This Illustrious based checkpoint merge produces 2D Anime pictures with a clean and detailed style. It is the result of merging 2 of my current favorite anime checkpoints.<h3 id="update">Update</h3>V-pred V3.0This release features a cleaner anime look with vibrant colors. The anatomy and prompt adherence also improved and additional models were added to the merge. For best results, the following quality tags are recommended: Positive tags<pre><code>masterpiece, best quality, absurdres, highres, very aesthetic, high quality, detailed, insanely detailed, beautiful, very awa, anime screencap</code></pre>Negative tags<pre><code>lowres, bad quality, worst quality,bad anatomy, sketch, jpeg artifacts, ugly, poorly drawn,blurry, transparent background, censored, (simple background:1.6), artist name, signature, watermark</code></pre>V-pred V2.0 Made some improvements to to the style and prompt adherence and added an additional model to the merge.V-pred Version 1.0 This a Noob AI V-pred based version of Hotaru Blend. It is still work in progress and the style is not yet where I want it. But it is already working very well. It has a good prompt adherence and results are reproducible to some degree.Recommended settings: Sampler: Euler A + SGM Uniform or DDPM + Align your Steps CFG: 3-5 (I prefer 3.5) Steps: 25merged models copycat-noob obsV1(vpred) hikariNoobVPred_121 CocoaMix style Yume (Illustrious) v-pred V2 (in V-pred V2.0)chromalinemixNoobaiV_40 (in V-pred V3.0) cocoamixYumeIllustrious_v50 (in V-pred V3.0) raehoshiIllustXL_vpredV10 (in V-pred V3.0) + additional styling lorasVersion 4.0 DMD2 This release is the same model as Version 4 with an additional DMD2 lora merged into for faster image generation. The quality is nearly identical to the main model, except that this version has a bit more saturated colors.Recommended settings:Sampler: Euler A + SGM Uniform Steps: 8-12 (recommended 10) CFG: 1.5 - 2 (a bit higher should also work)Version 4.0For this version I added an additional personal merge model which improves the creativity of the model. The colors are now a bit less vibrant and instead have "smoother" look now. Version 3.0This release provides some overall improvements. I tried to make the style more like 2D Anime illustrations.I made 2 quality embeddings for positive and negative prompt, that can be used an an alternative to the below mentioned Quality tags. You can find them here: <a target="_blank" rel="ugc" href="https://civarchive.com/models/1821398?modelVersionId=2061138">https://civarchive.com/models/1821398?modelVersionId=2061138</a>Version 2.0 This release improves the overall details a bit, most noticeable with the hair and reflections. The prompt adherence should also be better then the previous version.The recommended settings are the same as for Version 1.<h3 id="recommended-settings">Recommended settings</h3>Sampler: Euler A / Scheduler: SGM UniformSteps: 25-30CFG: 3-5Resolutions: SDXL recommended resolutions (832x1216, 896x1152...)Highres fix depending on use caseQuality tags:Positive:<pre><code>masterpiece, best quality, amazing quality, very aesthetic, absurdres, cute face, slender, detailed eyes, detailed hair</code></pre>Negative:<pre><code>lowres, bad quality, worst quality, bad anatomy, sketch, jpeg artifacts, ugly, poorly drawn, blurry, watermark, bad eyes, bright pupils, white pupils, bad hands, bad face, censored, mosaic censoring, bar censor, censor bar, text, signature</code></pre>merged models <a target="_blank" rel="ugc" href="https://civarchive.com/models/1223540/hdarainbowillus">HDA_Rainbow_Illus v1.4</a> <a target="_blank" rel="ugc" href="https://huggingface.co/JujoHotaru/BreedSeriesForXL/blob/main/model/HotaruBreedXL_Anime_v20.safetensors">HotaruBreedXL_Anime_v20</a> <a target="_blank" rel="ugc" href="https://civarchive.com/models/1671563/dpo-full-original-checkpoint">DPO full original checkpoint </a>(for DPO fix)+additional style lorasPermission: no commercial use, no selling, use only on Civitai (please also check the sources for their respective permissions)

Hotaru Blend

Version 4.0 DMD2 This release is the same model as Version 4 with an additional DMD2 lora merged into for faster image generation. The quality is nearly identical to the main model, except that this version has a bit more saturated colors.Recommended settings:Sampler: Euler A + SGM Uniform Steps: 8-12 (recommended 10) CFG: 1.5 - 2 (a bit higher should also work)

Files with SHA256: c8ed1bde5e5dd7ef660558ad34894cf796390a48162cde0b5c924d314b2631bd