I am admittedly kind of a noob at this, but if you enjoy my humble work, any donations are greatly appreciated: <a target="_blank" rel="ugc" href="https://ko-fi.com/insult_to_ninjary">https://ko-fi.com/insult_to_ninjary</a>Your generosity helps me stay motivated to keep improving and to keep trying new things!Trigger Words:Main trigger: smw2yiHelper prompts: parody, style parodyExtra help: pixel art; pixelated<ul><li>The LoRA can potentially work with no triggers, but it may not be strong/consistent.</li><li>For a stronger effect, use the main trigger: smw2yi. You can also weight it numerically or with parentheses - I usually do (((smw2yi))).</li><li>If the (weighted) main trigger is still not enough by itself, next try one or both of the helper prompts along with the main trigger. They can be weighted as well - I usually do them all together, like (((smw2yi, parody, style parody))).</li><li>I didn't include them as official triggers because they usually weren't necessary in my testing, but either pixel art OR pixelated (probably don't need both at the same time) can be used along with the other trigger words to help achieve a proper pixel aesthetic if you're having trouble getting it (but keep in mind that some base models just won't be able to do it no matter what).</li><li>Please note that the main trigger here is only for the Illustrious model - it won't work for the SD1.5 model. The helper prompts may or may not work for SD1.5 (I never tried them), but they shouldn't be necessary as long as you are using a compatible base model.</li></ul>At a Glance (Illustrious):<ul><li>Base model: Illustrated style (I use WAI-illustrious-SDXL)</li><li>LoRA weight: 1 (but slowly go lower or higher if necessary)</li><li>Suggested parameters: Euler a, Automatic, 30 steps, 832x1216/1216x832, CFG 5 (but feel free to experiment!)</li><li>Hires. fix: 1.5x, use a quality upscaler (like 4x-UltraSharpV2), denoising strength 0.5 - 0.3</li><li>Extra: use ADetailer to fix faces and/or hands if necessary</li></ul>Illustrious Version Description:This LoRA produces art in a style similar to Super Mario World 2: Yoshi's Island. So what you'll get is kind of a colorful doodle pixel art style. I really love it! And since it's a pixel art style, precise fidelity is not quite as important, so you shouldn't really need ADetailer as much (but of course still use it if you want to!).Most of the time, the LoRA should produce pixel art even if you didn't specify it in your prompt. This is especially true if you are using the main trigger and the two helper prompts all together. However, there are times when it randomly doesn't produce pixel art. It still looks nice and colorful and cartoony, but it's just not pixelated. If that's a problem for you, either just try again on a different seed, or include something like "pixel art" or "pixelated" in your prompt.The LoRA can also tend to produce images that appear too "zoomed out." Most likely a product of having a lot of screenshots in the training dataset. The helper prompts also seem to intensify this phenomenon. To counteract it (if you want to), you can prompt for the specific composition you want (portrait, upper body, cowboy shot, etc.), and this usually works well enough.Finally, the LoRA seems to do much better with outdoor settings compared to indoor settings. This makes sense, as the game screenshots used in training were mostly outdoors. Thus, it's very good at capturing the style of things you might see in the Yoshi's Island game. But for things it's not familiar with, it just kind of guesses how to apply the style. Thankfully, it's usually pretty good at guessing in a way that looks appropriate and pleasing, and at worst, it will still apply a generic pixel art style that looks acceptable.For more potentially relevant information, you can check out the original SD1.5 description immediately below. Below that, there's a troubleshooting section in case you're having any problems.SD1.5 Version Description:***NOTE: For best results, please use one of the 2 suggested base models with this LoRA: AniDosMix or Koji. Of course you can experiment if you want, but I tried this LoRA with many different base models, and unfortunately it's kind of picky. The vast majority of models didn't work very well. :( [Way later edit: my <a target="_blank" rel="ugc" href="https://civarchive.com/models/2009146/itns-greatest-hits-merge">Greatest Hits</a> merge is a pretty solid general purpose checkpoint that should also work as a fairly neutral base model.]With that out of the way, this LoRA was trained to make art that looks like the SNES game Super Mario World 2: Yoshi's Island. But interestingly, the results you get usually won't look exactly like gameplay screenshots from the game. Instead, you'll get this neat crayon/doodle effect on your art that is very reminiscent of the game's art style, but with a bit of a modern twist provided by your base model.If you're having trouble getting good results, here are some tips:1. If you're not noticing any effect from the LoRA, switch to one of the 2 recommended base models (AniDosMix or Koji) [way later edit: or try my <a target="_blank" rel="ugc" href="https://civarchive.com/models/2009146/itns-greatest-hits-merge">Greatest Hits</a> merge!]. Alternatively, you could try gradually increasing the weight of the LoRA. In theory that should help out, but I haven't tested it extensively.2. If your images are coming out too landscapey, with either really small characters or no characters at all, try gradually decreasing the weight of the LoRA. The recommended base models work ok at the default weight of 1, but you could also lower it to probably around 0.7 or 0.8 and still get some nice results.3. Use a VAE, especially if your chosen base model doesn't have one baked in!4. Upscale or use Hires. fix with a quality upscaler.Troubleshooting:If you're having persistent problems with anatomy, fine detail, and/or other jank, just try your best with all the usual fixes - different prompting (including negative prompt), try a different base model, use a different VAE, try different clip skip values, adjust the LoRA weight, use adetailer, inpaint, etc.Outro:If you make anything awesome, please consider posting them to the gallery! That goes for all of my models! I really like seeing what kind of stuff you all are able to make!If you have any questions, feel free to ask! I will try my best to help! Also, if you have any ideas for future models, please don't hesitate to share them with me!

SMW2: Yoshi's Island Influence

SMW2YI.safetensors

v1.0

IL_v1.0

I am admittedly kind of a noob at this, but if you enjoy my humble work, any donations are greatly appreciated: <a target="_blank" rel="ugc" href="https://ko-fi.com/insult_to_ninjary">https://ko-fi.com/insult_to_ninjary</a>Your generosity helps me stay motivated to keep improving and to keep trying new things!***NOTE: For best results, please use one of the 2 suggested base models with this LoRA: AniDosMix or Koji. Of course you can experiment if you want, but I tried this LoRA with many different base models, and unfortunately it's kind of picky. The vast majority of models didn't work very well. :(With that out of the way, this LoRA was trained to make art that looks like the SNES game Super Mario World 2: Yoshi's Island. But interestingly, the results you get usually won't look exactly like gameplay screenshots from the game. Instead, you'll get this neat crayon/doodle effect on your art that is very reminiscent of the game's art style, but with a bit of a modern twist provided by your base model.If you're having trouble getting good results, here are some tips:1. If you're not noticing any effect from the LoRA, switch to one of the 2 recommended base models (AniDosMix or Koji). Alternatively, you could try gradually increasing the weight of the LoRA. In theory that should help out, but I haven't tested it extensively.2. If your images are coming out too landscapey, with either really small characters or no characters at all, try gradually decreasing the weight of the LoRA. The 2 recommended base models work ok at the default weight of 1, but you could also lower it to probably around 0.7 or 0.8 and still get some nice results.3. Use a VAE, especially if your chosen base model doesn't have one baked in!4. Upscale or use Hires. fix with a quality upscaler.If you make anything awesome, please consider posting them to the gallery! That goes for all of my LoRAs! I really like seeing what kind of stuff you all are able to make!If you have any questions, feel free to ask! I will try my best to help! Also, if you have any ideas for future LoRAs, please don't hesitate to share them with me!

Files with SHA256: 51808008dd736371c8ad281971e4f11f5ac85356b0cdc8e5f952eed0f95ca046