Hakushi Mix is designed to be used with various blended art styles. "Hakushi" means "Blank Paper" The quality tags and negative prompts used in the samples are not necessarily the best or only options. Please adjust them as needed to match your preferred style.<h3 id="v1.41">v1.41</h3>Fixed the issue in v1.4 where monochrome or sketch-like images were sometimes generated.In this version, using a CFG scale of 6 or higher may result in overly strong colors depending on the composition. Try starting around CFG: 5 and see how it goes.<h3 id="v1.4">v1.4</h3>Reduced the color unevenness that was noticeable in v1.3.Improved the stability of human figures at high resolutions... maybe.There is a known issue where monochrome images may be generated depending on the prompted composition. This tends to occur when prompts are minimal or lack color specification, though the exact conditions are unclear.Adding "monochrome" to the negative prompt can help reduce this, but it may still occur, so please be aware.<h3 id="v1.3">v1.3</h3>Merged with Illustrious v2.0. High-resolution stability may be improved compared to v1.2, but resolutions like 1344×1792 can still easily cause issues with fingers and limbs. If the results are particularly broken, try lowering the resolution.At higher resolutions, fingers are prone to breaking if not guided properly. Giving the character something to hold or specifying a clear pose tends to improve stability. Limbs are especially likely to break when multiple motion-related prompts are combined (e.g. arms up, dancing, dynamic pose). Extra arms can be suppressed by including "extra arms" in the negative prompt, which also helps reduce the occurrence of extra legs. If legs still break even after that, adding "extra legs" to the negative prompt can help suppress the issue.<h3 id="v1.2">v1.2</h3>Merged with Illustrious v1.1. Additional adjustments were made to enhance detail and enable more delicate, refined strokes.High-resolution output is possible for simple compositions. However, complex poses may lead to visual breakdowns such as extra limbs or duplicated figures, so avoid pushing it too far.Support for natural language prompts has not been thoroughly tested.Be aware that prompts emphasizing the background, such as those including "scenery", tend to make the main subject appear smaller.Also, leaving the background unspecified may result in a simple or plain background.<h3 id="v1.1">v1.1</h3>This version merges the v1.0 model with 25% of the Illustrious v1.0 base model. There is a slight improvement in image quality, but the difference becomes almost unnoticeable when using Hires Fix for upscaling.<h3 id="v1.0">v1.0</h3>A merged model based on Illustrious v0.1. <h3 id="recommended-settings:">Recommended Settings:</h3><ul><li>Steps: 15~35</li><li>CFG Scale: 5~7</li><li>Sampler: Euler a / Euler / DPM++ 2M / DDIM etc</li><li>Schedule: Automatic / SGM Uniform / Simple etc</li><li>Size: 1024x1024, 832x1280, 896x1152, 1280x1280, 1024x1536, 1088x1472, *1536x1536, *1280x1856, *1344x1792 etc*From v1.2 onward (avoid pushing it if results become unstable)</li><li>Additional tools: Use Hires Fix and ADetailer as needed.</li></ul><h2> 日本語</h2>Hakushi Mix は様々な画風をブレンドして使用することを前提として作られています。 また、サンプルで使われている品質タグとネガティブプロンプトは必ずしも正解とは限りません。 好みに合った画風になるよう各自で調整してくだされば幸いです。<h3 id="v1.41">v1.41</h3>1.4でモノクロやラフのような画像が生成されてしまう現象を修正しました。今回はCFGが6以上だと構図によっては発色が強すぎるかもしれません。CFG:5あたりで様子を見てみてください。<h3 id="v1.4">v1.4</h3>1.3で気になっていた色ムラのようなノイズを軽減。高解像度での人体破綻の軽減（たぶん）。指示した構図によってはモノクロの画像が生成されてしまう現象を確認しています。指示が少ない場合や色を指定していない場合に起こりやすい気がしますが、条件は定かではありません。ネガティブに"monochrome"を入れればある程度抑制できますが、それでも生成されてしまう場合があるのでご注意ください。<h3 id="v1.3">v1.3</h3>Illustrious v2.0をマージ。v1.2より高解像度の耐性は上がったかもしれない？依然として1344×1792などの解像度では指や手足が破綻しやすいため、酷い場合は解像度は抑えめにするように。高解像度では、指は何も指示していないと破綻しやすい傾向がある。何かを持たせたり、しっかりポーズを指定すれば安定しやすい。また、手足は複数の指示があると破綻しやすい（例えばarms up, dancing, dynamic poseを組み合わせるような場合）。増えた腕はネガティブプロンプトにextra armsを入れることで抑制でき、ついでに増えた足も抑制される。それでも足が破綻する場合は、extra legsを追加すると抑制できる。<h3 id="v1.2">v1.2</h3>Illustrious v1.1をマージ。その他ディテールの強化を図り、より繊細なタッチを表現できるよう調整。単純な構図であれば高解像度の出力も可能。ただし複雑なポージングになると手足が増えたり、人物が増えたりと破綻しやすいのであまり無茶はしないように。自然言語への対応力は検証不足。sceneryなど背景に重点を置いたプロンプトを入れると、被写体が小さくなりやすい傾向があるので注意。また背景に関して無記入のままだとsimple backgroundになりやすい場合がある。<h3 id="v1.1">v1.1</h3>v1.0にIllustrious v1.0ベースのモデルを25%マージしたもの。 若干画質が向上したような気がする程度で、Hires Fixによるアップスケールを施した場合その差異はほぼ感じられなくなる。<h3 id="v1.0">v1.0</h3>Illustrious v0.1準拠のマージモデル。 <h3 id="recommended-settings:">Recommended Settings:</h3><ul><li>Steps: 15~35</li><li>CFG Scale: 5~7</li><li>Sampler: Euler a / Euler / DPM++ 2M / DDIM 等</li><li>Schedule: Automatic / SGM Uniform / Simple 等</li><li>Size: 1024x1024, 832x1280, 896x1152, 1280x1280, 1024x1536, 1088x1472, *1536x1536, *1280x1856, *1344x1792 等*v1.2以降（破綻する場合は無理をしないこと）</li><li>Additional tools: 必要に応じてHires FixやADetailer</li></ul>My Archive<a target="_blank" rel="ugc" href="https://huggingface.co/tonikakushiroi/HakushiMix_IllustriousXL/tree/main">https://huggingface.co/tonikakushiroi/HakushiMix_IllustriousXL/tree/main</a>

Hakushi Mix

hakushiMix_v141.safetensors

v1.41

v1.4

v1.3

v1.2

v1.1

v1.0

⚠️This is not my model, it's a friendly reprint, so please, check and support the original creator!⚠️Hakushi Mix - Created by <a target="_blank" rel="ugc noopener noreferrer nofollow" href="https://civitai.com/user/tonikakushiroi784">tonikakushiroi784</a> from civitaiCheck it out!:<s>-------------------------------------------------------------------------------------------------------</s><a target="_blank" rel="ugc noopener noreferrer nofollow" href="https://civitai.com/models/1337666?modelVersionId=1856064">https://civitai.com/models/1337666?modelVersionId=1510562</a><s>-------------------------------------------------------------------------------------------------------</s>This model series is absolutely amazing! Beautiful eyes, good background, very consistent, everything is good with this one! Of course one of the best or the best illustrious model ♡good job, tonikakushiroi784 :)(These image previews are in bad quality because the model has a absurd highres and derailment! Unfortunately, tensor is very crappy and don't support them for now :P)<s>-------------------------------------------------------------------------------------------------------</s>DISCLAIMER⚠️I don't earn NOTHING with reprinted models on tensor art! Every models that I upload here are just for accessibility and generation for other users only! :)⚠️(if you are the creator and feels uncomfortable by this friendly reprint, please, claim your model on tensor art discord ^_^)

Files with SHA256: d0caf6453f6d6338c0270aa9aeab678b0b1daa1b5f6cc3b3817874cdff95648a