This is a LORA aims to simulate the art style of the artist known as MKaz. If you are interested in the original illustrations, you can check out their FA account.

The artist’s works exude the stereotypical early 90s editorial style of drawing anthropomorphic animals. While the artist has several distinct artstyles (toon style, pure illustration), this LORA focuses solely on simulating their most predominant illustration work type: photo compositing, where the anthropomorphic characters are superimposed onto real photo backgrounds. Additionally, the anthropomorphic characters that are being superimposed are usually artistically nude and outright erotisim is rare.

In the artist's usual photo compositing works, the photographic background typically dominates the image area and carries much of the visual weight.
However, this LORA is not trained to faithfully reproduce the aforementioned composition ratio. Instead, it is much more character focus, therefore during the generation it is expected that the character will occupy a huge part of the image space.

Generating highly convincing photo background is not the main goal of this LORA and candidate models were not benchmarked with that criteria in mind for public release.
One of the greatest weaknesses of this LORA is the grey border that can appear along the right and bottom edges of generated images. This issue seems to happens consistently at Hires.fix upscale ratio that result in the generated image which is indivisible by 8 and also stem from the non-standard aspect ratio used to train this LORA.
If you run into this issue try to adjust the upscale ratio accordingly to ensure the generated image resolution can be cleanly divided by 8, this will usually resolve the issue.

The showcase images are cherry picked and upscaled.

Trigger word: MkazStyle
Recommended tag: furry, anthro
Recommended LORA Strength: around 0.75
Prompting Style: e621 or danbooru tags, requires moderate amount of tags in the prompt to guide the LORA toward a decent output.
Sampling method: Euler, Euler A
CFG strength: 4~7
Significant tag (order from highest exposure to lowest exposure): solo, masculine, sheath, canine, feline, feminine, photo background, greyscale photo background, taur, paws, flaccid, cartoon background
Recommended Resolution (taur): for taur character, it is recommended to use square or landscape aspect ratio so the LORA has more relative horizontal space to work with when generating taur body. IMO square aspect ratio usually result in better taur character artstyle and proportion wise. Please note that taur character is more prone to be generated with defects.

LORA description
Trained on YiffyMix V63 checkpoint, the images showcase here are all generated using YiffyMix V64 checkpoint. Rank/Alpha ratio of 32 to 16.
Predispose to generate nude masculine character with sheath and image with photo background.

LORA known weakness:
1. Grey color band at the edges of the image: if you run into this issue, try adjusting your Hires.fix upscale ratio to ensure the generated resolution can be divided cleanly by 8 and see if the issue goes away.
2. Mediocre photo background: since the base checkpoint wasn't fine-tuned for photorealism, expect mediocre background details.
3. Struggles to consistently generate correct hand-to-object interactions: This has been a rather challenging issue persisted in the SDXL family. The LORA alone can't fix it, so expect to rely on Inpainting or ControlNet to clean up these issues if you found a particular great output with such defect.
4. Artifacting on small features: Expect visual artifacts, particularly on hands, mouths, and the far eye/face area in three-quarter angles.
5. Inconsistent pupil styling: Generated pupils occasionally develop artifacts or overly embellish details, straying from the original art style's simple black pupils.
6. Poor real-world object generation: Everyday items like cars and street hardware routinely fail to generate cleanly, resulting the blobby mess on the output.
7. Inconsistent finger: character could be generated with 4 fingers at one hand and 5 fingers at another.
8. Struggles with continuous wrapping and plausible placement of rope-like accessories: eg: camera strap would frequently break continuity or clipping into character.
9. Taur character has a higher failure rate: generations of taur character usually result in more conspicuous defects.
10. Struggling to generate multiple characters and their interactions: This LORA was trained and benchmarked primarily on artstyle fidelity, this priority may have cost the LORA's capability to generate complex poses and multi-character scenes.