High-resolution image generation (4–16 megapixels) on memory-constrained mobile devices is achievable by generating a base image, upsampling, patchifying in latent space, and applying semantic noise injection to refine texture and seamlessly blend patch boundaries.
Pixel Rush generates a base 1K image, upsamples it with off-the-shelf super-resolution, then encodes to latent space and divides into patches. Each patch gets guided semantic noise to add high-resolution texture, and boundary-aware noise blending resolves seam artifacts — all within the original latent space dimensions. This yields ~35× speedup (from ~10 minutes to ~20 seconds) while keeping memory footprints mobile-compatible. ✦ AI generated
Fati Periqi · The TWIML AI Podcast · 2026-08-12 · original ↗
starts at this moment · 36:37
“Can we do like 4 megapixel image generation, 16 megapixel image generation?”
It starts with a prompt and there is this base generation like any model... and then it generates let's say a base image... 1k image. Then we have this cascade upsampling stage... it generates let's say higher resolution image. So when we do that we have now let's say 16 megapixel in the image... then we take that image and then we apply an encoder, a VAE, then we go into latent space in that latent space... we divide that latent space into manageable chunks we patchify them... when we are doing blending across such latent space data representation of for these 16 patches we like to again do another noise injection when we combine them together and but in this case it is not like all over the patch same in the latent space... we kind of allow algorithm to generate more noise towards the boundary but in the center of the patches be maybe less at a lesser degree... it's maybe 35 times faster. You know, from let's say 10 minutes to around 20 seconds type of acceleration.
verbatim transcript · starts at 36:37
36:18want people to uh go and have to find fine-tune those models but we are saying that hey you can still use any of those models that you have and then uh follow our pipeline we discussed in the paper so you can use that model still generate let's say four times 16 times more pixels >> and so talk a little bit about the generation process like what what's
36:47different about the way you've approached this >> yeah absolutely so it again starts with a prompt and there is this base generation like any model it could be let's say flat model and then it generates let's say a base image what base image what I mean by that let's say 1k image and Then we have this cascade upsampling stage. That is the part that uh is no about this paper. That's why I
37:14say CPR paper. This cascade upsample um what it takes it takes this image which is like no RGB pixels not latent space and then it uh case for instance any using any OP image super resolution solution. It could be by cubic up sampling or you know kind of it could be something smarter it generates let's say higher resolution image. So when we do that okay we have now let's say 16
37:46megapixel in the image not one megapix we have like a lot of pixels and then we take that image sam and then we apply an encoder a va then we go into a know kind of latent space in that latent space of course I mentioned that we want to we are concerned about the memory we now divide that uh latent space into manageable chunks we pify them. So then
38:17we do improve those latent space uh features but then uh this this we are still in the latent space we go to the um through a VA decoder in this case to the pixel space something we need to be very careful here yeah there are solutions also you know kind of using pification like I'm going to take and create a you know kind of like patch then another patch another patch then
38:46When you do that you create artifacts actually you know like scenes visible >> meaning when you do that in your origin space you create patches what's different here is that you're doing it in the latent space >> absolutely there are many reasons one is latent space is much smaller special uh dimensionalty than the original pixel space the other one is in lat space we can induce nodes
39:14[laughter] And that is a smart way of leveraging noise. Uh yeah. Uh that is the other reason. Yeah. And so you have in in the image describing the cascade uh coarse latent refinement stage and then high quality latent. Are are these two separate latent spaces or is it one latent space? Like what does this mean? help me understand like wrap my head around >> lat space is the you know lat space of
39:49the kind of original model it is not a separate space. So what does the refinement stage do? >> Refinement space that adds some guided semantic noise and then it uh iterates a couple of time that is the new part that's the part that we provide. So it starts with this latent features. It gets some additional uh kind of uh flexibility through this injected semantic noise. We want that because we