Want more LoRAs? Show your support by using my resources! I love seeing what people make.

Meri Kuri, Hallelujah, Banzai Tree !! Hope you creators are having a great holiday season. Tis the season for more LoRAs, I think!

Hello! I hope you guys are doing well. After working a bit on my SREF style collection for 1.5, I dreamed about training it on more advanced models... unfortunately my computer is a potato. Well, not really, I can run advanced models with ease. However, training them locally is a different story, I always seem to run out of memory no matter what I do. I ended up training on Civitai's LoRA training suite and the results seem to be quite serviceable.

For those of you who are unfamiliar with Midjourney's style parameters, Midjourney has a gigantic set of codes that can trigger all sorts of randomized, stylistic effects. I'm choosing styles that stick out to me personally, there's no set way I'm choosing the styles I train. For every style, the original code will be attached in case you want to use the styles properly.

hwsref1 ---sref 1333447729

A semi-realistic, somewhat hazy art style. Textures like the clothes and hair are VERY realistic. It has a very desaturated and "candid" aesthetic, if that makes sense.

hwsref2 --sref 632624334

A very moe and glossy anime art style, with most characters having round and almost doll-like heads. Hair, clothing, accessories and character surroundings are very shiny and almost holographic.

hwsref3 --sref 1742554747

A clean and stylized retro illustration style with bold linework, flat shading, and a vintage comic aesthetic.

hwsref4 --sref 2467884816

Super-saturated, painterly and detailed anime art style; it's really giving that stereotypical "90s anime" aesthetic. Nijijourney likes to make pictures with this code in mostly cool tones, night scenes, and lots of red splotches that could look like blood too. Don't worry, said splotches will only appear if you prompt for them specifically.

hwsref5 --sref 3505859829

This style is very sketchy, with loose and rough lines. Even blend of western comic and manga influences in the style reference, different checkpoints add a more western or eastern bias. Many images generated with Niji 6 have limited palettes; like the previous LoRA those effects will not trigger unless you specifically prompt for them.

hwsref6 --sref 2569415758

A real quirky and moody art style. Characters don't have detailed textures but they make up for that with really interesting shapes, a psuedo-3D style and some really cool lighting effects. Rim lighting and backlighting are common.

hwsref7 --sref 2252558232

The first entry when it comes to LoRAs made exclusively for Illustrious, I won't make any more LoRAs for 1.5 at the moment unless there's a demand for them. This style seems to be pretty popular according to some Midjourney style reference code aggregators and I see why! It's very colorful, scribbly... a really nice consistent hand-drawn style that reminds me heavily of children's book illustrations from years ago. Quentin Blake, anyone?

hwsref8 --sref 4169240941

According to a Midjourney sref aggregator, this is the most popular sref of all time. I might end up training some of the popular codes. If you thought hwsref2 was shiny and glossy, this amplifies it to 10. It's very refractive, and chock full of shiny crystalline aesthetics. This style can add some real shiny and stylish beauty to your generations.

hwsref9 --sref 680572301

Another popular code. This one is VERY saturated, very surreal... it has a bit of a pop art, maybe some memphis, late 80s/early 90s aesthetics going on? It's a very cool look.

hwsref10 --sref 2040989065

A very bold, crunchy, offbeat and high contrast anime retroslop style, for those of you who really like your "Akira" vibes or you're the kind of person that likes taking drugs and popping in forgotten 20th century anime OVAs. Compared to other "retro anime" LoRAs, this one has its own unique energy going for it. Some of the shapes, backgrounds and composition are reminiscent of ukiyo-e paintings. Midjourney oftens add weird garbled watermarks and logos onto images.

Usage Tips

The trigger word is "hwsrefX". Replace the X with the right number listed.
Any model with Illustrious based ancestry, including NoobAI EPS models are compatible.
Use Booru language when prompting.
LoRA strength can be 0.8 to 1.5. Higher strengths may cause more creative details.
Use Euler a sampler at supported resolutions for Stable Diffusion XL. Hires fix is not required.
Use ADetailer to fix imperfect faces.

Search AI Models & Addons

very few results