Sign In

Centerflex

247
1.5k
29
0
Verified:
SafeTensor
Type
Checkpoint Trained
Stats
359
Uploaded
Jul 5, 2023
Base Model
SD 1.5
Usage Tips
Clip Skip: 1
Trigger Words
Ligne claire illustration
By Jean-Pierre Gibrat
By Roy Krenkel
By Milo Manara
By Francois Schuiten
Hash
AutoV2
5380820A9D
0
0
0
0
0

This is a non-anime model that can handle illustrated styles in addition to photorealistic (trends to semirealistic by default).

Prompting guide

For Illustrated images

For Illustrated, cartoon, or comic/graphic novel style results, make sure "illustration" is in the positive prompt.

Recommended negative prompt for illustrated images:

3D digital, photo, blurred, grainy, low quality, artist name, signature, watermark, text, web address

You can embellish an illustration with further styles, such as the specific artist names listed in the triggers for this model, or with a hybrid illustrated style using the keyword phrase "ligne claire illustration"

For Photorealistic images

Recommended negative prompt for realistic photographs:

3D digital illustration, blurred, grainy, low quality

Artist styles

Artist styles that this model was explicitly trained to on include Roy Krenkel, Jean-Pierre Gibrat, Milo Manara, Willy Vendersteen, Francois Schuiten, Edgar P Jacobs, Herge, Jacques Tardi, and Moebius.

To invoke a specific artist styles, simply add "by Artist Name" to the positive prompt, e.g., "illustration of a woman on a tropical beach, by Roy Krenkel"

Artist names can have a very strong impact; try them with weak weights, e.g. "(by Roy Krenkel:0.1)" or "(by Jean-Pierre Gibrat:0.1)" and see how a dash of their style influences the composition and color scheme.

Ligne claire

This model contains a general activation keyword phrase to evoke a certain hybrid comic book style and composition. This phrase is "ligne claire" meaning "clear line" after the Franco-Belgian comic book tradition.

For example: "A ligne claire illustration of a female science fiction bounty hunter."

Background/history

A funny thing happened when I was training my photorealistic Centerfold Blend. I found that when I incorporated illustrated images in the training data (properly tagged, of course) the model produced better photorealistic images.

Centerflex is an extension of an older version of Centerfold Blend, with a great deal more non-photographic training data. It is intended to be more flexible than Centerfold (hence the name Center+Flex), especially regarding illustrated styles. It was trained with a variety of artist styles in addition to high quality photographs of people.