A 3D artist working on the indie game Ashenmoor posted a walkthrough of a single character — a golem — from hand-drawn concept to a rigged, animated character running in Godot. He studied 3D for five years and has been working in the field since 2017. The AI sits in the middle of the pipeline, not at either end.
The loop, in his order:
- Painted a quick concept sketch by hand in Krita
- Refined that image with AI (Krita AI Diffusion, plus a Krita-to-fal.ai add-on he built himself)
- Made a few manual touch-ups on the 2D image
- Generated the 3D model from that image with Hunyuan 3D 3.1
- Rigged and animated it in Blender
- Brought it into Godot
The part worth copying is the sequencing. Asked how he controlled the style, his answer was that he made sure the input image already looked right before handing it to the mesh generator — starting from his own sketch and refining it, rather than generating a concept and hoping the 3D step preserves it. Style control is an input problem, not a model problem.
Two things recur across the thread. The manual work doesn’t disappear, it moves: retopology, UVs, textures and rigging stay human, and a commenter noted the result was faithful to the concept but “a little muddy” — image-to-3D currently gets you a blockout, not a shippable asset. And a second artist in the thread describes landing on the same shape independently: photos instead of sketches, AI to simplify them, then automated retopology with hand-fixed texturing.
The other signal is anxiety, not craft. One developer doing similar low-poly work said he expects the process to draw the same anti-AI sentiment regardless of how it’s built. That social tax is now a line item in this pipeline’s cost model, and it shows up before any of the art does.