I will admit that this lora doesn't capture the face very will but I do think it's due to the lack of images, I did try another version with a couple of close ups and it ended up being worse somehow, I will give it another go but no until there are a couple more images out thereA lora model of <a target="_blank" rel="ugc" href="https://marvelrivals.fandom.com/wiki/Hela">Hela from Marvel Rivals</a>, trained over 18 images for 1024 using <a target="_blank" rel="ugc" href="https://civarchive.com/models/1004512?modelVersionId=1125858">Initium</a> as a training model<ul><li>Trigger word: mhelar</li><li>Looks tags: green eyes, green lips</li><li>Armor tags: spiked helmet, armor, shoulder armor, black bodysuit, green trim, flaming cape, green cape, claws, high heels</li><li>Optional tags: flaming sword, green sword</li><li>Weight: 1 is perfect</li><li>You can also use the armor in combination with other characters</li></ul>If you wanna support me <a target="_blank" rel="ugc" href="https://ko-fi.com/infamousfish">ko-fi</a>Additional information:<ul><li><a target="_blank" rel="ugc" href="https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu-forge">Webui </a>I use</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/888145/aaa-anime-sdxl-vae">AAA Anime VAE SDXL v1.5</a> as VAE</li><li>Sampling method: Euler a</li><li>Schedule type I set it to Karras</li><li>Sampling steps I go 20-30</li><li>4xNomos8kDAT as an Upscaler</li><li>Hires steps I go 8-15 (I don't always use hires fix)</li><li>Denoising strength I go 0.4</li><li>Upscale by 1.2 to 2.0 depending on the starting image dimensions</li><li>Imagine dimensions vary depending on what I'm trying to do</li><li>I use 6 CFG Scale</li><li><a target="_blank" rel="ugc" href="https://github.com/wkpark/uddetailer">uddetailer</a> for faces, I use mediapipe_face_mesh set Dilation to 50 and Y offset to 30 to get the full face</li></ul>

Hela [Marvel Rivals]

1161115_training_data.zip

hela_marvel_rivals_v2-000013.safetensors

I will admit that this lora doesn't capture the face very will but I do think it's due to the lack of images, I did try another version with a couple of close ups and it ended up being worse somehow, I will give it another go but no until there are a couple more images out thereA lora model of <a target="_blank" rel="noopener noreferrer nofollow" href="https://marvelrivals.fandom.com/wiki/Hela">Hela from Marvel Rivals</a>, trained over 18 images for 1024 using <a target="_blank" rel="noopener noreferrer nofollow" href="https://civitai.com/models/1004512?modelVersionId=1125858">Initium</a> as a training model<ul><li>Trigger word: mhelar</li><li>Looks tags: green eyes, green lips</li><li>Armor tags: spiked helmet, armor, shoulder armor, black bodysuit, green trim, flaming cape, green cape, claws, high heels</li><li>Optional tags: flaming sword, green sword</li><li>Weight: 1 is perfect</li><li>You can also use the armor in combination with other characters</li></ul>If you wanna support me <a target="_blank" rel="noopener noreferrer nofollow" href="https://ko-fi.com/infamousfish">ko-fi</a>Additional information:<ul><li><a target="_blank" rel="noopener noreferrer nofollow" href="https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu-forge">Webui </a>I use</li><li><a target="_blank" rel="noopener noreferrer nofollow" href="https://civitai.com/models/888145/aaa-anime-sdxl-vae">AAA Anime VAE SDXL v1.5</a> as VAE</li><li>Sampling method: Euler a</li><li>Schedule type I set it to Karras</li><li>Sampling steps I go 20-30</li><li>4xNomos8kDAT as an Upscaler</li><li>Hires steps I go 8-15 (I don't always use hires fix)</li><li>Denoising strength I go 0.4</li><li>Upscale by 1.2 to 2.0 depending on the starting image dimensions</li><li>Imagine dimensions vary depending on what I'm trying to do</li><li>I use 6 CFG Scale</li><li><a target="_blank" rel="noopener noreferrer nofollow" href="https://github.com/wkpark/uddetailer">uddetailer</a> for faces, I use mediapipe_face_mesh set Dilation to 50 and Y offset to 30 to get the full face</li></ul>

Files with SHA256: 358794c80de957e6c80d1a9c607f53c357f49c0bd42033fb42076bb8d0ce7005