Here's something a little different after a few artist style LoRAs... this (work-in-progress) LoRA collection is exclusively trained on various style reference codes found on Nijijourney version 6. For every LoRA made, I'll paste the original code. There's no set way I'm choosing these styles, just some from the internet, some I've stumbled across personally. hwsref1 ---sref 1333447729A detailed style that blends a bit of anime and realism, emphasis on the latter. Textures like the clothes and hair are VERY realistic. It has a very desaturated and "candid" aesthetic, if that makes sense. <a target="_blank" rel="ugc" href="https://civarchive.com/models/4855/anythingelse-v4">AnythingElse V4</a> seems to like this LoRA the most. hwsref2 --sref 632624334A 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. A lot of checkpoints are compatible, but <a target="_blank" rel="ugc" href="https://civarchive.com/models/9409?modelVersionId=90854">AnythingV5</a> renders it cleanest.hwsref3 --sref 1742554747A clean and stylized illustration style with bold linework, flat shading, and a vibe that sort of reminds me of vintage comics. Dynamic head angles are common. This LoRA can be unpredictable on many checkpoints. <a target="_blank" rel="ugc" href="https://civarchive.com/models/9409?modelVersionId=90854">AnythingV5</a> is best for this LoRA.hwsref4 --sref 2467884816Super-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. This LoRA controls all that to keep your generations looking normal. <a target="_blank" rel="ugc" href="https://civarchive.com/models/4468/counterfeit-v25">Counterfeit-V3.0</a> works best for this one.hwsref5 --sref 3505859829This 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. <a target="_blank" rel="ugc" href="https://civarchive.com/models/9409?modelVersionId=90854">AnythingV5</a> works best for this LoRA.hwsref6 --sref 2569415758A 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. Updated my training techniques so it's compatible with a lot of checkpoints. <a rel="ugc" href="https://civarchive.com/models/23900/anylora-checkpoint">AnyLoRA</a> seems to work the best for me. Usage Tips<ul><li>The trigger word is "hwsrefX". Replace the X with the right number listed.</li><li>Most NAI-based checkpoints are generally compatible with this LoRA. Check out the recommended resources to see which ones work the best.</li><li>Use Booru language when prompting.</li><li>LoRA strength can be 0.8 to 1.5. Higher strengths may cause more creative details.</li><li>For more anime/illustrated styles, use any DPM or DDIM sampler. For more realistic styles, use Euler/Euler a samplers at 512x768 base resolution. Hires fix is not required.</li><li>Use <a target="_blank" rel="ugc" href="https://github.com/Bing-su/adetailer">ADetailer</a> to fix imperfect faces.</li></ul>

Hoodwinker's SREF Styles

hwsref3.safetensors

hwsref3

Files with SHA256: 0638969c935ff2af8d3dc63385bb9e7223ec1e4973a45008e6890da7c09ffa8b