[SD 1.5][Checkpoint]Min Illust Background 2250
<p>Originally posted to <a href="https://huggingface.co/ProGamerGov/Min-Illust-Background-Diffusion" rel="ugc" target="_blank">HuggingFace by ProGamerGov</a></p><p>This fine-tuned Stable Diffusion v1.5 model was trained for 2250 iterations with a batch size of 4, on a selection of artistic works by Sin Jong Hun. Training was performed using <a href="https://github.com/ShivamShrirao/diffusers" rel="ugc" target="_blank">ShivamShrirao/diffusers</a> with full precision, prior-preservation loss, the train-text-encoder feature, and the new <a href="https://huggingface.co/stabilityai/sd-vae-ft-mse" rel="ugc" target="_blank">1.5 MSE VAE from Stability AI</a>. A total of 4120 regularization / class images were used from <a href="https://huggingface.co/datasets/ProGamerGov/StableDiffusion-v1-5-Regularization-Images" rel="ugc" target="_blank">here</a>. Regularization images were generated using the prompt "artwork style", 50 DDIM steps, and a CFG of 7.</p><p>Use the tokens <strong>sjh style</strong> in your prompts for the effect. Note that the effect also appears to occur at a much weaker strength on prompts that steer the output towards specific artistic styles.</p><p>This model will likely not perform well on generating portraits and related tasks, as the training data was primarily composed of landscapes.</p><ul><li><a href="https://huggingface.co/ProGamerGov/Min-Illust-Background-Diffusion/resolve/main/v1_size_512x768_t3x4.png" rel="ugc" target="_blank">Full Image</a></li><li><a href="https://huggingface.co/ProGamerGov/Min-Illust-Background-Diffusion/resolve/main/v1_size_512x512_t4x10.png" rel="ugc" target="_blank">Full Image</a></li><li><a href="https://huggingface.co/ProGamerGov/Min-Illust-Background-Diffusion/resolve/main/v1_512x512_t4x5.png" rel="ugc" target="_blank">Full Image</a></li></ul><p>Example images were generated with the v1 2250 iteration model using 50 steps of DPM++ 2M Karras with a format of:</p><p><prompt>, sjh style</p>
页:
[1]