Funny that you assume the image is anything like what the game will look like. The image itself is also slop.
I am a 3D artist that does game props. I’ve tested the 3D AIs out there, and watched a couple YouTubers try the paid options just to see what I’m competing with.
Generating an entire scene (by having chat gpt use blender) is both expensive and takes hours, and is consistent with image slop. It looks good at first glance but on closer inspection is full of artists. Things are clipped into each other, rotated incorrectly, rooms and house layouts don’t make sense, the materials are flat and basic, normals are the wrong way, etc.
Things like meshy ai are surprisingly accurate with generating a prop from a 2D image but it’s extremely unoptimised and will never be able to be used in a game. Like I tested it on a render of one of my models and it did a decent job recreating it, but my model was 64 triangles. The ai copy was 265,000. For reference, Blender starts to become unusable around 2m triangles, and Overwatch characters are around 34k on average. Which is considered extremely high for a game asset. And the ai model was untextured. Uv unwrapping and texturing is probably about 70% of the work, modelling it is the easy part.
I assume it’ll just pick out one of the 8 hairstyles and eyes closest ti your description like a shitty Mii ripoff, slap a few personality tags on it and call it a day.
Funny that you assume the image is anything like what the game will look like. The image itself is also slop.
I am a 3D artist that does game props. I’ve tested the 3D AIs out there, and watched a couple YouTubers try the paid options just to see what I’m competing with.
Generating an entire scene (by having chat gpt use blender) is both expensive and takes hours, and is consistent with image slop. It looks good at first glance but on closer inspection is full of artists. Things are clipped into each other, rotated incorrectly, rooms and house layouts don’t make sense, the materials are flat and basic, normals are the wrong way, etc.
Things like meshy ai are surprisingly accurate with generating a prop from a 2D image but it’s extremely unoptimised and will never be able to be used in a game. Like I tested it on a render of one of my models and it did a decent job recreating it, but my model was 64 triangles. The ai copy was 265,000. For reference, Blender starts to become unusable around 2m triangles, and Overwatch characters are around 34k on average. Which is considered extremely high for a game asset. And the ai model was untextured. Uv unwrapping and texturing is probably about 70% of the work, modelling it is the easy part.
Holy shit that’s so much worse than I thought lmao
I assume it’ll just pick out one of the 8 hairstyles and eyes closest ti your description like a shitty Mii ripoff, slap a few personality tags on it and call it a day.