AIToday
Large Language ModelsVideo GenerationZenn AI/MLPublished: Oct 7, 2026, 22:00 JST

Spine 4.2 monster set built by AI chat in Riggy tool

Spine 4.2 monster set built by AI chat in Riggy tool

3 Key Points

  1. What happened

    Using Spine 4.2 and a tool called Riggy, a developer got AI to rig and animate four RPG monster moves from one layered PSD; the first prompt produced all four, and about 1 s attack landed usable on the first try.

  2. Why it matters

    The developer says the difference between the smooth attack and the troublesome death came down to prompt wording, suggesting AI animation output depends less on the tool than on how precisely the motion is described.

  3. What to watch

    The write-up credits mouse edits and rig fixes, not the AI, for the final polish, so it hinges on how much finishing work the chat still leaves. The finished rig is usable as standard Spine 4.2 JSON.

WHO IT HITSThis lands on game artists and technical animators working in Spine who are weighing whether AI chat tools can absorb the first pass of rigging and motion. The write-up suggests their real workload may shift toward precise prompt writing and manual mouse-level fixes.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

The write-up describes starting from a single layered PSD and having Riggy's chat produce a whole combat set — idle, attack, hit and death — for an RPG monster, with rig and animation both AI-generated. What separates the smooth attack from the difficult death, according to the developer, is not the model but the instruction: describing the length of the motion, which parts move and which stay still, the order of the action, how the rest of the body participates, and the game-side event needed at the moment of impact.

For fixes, the developer found that vague requests kept failing and concrete ones worked. Telling the model what happens at which second and in which direction — for example that at 0.5–0.7 s the feet lift and the club should stay low, or that from 1.0 s to the end the club should be laid flat by 1.5 s — corrected problems in one pass, because the model assembles poses numerically. Explicitly approving animations already accepted also protects them from later edits, though changes to the rig itself, such as which bone a part hangs from or the draw order, affect every animation and force a full re-check.

The article also flags rig problems that exist with or without AI: a belt painted on the torso but parented to the hip bone stayed behind when the torso leaned forward, arm draw order caused the front arm to slide behind the body, and the ground line was about 70 px above the bottom edge of the image, so instructions based on the image edge pushed the feet into the floor. Small adjustments such as re-parenting bones or nudging a key pose were faster and free with the mouse in Motion mode. Whether AI chat carries more of this work forward is likely to depend on getting the rig right first and on how precisely each motion is described.

FAQ
What tool and version did the developer use for the monster animations?
The animations were made in Spine 4.2, with the rig and animations generated by AI through the chat of a self-made tool called Riggy.
Which animation was the hardest to get right?
The death animation was the most difficult, while the attack came out usable on the first try, which the developer attributes mostly to prompt wording.
Can the finished animations be used in a game?
The export is standard Spine 4.2 JSON and atlas, so it can be finished by hand in the Spine editor or loaded into a game. Shipping a game with the Spine runtime requires a license from Esoteric Software.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleopenTPU uses AI to design AI chips, runs Qwen3 on FPGA