3 ms·

> I don't really see other usage of this My hot-take: this is the future of high-fidelity prompt-based image generation, and not diffusion models. Cycles (or a

> I don't really see other usage of this My hot-take: this is the future of high-fidelity prompt-based image generation, and not diffusion models. Cycles (or any other physically based renderer) is superior to diffusion models because it is not probabilistic, so scene generation via LLM before handing to off to a tool leads to superior results, IMO - at least for "realistic" outputs.

ghurtado 1y ago
> My hot-take: this is the future of high-fidelity prompt-based image generation and not diffusion models Why are those two the only options?
- overfeed 1y ago
  > Why are those two the only options? I made no such claim. The only thing I declared is my belief in the superiority of PBR over diffusion models for a specific subset of image-generation tasks. I also clearly framed this as my opinion, you are free to have yours.
  - ghurtado 1y ago
    > also clearly framed this as my opinion, you are free to have yours. Yes, thank you for your generosity. You very clearly framed your opinion as a dual choice ("this instead of that", "rather than", "either or"). The most natural way to read your comment is that way: one or the other. That's the way the English language works. If you meant something else, you should have said something else.
    - overfeed 1y ago
      You may have missed my first 3 words that did the level-setting. Infact, the first word should have been enough to signal subjectivity, that's just how the english language works. Hot take: a piece of commentary, typically produced quickly in response to a recent event, whose primary purpose is to attract attention
raincole 1y ago
Of course no one knows the future, but I think it's very plausible that the future of films/games (especially games) tech resembles something like this: 1. Generation something that looks good in 2D latent space 2. Generation 3D representation from 2D 3. Next time the same scene is shown on the screen, reuse information from step 2 to guide step 1
- overfeed 1y ago
  That's an interesting idea! I'm thinking step 2 might be inserting 3d foreground/hero objects in front of a 2d background/inside a 2d worldbox

ghurtado 1y ago

> My hot-take: this is the future of high-fidelity prompt-based image generation and not diffusion models Why are those two the only options?

overfeed 1y ago
> Why are those two the only options? I made no such claim. The only thing I declared is my belief in the superiority of PBR over diffusion models for a specific subset of image-generation tasks. I also clearly framed this as my opinion, you are free to have yours.
- ghurtado 1y ago
  > also clearly framed this as my opinion, you are free to have yours. Yes, thank you for your generosity. You very clearly framed your opinion as a dual choice ("this instead of that", "rather than", "either or"). The most natural way to read your comment is that way: one or the other. That's the way the English language works. If you meant something else, you should have said something else.
  - overfeed 1y ago
    You may have missed my first 3 words that did the level-setting. Infact, the first word should have been enough to signal subjectivity, that's just how the english language works. Hot take: a piece of commentary, typically produced quickly in response to a recent event, whose primary purpose is to attract attention

raincole 1y ago

Of course no one knows the future, but I think it's very plausible that the future of films/games (especially games) tech resembles something like this: 1. Generation something that looks good in 2D latent space 2. Generation 3D representation from 2D 3. Next time the same scene is shown on the screen, reuse information from step 2 to guide step 1

overfeed 1y ago
That's an interesting idea! I'm thinking step 2 might be inserting 3d foreground/hero objects in front of a 2d background/inside a 2d worldbox