I made lesser known guy or obscure character, Voldon, from lesser known Pokemon game, Pokemon Battle Revolution. This guy is so obscure that finding a decent dataset about him is next to impossible. If you find him via Google Images, good luck on finding him. I managed to only get literally 1 decent image. But in the end, I used <s>30+</s> 80+ images and image enhancer to make datasets a bit better in terms of quality, plus making a decent data synthetic of him. Trigger Word<pre><code>voldon</code></pre>Settings I Use:<ul><li>Weight: 0.6 - 0.85</li><li>Sampling Method: DPM++ 2M SDE Karras</li><li>Sampling steps: 23 - 25</li><li>Resolution: 768x768 to 960x960</li><li>CFG Scale: 7</li></ul>I use Mature Male Mix and AnyLoRA with Ghibli Style LoRA. Also, I used AnyLoRA + Pokemon Style LoRA.

Voldon | Pokemon | SD 1.5 [Old Model]

I managed to update this model that's quite challenging to update due to very limited amount to update. I experimented new workflow to increase the quantity of the images though cyclic image generator and create model. In short, I repurposed some image that I generated, made into new model, and evaluated the model, then rinse and repeat. Here's some info about my model;<ul><li>Increase the image quantity in a dataset from 30+ images to 80+ images.</li><li>Increase LoRA size</li><li>Increase train resolution to 640</li></ul>Notes:This model is still bit unstable in terms of finger rendering, inflexible in terms of clothing, inacurracies facial feature in realistic/semi-realistic. So, I still update this model in future update.v1.0:Here's some information about this model:<ul><li>Using 30+ images mostly from Youtube and Google Images</li><li>Epochs: 15</li><li>Steps: 10</li></ul>

Files with SHA256: ca01b5d87d7588b349865ccdda642ebf55a0f3ce2bea12d5306121be53505deb