[Large Image] Generative Models

Extremely slow rendering of an image file (JPG) set up for a 4X “restore” from a 3000x4000 to 12000 x 16000. It’s been running a half-hour now, progress bar has barely moved. Gigapixel 8.3.4. No end in sight. Local process (not cloud). I’m not running anything else at the moment.

Computer specs = Lenovo ThinkStation P520C, Intel(R) Xeon(R) W-2225 CPU @ 4.10GHz, 128GB of RAM, Windows 11 Pro for Workstations (64-bit) version 24H2, OS build 26100.3915. NVIDIA T1000 8GB.

What type of GPU is in your system?

NVIDIA T1000 8GB

We’re now well beyond an hour of no movement on progress bar for rendering.

The generative models are going to be very slow on that gpu.

I might expect a 3000x4000 image to take 3-4 hours (maybe more), assuming it hasn’t failed.

1 Like

Good to know. It seems to be very slowly moving the progress bar across, almost to half-way point now. Not using much in the way of system resources. (Any suggestions on a more robust / faster GPU?)

at least an RTX because of the tensor cores! :slight_smile:
you can work on your photo remotely on my pc for a moment if you like.

1 Like

So many nice people who have already said it all. :smiley:

1 Like


Just over three hours to render, unimpressive result. Granted, the source was a cell phone image, so I really wasn’t expecting much … but still … sigh …

Three hrs?

When using the basic Models i dont count the time because it does not matter for single images (less than 5 minutes).

Its much longer for the generative Models (less than 20 - 40 minutes).

I’m using a RTX 4090, but other not that big GPU does the job too in much less time.

Hello!

Recover and Redefine are optimized for images with pixel dimensions of 1024 or smaller.
When working locally with small images:

  • Recover focuses on accurately restoring photo detail
  • Redefine is tuned to enhance artwork with a stylized, creative edge
    Keep in mind: both models are resource-intensive and have elevated hardware requirements—at least 8GB of VRAM is recommended to avoid processing

For best results when rendering on your own machine, resize the image to fit within 1024×1024 pixels. If your system can’t meet the demands, cloud rendering is available as a fallback.

For most upscaling tasks, we recommend using the core models. After importing an image, enable Auto mode to establish a strong starting point, then adjust settings as needed based on your goals.