Comment by A_D_E_P_T
1 day ago
What's interesting is that you can't do the same with long-form AI prose. You can ask any given LLM to write in different styles, but it will always fall back on the same rhetorical and structural techniques, so the "new style" is strictly surface-level. In fact, it's getting worse over time -- especially in Claude from Opus 4.6 --> Opus 5.
I suspect that the same thing is happening here -- as other comments note: lack of distinctiveness, uniformity -- but that the low information content of posters prevents it from becoming immediately obvious.
You can't really do this with images either. The least AI-looking images in the post are ones that have the least amount of content, the fewest things to mess up. Once you introduce more detail, it starts converging on the same flaws you usually see, just covered up by a thin veneer of a different artstyle.
It's a thing inherent to diffusion models I think, a kind of uniformity of content across the image.
The similarity may be partly the image model's fault – especially if it's been post-trained/distilled towards performance, correctness and "quality" (for some value of "quality" anyway) which inevitably occurs at the expense of creativity and variation.
But I think it might be more about the LLM's lack of creativity in coming up with the prompt for the image (or "embellishing" a user's prompt), or directly the embedding vector if the LLM is itself the image model's text encoder. Then whatever the image gen outputs is simply an instance of the GIGO principle. It would be interesting to test whether similar cliches and motifs also occur if you ask the model to create an SVG rather than a raster image.