alt Hacker NewsThere's an interesting question about scope.
The IGA FAQ states, in its entirety on this topic: "Games developed using generative AI are strictly ineligible for nomination." [1]
Sandfall probably interpreted this reasonably: no AI assets in the shipped product. They say they stripped out AI placeholders before release (and patched the ones they missed). But the IGA is reading it strictly: any use during development disqualifies.
If that's the standard, it gets interesting. DLSS and OptiX are trained neural networks in an infrastructure-shaped raincoat—ML models generating pixels that were never rendered. If you used Blender Cycles with OptiX viewport denoising while iterating on your scenes, you "developed using generative AI."
By a strict reading, any RTX-enabled development pipeline is disqualifying. I wonder if folks have fully thought this through.
[1] https://www.indiegameawards.gg/faq (under "Game Eligibility")
Replies
Before people chime in to claim "That isn't what we meant"...
DLSS and Cycles denoising are, well, denoising. It's the same process as denoising in Stable Diffusion, essentially, and was trained in the same way.
> Sandfall probably interpreted this reasonably: no AI assets in the shipped product.
Developed and shipped are different words.
> But the IGA is reading it strictly
You meant they meant it strictly? They wrote the policy.
Nobody's probably thought this through, but if I had to guess, the first revision to the rule will be "no _assets_ generated with gen AI" because the most upset parties about gen AI use in gamedev are asset creatives who create textures, models, and audio, perform music and voice acting, etc.
Upscaling technologies are transformative but post-processing. The uproar isn't over what happens in the render pipeline but in the creative process.
Same reason why auto-LoD generation wouldn't and hasn't pissed anyone off: it's not generating LoDs of a mesh that's problematic, it's generating the source model that an artist would create.