Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -9,13 +9,21 @@ tags:
|
|
9 |
size_categories:
|
10 |
- 10K<n<100K
|
11 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
A filtered version of the [vivym/midjourney-prompts](https://huggingface.co/datasets/vivym/midjourney-prompts) dataset
|
13 |
|
14 |
-
Filtering criteria
|
15 |
- top 10% in length (assuming that longer prompts = more effort and higher quality)
|
16 |
- not niji style
|
17 |
- used on an image to be upscaled (assuming that users are more likely to upscale an image that is aesthetically pleasing)
|
18 |
- used on midjourney version 5.0+
|
19 |
- deduplicated
|
20 |
|
|
|
21 |
filter.py script provided (go to dataset files) so you can create your own dataset however you like
|
|
|
9 |
size_categories:
|
10 |
- 10K<n<100K
|
11 |
---
|
12 |
+
|
13 |
+
|
14 |
+
<img src="https://cloud-3i4ld6u5y-hack-club-bot.vercel.app/0home.png" alt="Akash Network logo" width="400"/>
|
15 |
+
|
16 |
+
Thank you to the [Akash Network](https://akash.network/) for sponsoring this project and providing A100s/H100s for compute!
|
17 |
+
|
18 |
+
# About
|
19 |
A filtered version of the [vivym/midjourney-prompts](https://huggingface.co/datasets/vivym/midjourney-prompts) dataset
|
20 |
|
21 |
+
# Filtering criteria
|
22 |
- top 10% in length (assuming that longer prompts = more effort and higher quality)
|
23 |
- not niji style
|
24 |
- used on an image to be upscaled (assuming that users are more likely to upscale an image that is aesthetically pleasing)
|
25 |
- used on midjourney version 5.0+
|
26 |
- deduplicated
|
27 |
|
28 |
+
# Run yourself
|
29 |
filter.py script provided (go to dataset files) so you can create your own dataset however you like
|