<h3 id="no-unauthorized-reproduction.-it-affects-my-motivation.-h6gk77knu">No unauthorized reproduction. It affects my motivation.</h3>😭It would be encouraging if you could press the like button.😭 v1.0 = standard fp16 model. / v1.0_fp8 = Low-precision fp8 model. For low-spec PCs. v1.0_hyper8_fp8 = fp8 &amp; hyperSDXL merge model. For low-spec PCs &amp; fast generation. vpred05_v1.5 &amp; vpred06_v1.6 vpred065S_v1.65s = v-pred fp16 model.<h3 id="24112024-update-7bbdg5fhf">24/11/2024 update</h3><ul><li>Note / 日本語記載は下のほうにあります。 <a target="_blank" rel="ugc" href="https://tensor.art/models/800369174514767719/illustrious-model-illusob-xl-mahoroba-mix-vpred065S_v1.65S">It is also available on TensorArt for you to try.</a> No quality tag &amp; negative prompt required. Of course, quality tags would improve results. If noise occurs, reduce the CFG scale gradually or use Hi-Res to eliminate the noise.It was confirmed that this v-pred model does not generate images correctly with “A1111 dev”. We looked for a way to generate images with “reforge” and “comfyui” and correctly generate images with “A1111 dev”, but found various problems such as not being able to use hyper8steps, so we decided not to support “A1111 dev”. The order of the tags should be the same as <a target="_blank" rel="ugc" href="https://civarchive.com/models/795765">illustrious-XL</a> and should be fine. Rating tags have probably lost their functionality. This is not my fault. The corresponding LoRAs remain the same, those that were available can be used, and those that are not cannot be used. When generating dark scenes, the CFG scale must be lowered slightly or the colors may look wrong.Sample images are produced with the following settings. Sampling method : Euler / Schedule type : Simple / CFG Scale : 4-5 / Sampling steps : 20 <a target="_blank" rel="ugc" href="https://civarchive.com/models/428790?modelVersionId=477793">ByteDance Hyper-SDXL - 8 Steps</a> This allows for fast generation with the following settings. Sampling method : Euler / CFG Scale : 1 / Sampling steps : 10<a target="_blank" rel="ugc" href="https://huggingface.co/tianweiy/DMD2">dmd2_sdxl_4step_lora_fp16</a> This allows for fast generation with the following settings. Sampling method : Euler / CFG Scale : 1 / Sampling steps : 5</li></ul><hr /><ul><li>Generation environment reForge version: f1.0.5-v1.10.1RC-latest-801-g62a6cdf1 python: 3.10.11 / torch: 2.3.1+cu121 / xformers: 0.0.27 / gradio: 3.41.2 </li></ul><hr /><ul><li>license I would like to follow the <a target="_blank" rel="ugc" href="https://civarchive.com/models/795765">illustrious-XL</a> license. </li></ul><hr /><ul><li>merge recipe <a target="_blank" rel="ugc" href="https://civarchive.com/models/920730?modelVersionId=1094346">illusob-xl-merge-base-vpred065S_v165s</a> = <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294?modelVersionId=1093948">noobaiXLNAIXL_vPred065SVersion</a> x (1-alpha) + <a target="_blank" rel="ugc" href="https://civarchive.com/models/835655?modelVersionId=1023901">illustriousXLPersonalMerge_v30Noob10based</a> x alpha (0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.0) + (<a target="_blank" rel="ugc" href="https://civarchive.com/models/835655?modelVersionId=1023901">illustriousXLPersonalMerge_v30Noob10based</a> - <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294?modelVersionId=1022833">noobaiXLNAIXL_epsilonPred10Version</a>) x alpha (0.0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.5) + (ExilluNoob05-B_v01 - <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294?modelVersionId=968495">noobaiXLNAIXL_epsilonPred05Version</a>) x alpha (1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,1.0,1.0,1.0) Difference LoRA A = Community pony type Model - <a target="_blank" rel="ugc" href="https://civarchive.com/models/257749?modelVersionId=290640">ponyDiffusionV6XL_v6StartWithThisOne</a> Difference LoRA B = Community animagine type Model - <a target="_blank" rel="ugc" href="https://civarchive.com/models/260267?modelVersionId=403131">animagineXLV31_v31</a> This model = illusob-xl-merge-base-model + Difference LoRA A ( 0,0,0,0,0,0.4,0.4,0.4,0.4,0,0,0 ) + Difference LoRA B ( 0,0,0,0,0,0.4,0.4,0.4,0.4,0,0,0 ) + self-made illustrious LoRA A ( 0,0,0,0,0,0.9,0.9,0.9,0.9,0,0,0 ) + self-made illustrious LoRA B ( 0,0,0,0,0,0.3,0.3,0.3,0.3,0,0,0 ) + self-made Pony LoRA A ( 0,0,0,0,0,1.2,1.2,1.2,1.2,1,1,1 ) </li></ul><hr /><ul><li>メモ <a target="_blank" rel="ugc" href="https://tensor.art/models/800369174514767719/illustrious-model-illusob-xl-mahoroba-mix-vpred065S_v1.65S">TensorArtに置いてありますのでお試し生成ができます。</a> クオリティタグやネガティブプロンプトを使わなくてもいいモデルです。もちろんクオリティタグを使ったほうが結果が良くなる場合があります。 ノイズが発生した場合は、CFGスケールを徐々に小さくするか、ハイレゾを使用してノイズを除去してください。このv-predモデルはA1111 devでは画像が生成できないことを確認しています。 reforgeとcomfyuiで画像が生成でき、A1111 devで画像を生成できる方法を模索しましたがhyper8stepsが使えなくなる等の別の問題が多数発生することになったためA1111 devでは使えないこととします。 タグの順番は <a target="_blank" rel="ugc" href="https://civarchive.com/models/795765">illustrious-XL</a> と同じで問題ないと思います。 どうやらレーティングタグはあまり機能しなくなっています。これは私の仕業ではありません。 対応するLoRAも従来通り、なぜか使えるものは使えるし使えないものは使えません。暗いシーンを生成する場合、CFGスケールを少し下げないと色がおかしく見えることがあります。 サンプル画像の生成設定です。 Sampling method : Euler / Schedule type : Simple / CFG Scale : 5 / Sampling steps : 20 <a target="_blank" rel="ugc" href="https://civarchive.com/models/428790?modelVersionId=477793">ByteDance Hyper-SDXL - 8 Steps</a> を使って高速生成が可能です。 Sampling method : Euler / CFG Scale : 1 / Sampling steps : 10 <a target="_blank" rel="ugc" href="https://huggingface.co/tianweiy/DMD2">dmd2_sdxl_4step_lora_fp16</a> を使って高速生成が可能です。 Sampling method : Euler / CFG Scale : 1 / Sampling steps : 5</li></ul><hr /><ul><li>生成環境 version: f1.0.5-v1.10.1RC-latest-779-ge4ef43fc / python: 3.10.11 / torch: 2.3.1+cu121 xformers: 0.0.27 / gradio: 3.41.2 / checkpoint: 1c5a1cce01</li></ul><hr /><ul><li>ライセンス 大本である illustrious-XL に準じたいと思います。</li></ul><hr />この場を借りて素晴らしいモデルを作られた方々に敬意を表します。 名前を付けるのが苦手なので安直な名前で illustriousnoob から triousno をとり illusob、イルソブとしました。ミックス名はこの絵柄で活動してるあの人の名前を捩った感じです。 ナンバリング規則を変えました、v1.0を基準にv-predの番号を足した数にしてわかりやすくしました。

[illustrious model] illusob-xl-mahoroba-mix (noobai v-pred0.65S merge model)

For low spec PCs.fp8 converted.

illusob-xl-mahoroba-mix_v10_fp8.safetensors

epred_v1.0_fp8

Files with SHA256: 0c9372902af45e041b8666a81525c5064dbfe8328b648c74ff9fd56f1c371f18