

Coverage timeline

Follow the coverage to explore different perspectives.

Nov 20
  1. Hacker News AI
    AI Can Draw Any Frame. Animation Is Not Its Strong Suit

    The author ran 153 generations across GPT-Image-2, 2.5-flare and 2.5-sunburst on four tasks: a spinning coin, a crow sword slash, cutting the crow into layers, and extracting assets from a game mock-up. All three models failed the coin spin because a single frame cannot encode sequence, while poses, body parts and single assets worked; frame-by-frame animation kept the character but drifted in size by up to 27.4%, and cutting the crow into seven layers once allowed zero-drift rigged animation with no further calls. From the mock-up, all 60 of 60 assets were usable and the rebuilt screen reached about two-thirds of the way to a perfect copy, with the author concluding none of the output is game-ready yet.

Activity for this story

Not enough continuous observations to show a trend yet.