<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/267c3cdf-62ed-4843-bd62-8cc6cbcb2eba/original=true/267c3cdf-62ed-4843-bd62-8cc6cbcb2eba.jpeg" /><h3 id="model-description">Model Description</h3>A continuation of <a target="_blank" rel="ugc" href="https://huggingface.co/ChenkinRF/ChenkinNoob-XL-v0.2-Rectified-Flow">ChenkinRF 0.2</a>For main model description please refer to it.<ul><li>Developed by: <a target="_blank" rel="ugc" href="https://huggingface.co/CabalResearch">Cabal Research</a> (Bluvoll, Anzhc)</li><li>Compute provided by: Chenkin, Heathcliff</li><li>License: <a target="_blank" rel="ugc" href="https://freedevproject.org/faipl-1.0-sd/">fair-ai-public-license-1.0-sd</a></li><li>Finetuned from model: <a target="_blank" rel="ugc" href="https://huggingface.co/ChenkinRF/ChenkinNoob-XL-v0.2-Rectified-Flow">ChenkinNoob-XL-v0.2-Rectified-Flow</a></li></ul><h2 id="bias-and-limitations">Bias and Limitations</h2>Standard biases and limitations of Danbooru dataset apply, dataset consists of danbooru up to January 2026.<h3 id="getting-started-guide">Getting Started Guide</h3><ul><li><a target="_blank" rel="ugc" href="https://www.notion.so/Chenkin-Noob-XL-RF-User-Guide-2fc03a034c1f80c18c00ffc93d4ad2a4?source=copy_link">English</a></li><li><a target="_blank" rel="ugc" href="https://my.feishu.cn/wiki/ITBKwIoD5itZXukljjacDvc2nYd?from=from_copylin_">中文</a></li></ul><h2></h2><h1 id="recommendations">Recommendations</h1><h3></h3><h3 id="inference">Inference</h3>Comfy<a target="_blank" rel="ugc" href="https://cdn-uploads.huggingface.co/production/uploads/633b43d29fe04b13f46c8988/7vuzsT-ANUL-3mIXXIjbM.png"><img src="https://cdn-uploads.huggingface.co/production/uploads/633b43d29fe04b13f46c8988/7vuzsT-ANUL-3mIXXIjbM.png" alt="image" /></a>(Workflow is available alongside model in repo)Same as your normal inference, but with addition of SD3 sampling node, as this model is Flow-based.Recommended Parameters: Sampler: Euler, DPM++ SDE, etc. Steps: 20-28 CFG: 3-6 Shift: 3-8 Schedule: Normal/Simple/SGM Uniform/Beta Positive Quality Tags: <code>masterpiece, best quality, aesthetic</code>Negative Tags: <code>worst quality, normal quality, bad anatomy, low resolution</code>A1111 WebUI(All screenshots are repeating our other RF release, as there is no difference in setup)Recommended WebUI: <a target="_blank" rel="ugc" href="https://github.com/Panchovix/stable-diffusion-webui-reForge">ReForge</a> - has native support for Flow models, and we've PR'd our native support for Flux2vae-based SDXL modification.How to use in ReForge:<a target="_blank" rel="ugc" href="https://cdn-uploads.huggingface.co/production/uploads/633b43d29fe04b13f46c8988/UV5Yp66H7YlccdQqborPf.png"><img src="https://cdn-uploads.huggingface.co/production/uploads/633b43d29fe04b13f46c8988/UV5Yp66H7YlccdQqborPf.png" alt="изображение" /></a> (ignore Sigma max field at the top, this is not used in RF)Support for RF in ReForge is being implemented through a built-in extension:<a target="_blank" rel="ugc" href="https://cdn-uploads.huggingface.co/production/uploads/633b43d29fe04b13f46c8988/LpMF0lmC96X001Au9fFU_.png"><img src="https://cdn-uploads.huggingface.co/production/uploads/633b43d29fe04b13f46c8988/LpMF0lmC96X001Au9fFU_.png" alt="изображение" /></a><a target="_blank" rel="ugc" href="https://cdn-uploads.huggingface.co/production/uploads/634cf025fe861cc73a2e7dd0/siEBfX41E6D_DUEqaXMgs.png"><img src="https://cdn-uploads.huggingface.co/production/uploads/634cf025fe861cc73a2e7dd0/siEBfX41E6D_DUEqaXMgs.png" alt="imagen" /></a>Set parameters to that, and you're good to go.Recommended Parameters: Sampler: Euler Comfy, Euler, DPM++ SDE Comfy, etc. ALL VARIANTS MUST BE RF OR COMFY, IF AVAILABLE. In ComfyUI routing is automatic, but not in the case of WebUI. Steps: 20-28 CFG: 3-6 Shift: 3-8 Schedule: Normal/Simple/SGM Uniform/Beta Positive Quality Tags: <code>masterpiece, best quality, aesthetic</code> Negative Tags: <code>worst quality, normal quality, bad anatomy, low resolution</code>ADETAILER FIX FOR RF: By default, Adetailer discards Advanced Model Sampling extension, which breaks RF. You need to add AMS to this part of settings:<a target="_blank" rel="ugc" href="https://cdn-uploads.huggingface.co/production/uploads/633b43d29fe04b13f46c8988/RQMtfm5Xi3V7oNsqXoZJN.png"><img src="https://cdn-uploads.huggingface.co/production/uploads/633b43d29fe04b13f46c8988/RQMtfm5Xi3V7oNsqXoZJN.png" alt="изображение" /></a>Add: <code>advanced_model_sampling_script,advanced_model_sampling_script_backported</code> to there.If that does not work, go into adetailer extension, find <a target="_blank" rel="ugc" href="http://args.py">args.py</a>, open it, replace builtinscripts like this:<a target="_blank" rel="ugc" href="https://cdn-uploads.huggingface.co/production/uploads/633b43d29fe04b13f46c8988/rmnS-i_kciJzTZmeR-mGP.png"><img src="https://cdn-uploads.huggingface.co/production/uploads/633b43d29fe04b13f46c8988/rmnS-i_kciJzTZmeR-mGP.png" alt="изображение" /></a>Here is a copypaste for easy copy:<pre><code>_builtin_script = (
 "advanced_model_sampling_script",
 "advanced_model_sampling_script_backported",
 "hypertile_script",
 "soft_inpainting",
)
</code></pre>Or use this fork of Adetailer - <a target="_blank" rel="ugc" href="https://github.com/Anzhc/aadetailer-reforge">https://github.com/Anzhc/aadetailer-reforge</a><h2></h2><h2 id="training">Training</h2><h3 id="training-details">Training Details</h3>Samples seen(unbatched steps): 52 million samples seen. Learning Rate: 2e-5 Effective Batch size: 1152 Effective Batch Size, 36 Batch Size, 4 Gradient Accumulation, 8 GPUs Precision: Mixed BF16 Optimizer: AdamW8bit with Kahan Summation Weight Decay: 0.01 Schedule: Constant with warmup Timestep Sampling Strategy: Uniform SD3 Shift: 2 Text Encoders: Frozen Keep Token: False Tag Dropout: 10% Uncond Dropout: 10% Shuffle: TrueAdditional Features used: Protected Tags, Cosine Optimal Transport.<h3 id="training-data">Training Data</h3>Danbooru up to January of 2026.<h3 id="lora-training">LoRA Training</h3>Pochi.toml is a basic TOML for usage with <a target="_blank" rel="ugc" href="https://github.com/67372a/LoRA_Easy_Training_Scripts/tree/refresh">https://github.com/67372a/LoRA_Easy_Training_Scripts/tree/refresh</a> MAKE SURE TO USE BRANCH REFRESH, comes ready to work.You can also use <a target="_blank" rel="ugc" href="https://github.com/bluvoll/Akegarasu-lora-scripts-RF/tree/main">https://github.com/bluvoll/Akegarasu-lora-scripts-RF/tree/main</a> to train LoRAs or Finetune the model, use Example.toml as a starter configuration for training, or the example in the huggingface repo.<h3 id="hardware">Hardware</h3>Model was trained on a 8xH100 node.<h3 id="software">Software</h3>Custom fork of <a target="_blank" rel="ugc" href="https://github.com/bluvoll/sd-scripts">SD-Scripts</a>(maintained by Bluvoll)<h2 id="acknowledgements">Acknowledgements</h2>The model is still overcoming the anatomy issues first seen in ChenkinNoobXL 0.2 Epsilon and the change caused by deprecated tags in danbooru 2025, at this point in time the model has become far sharper and detailed than expected, some newer characters are promptable with helper features, we expect this to improve over the next 5 or 7 epochs as we raise LR to 4e-5 due to the high batch size we run.<h3 id="testers">Testers</h3>Everyone in server who tested model throughout it's training and provided feedback, included but not limited to:<ul><li>Shinku</li><li>yoinked</li><li>low channel</li><li>Anzhc</li><li>lylogummy</li><li>Silvelter</li><li>brittle</li><li>Darren Laurie</li><li>L_A_X</li><li>Nebulae</li><li>Francisco</li><li>WANG</li><li>youhuang</li><li>ztxzhy</li><li>Drac</li><li>user</li><li>nian__gao233</li><li>DUO</li><li>Kai Wong</li><li>Requiredforsomereason</li><li>spawner</li><li>peoscrha</li><li>waww</li><li>itterative</li><li>Nama M</li><li>Talan</li><li>Magpie</li><li>BKM Desu</li><li>花火流光</li><li>tairitsujiang</li><li>123</li><li>2222k</li><li>spawner</li><li>青苇</li></ul><h3 id="showcase-images">Showcase Images</h3><ul><li>Itterative</li><li>Ryusho</li><li>Panchovix</li><li>Talan</li><li>Silvelter</li><li>Drac</li></ul><h3 id="hardware">Hardware</h3>Chenkin and Heathcliff for providing compute.

ChenkinNoob-XL Rectified-Flow

5 epochs of Additional trainingThe model is still overcoming the anatomy issues first seen in ChenkinNoobXL 0.2 Epsilon and the change caused by deprecated tags in danbooru 2025, at this point in time the model has become far sharper and detailed than expected, some newer characters are promptable with helper features, we expect this to improve over the next 5 or 7 epochs as we raise LR to 4e-5 due to the high batch size we run.

Files with SHA256: 54751e7a5b58e5afb176c0bcfe7b013ba4d2d48550d9be113f69eb42e541c206