Someone asked about running SDXL without a GPU, so here’s the workflow using Google Colab + AUTOMATIC1111, plus some bad news.
Back in the day, Colab was free for SD, but Google changed the policy. Now if you wanna run any SD notebook, you gotta get Colab Pro—$9.99 a month, gives you 100 compute units. Units roll over month to month, but expire 90 days after you use 'em.
The most recommended notebook in the community is TheLastBen’s Fast Stable Diffusion. Here’s the flow: open it, change the runtime type first. With Pro, you can pick V100 or A100. Heads up—A100 eats about 15 compute units per hour, V100 only 5. Stronger isn’t always better value, depends on what you’re doing. After picking, hit the start button top right. Runtime maxes out at 12 hours; closing the tab won’t end the session, but it’ll time out after 12 hours. Remember to manually stop it when you’re done, or you’ll burn through units for nothing.
Then run the cells in order. It’ll ask you to connect Google Drive, download models and dependencies to your cloud drive. Once downloaded, you don’t need to re-download next time. At the last step, wait for a URL to pop up—click it and you’ll get the A1111 interface.
Before generating, tweak these: resolution 1024x1024, sampling steps at least 30, sampler pick DPM++ 2M Karras or anything with Karras in the name.
