Huggingface: <a target="_blank" rel="ugc" href="https://huggingface.co/sheldonxxxx/MechaDream">https://huggingface.co/sheldonxxxx/MechaDream</a>Support me on Ko-fi to help me buy more GPU time: <a target="_blank" rel="ugc" href="https://ko-fi.com/techadventurex">https://ko-fi.com/techadventurex</a><h1>MechaDream</h1><h3>A Stable Diffusion model for Mecha</h3><img src="https://imagecache.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/b8834019-e857-4a47-7231-1da5a49fa500/width=525/b8834019-e857-4a47-7231-1da5a49fa500.jpeg" />All in one model (ckpt+lora+vae in one safetensors): <a target="_blank" rel="ugc" href="https://civarchive.com/models/47025">https://civarchive.com/models/47135</a><h2>Recommended Setting</h2>Checkpoint: Counterfeit-V2.5VAE: mecha_v2_e3-pruned.ckpt (<a target="_blank" rel="ugc" href="https://huggingface.co/sheldonxxxx/MechaDream/resolve/main/mecha_v2_e3-pruned.ckpt">https://huggingface.co/sheldonxxxx/MechaDream/resolve/main/mecha_v2_e3-pruned.ckpt</a>)Sampler: UniPC (Personal preference)CFG Scale: 15Negative Embeddings: EasyNegative, bad-hands-5Lora Weight: 0.8Hires.fix with scale &gt;= 2 is highly recommend<h2>My workflow</h2>Step 1: Batch generate imagesStep 2: Select the best image and inpaint any flawsStep 3: Upscale using MultiDiffusion🡳 Some comparison images between original image and after the above workflow 🡳<a target="_blank" rel="ugc" href="https://imgsli.com/MTcyNDM2">https://imgsli.com/MTcyNDM2</a><a target="_blank" rel="ugc" href="https://imgsli.com/MTcyNDM3">https://imgsli.com/MTcyNDM3</a><a target="_blank" rel="ugc" href="https://imgsli.com/MTcyNDM5">https://imgsli.com/MTcyNDM5</a><h3>Roadmap</h3>MechaDream v1 will be the base model for generic mecha image generationNew Lora models train on different anime Blu-ray will add flavors into my MechaDreamList of WIP anime:<ol><li>Gurren Lagann (Dataset is ready, will start training when I'm less busy)</li><li>Code Geass</li><li>Macross Frontier</li><li>Evangelion</li><li>Genesis of Aquarion (Maybe)</li><li>Valvrave</li><li>Eureka Seven (Maybe)</li><li>Gundam</li><li>Armored Core</li></ol>

MechaDream lora

Base model: Counterfeit-V2.5The Road to this LoraStep 1: Fine-tuned on a dataset consisting of 130,000+ images of mecha/robots from various sourcesStep 2: Mix the Lora models trained using different parametersStep 3: Test different mixing ratios to find the best model

Files with SHA256: e6c61280679511683cd32b3bc251c92eb68b95697d9dc0ad00b45c3f82e0be96