Dropping some science on the garden today.
I've been testing a pair of these plant sensors for the last few weeks and decided they were useful enough to get more. During testing I found that sensor 2 showed odd readings for light level. Placed right next to each other, when the light level was below 10k both sensors showed approximately the same value, but above 10k sensor 2 was off by a factor of about 2.75. It still followed the same pattern, so it seems like a calibration error.
I got 6 more sensors and placed them all next to each other. Now I'm giving the sensors a chance to stabilize and accumulate data. We'll see how consistent the readings are before putting them in their final locations.
#permies #permaculture #gardenstr #grownostr
I've been testing a pair of these plant sensors for the last few weeks and decided they were useful enough to get more. During testing I found that sensor 2 showed odd readings for light level. Placed right next to each other, when the light level was below 10k both sensors showed approximately the same value, but above 10k sensor 2 was off by a factor of about 2.75. It still followed the same pattern, so it seems like a calibration error.
I got 6 more sensors and placed them all next to each other. Now I'm giving the sensors a chance to stabilize and accumulate data. We'll see how consistent the readings are before putting them in their final locations.
#permies #permaculture #gardenstr #grownostr
(right-click loop and fullscreen for the best effect)
Here's what I learned from testing:
* You need enough VRAM to render all the frames at once, in a single batch.
* 16 frames at 512x512 uses 8.9 GB, 512x768 uses 11.2 GB for me.
* Use the 1.4 motion module. 1.5 almost always has watermarks, and often very little motion.
* Crank the CFG up to 10-25, otherwise the images are faded and lack detail. This seems like a bug in the extension because the original implementation works at normal CFG.
* Going over 75 tokens will change the image halfway through because of the way A1111 handles long prompts, so keep it less than 75.
* Face restore seems less stable between frames, try turning it off.
* Dynamic prompts should be removed, or they'll change for each frame (which might be neat if done intentionally).
* The motion modules were trained on 16 images. You can do more or less but don't expect good results. Using 24 frames often has too much change between frames but can still produce good results. 8 frames is good enough for an image to look "alive" but not long enough to show much motion.
* Remember to edit ui-config.json after running the plugin to set your defaults. Search for "animatediff" and set the motion module and output formats.
AnimateDiff combines well with One Button Prompt, but be careful not to go over 75 tokens. Try turning down the complexity to 3 or 4.
#stablediffusion #aiart #animatediff #grownostr
#permies #permaculture
It might helpful, or at least interesting, to know how a stable diffusion checkpoint was trained. I grabbed this script from reddit and packaged it up for easier use. It compares two checkpoints and shows the tokens with the largest difference between them. These are likely to be the words used more frequently during fine tuning.
#soilfoodweb #permies #permaculture #gardenstr
I didn't intend to delete the post of the beautiful garlic, oops. To make up for it, here's a beautiful red dragonfly I met in the garden today. It was so busy chowing down on something it caught that it didn't mind me sticking a camera in its face.
#stablediffusion
saviour of worlds
