Why do we need artworks?
Fast Line Art is using neural networks. They differ from the classical algorithms that use human-defined rules to do the task - they make up the rules themselves based on data, so they need data to train on. In the case of this particular network, it will need both sketches and line art – sketches as an example input, and line art as expected output. It needs quite a big amount of data – thankfully not as much as generative AI, but still, let’s say, several dozens.
The way it works is that the network takes the sketch, processes it and outputs a picture. Then our algorithm will compare the output picture with the expected line art, and the network will try to correct the errors. And that happens for all pictures many, many times over. For the network to work well, it needs to slowly accumulate small changes, and it needs varied input and output so it can work universally, and not just on one or two artworks or even artists.
If you are an artist that uses line art in your work, it will be a huge help if you can donate your time and work and draw a sketch and line art to be used in the training of the neural network.
For all the details about the project, please check out Introducing a New Project: Fast Line Art
License for your artworks
If you decide to participate, we ask you to license your work with this license:
Creative Commons Attribution License (CC-BY), with the requirement of attribution waived in the following specific circumstances:
- The Licensed Material and its derivatives are used by Stichting Krita Foundation, located at Korte Assenstraat 11 7411JP Deventer, the Netherlands.
- The Licensed Material and its derivatives are used by Halla Rempt Software, located at Korte Assenstraat 11 7411JP Deventer, the Netherlands.
- The Adapted Material is a machine learning model trained on the Licensed Material, which was created and trained and developed only by either Stitching Krita Foundation or Halla Rempt Software or both.
- The Adapted Material is an image that is a derivative of the Licensed Material that was created using a machine learning model trained on Licensed Material, where the machine learning model was created and trained and developed only by either Stichting Krita Foundation or Halla Rempt Software or both.
- The Adapted Material is a machine learning model licensed with GPL license.
Let’s call the license “CC-BY-KRITA-FLA” from now on.
Submissions
The neural network will learn easier and faster, if our dataset is as close to its task as possible.
-
We strongly suggest creating a new dedicated artwork instead of donating some of your older work, because of all the rules below.
-
You have to have all rights for the artwork (must be drawn by you, it cannot be a fanart, it shouldn’t be anything done for clients or using other people’s property, or based on a photo you don’t have rights to etc.).
-
The artworks must be licensed as in the section above (“CC-BY-KRITA-FLA”).
-
You need to submit artworks paired: sketch and lineart for the same picture.
-
The artworks must be submitted in PNG or KRA formats, and should be at least A4 300 dpi (2480 x 3508), but preferably A3 300 dpi (3508 x 4960) or more. (But note the brush sizes requirements). Both artworks needs to be the exact same size (best to just draw one on top of the other in the same document and save both to PNG the same way).
-
The sketch can be drawn with any reasonable brush with a reasonable size (in comparison to the lineart thickness - it doesn’t have to match exactly, but it has to be reasonable).
-
The sketch must be done in dark color over a white canvas. The sketch must be digital.
-
The lineart must be done on white canvas using black color with 100% opacity.
-
There are only a few brushes and brush sizes to choose from:
-
“b) Basic-1” at 12 px
-
“b) Basic-5 Size” at 12 px or 20 px
-
“h) Charcoal Pencil Medium” at 12 px
- When submitting the artwork, provide this information:
-
Name: your name or nickname for credit
-
Brush: which brush and brush size did you use in the lineart (separately for every line art piece if you submit more than one)
-
License: copy the license from the section above (CC-BY-KRITA-FLA).
- Sketch:
- no cross-hatching or shading,
- all the lines that don’t end up in a lineart (like construction lines etc.) should be lighter,
- perspective, composition, anatomy etc. in the picture don’t matter. What matters is whether the line art looks nice and fits the sketch.
- Line art:
-
same size as the sketch,
-
no cross-hatching or shading,
-
no additional details that weren’t included in the sketch,
-
very closely following the sketch. Imagine traditional inking over a pencil sketch: the sketch usually already has all the details, and you’re only following the lines with a pen to make them permanent and nice looking.
-
If you draw typical anime eyes (thick eyelashes line etc.), it’s actually probably better if you don’t include them in either the sketch or the lineart, since it will throw the neural network off. Likewise for all other details that look very different from the sketch, or need to have a special case.
- How to submit:
- You can submit here in this thread - please add a note in your comment that you submit under the license written above (CC-BY-KRITA-FLA).
- You can send an email to foundation@krita.org - please add a note in the email that you submit the artworks under the license written above (CC-BY-KRITA-FLA).
If you have any questions about submissions or the artworks etc., please ask them here and I’ll try my best to answer. If you have questions or comments about the project itself, please check out the other, general thread: Introducing a New Project: Fast Line Art
Thank you very much for your support, I really appreciate it.






