SDXL
20
Euler
3
,
all credits to cabal: https://civitai.com/models/2363696
Model DescriptionA continuation of ChenkinRF 0.2For main model description please refer to it.Developed by: Cabal Research (Bluvoll, Anzhc)Compute provided by: Chenkin, HeathcliffLicense: fair-ai-public-license-1.0-sdFinetuned from model: ChenkinNoob-XL-v0.2-Rectified-FlowBias and LimitationsStandard biases and limitations of Danbooru dataset apply, dataset consists of danbooru up to January 2026.Getting Started GuideEnglish中文RecommendationsInferenceComfy(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: 20-28CFG: 3-6Shift: 3-8Schedule: Normal/Simple/SGM Uniform/Beta Positive Quality Tags: masterpiece, best quality, aestheticNegative Tags: worst quality, normal quality, bad anatomy, low resolutionA1111 WebUI(All screenshots are repeating our other RF release, as there is no difference in setup)Recommended WebUI: ReForge - has native support for Flow models, and we've PR'd our native support for Flux2vae-based SDXL modification.How to use in ReForge: (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: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: 20-28CFG: 3-6Shift: 3-8Schedule: Normal/Simple/SGM Uniform/Beta Positive Quality Tags: masterpiece, best quality, aestheticNegative Tags: worst quality, normal quality, bad anatomy, low resolutionADETAILER FIX FOR RF: By default, Adetailer discards Advanced Model Sampling extension, which breaks RF. You need to add AMS to this part of settings:Add: advanced_model_sampling_script,advanced_model_sampling_script_backported to there.If that does not work, go into adetailer extension, find args.py, open it, replace builtinscripts like this:Here is a copypaste for easy copy:_builtin_script = ( "advanced_model_sampling_script", "advanced_model_sampling_script_backported", "hypertile_script", "soft_inpainting", ) Or use this fork of Adetailer - https://github.com/Anzhc/aadetailer-reforgeTrainingTraining DetailsSamples seen(unbatched steps): 52 million samples seen.Learning Rate: 2e-5Effective Batch size: 1152 Effective Batch Size, 36 Batch Size, 4 Gradient Accumulation, 8 GPUsPrecision: Mixed BF16Optimizer: AdamW8bit with Kahan SummationWeight Decay: 0.01Schedule: Constant with warmupTimestep Sampling Strategy: UniformSD3 Shift: 2Text Encoders: FrozenKeep Token: FalseTag Dropout: 10%Uncond Dropout: 10%Shuffle: TrueAdditional Features used: Protected Tags, Cosine Optimal Transport.Training DataDanbooru up to January of 2026.LoRA TrainingPochi.toml is a basic TOML for usage with https://github.com/67372a/LoRA_Easy_Training_Scripts/tree/refresh MAKE SURE TO USE BRANCH REFRESH, comes ready to work.You can also use https://github.com/bluvoll/Akegarasu-lora-scripts-RF/tree/main to train LoRAs or Finetune the model, use Example.toml as a starter configuration for training, or the example in the huggingface repo.HardwareModel was trained on a 8xH100 node.SoftwareCustom fork of SD-Scripts(maintained by Bluvoll)AcknowledgementsThe 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.TestersEveryone in server who tested model throughout it's training and provided feedback, included but not limited to:Shinkuyoinkedlow channelAnzhclylogummySilvelterbrittleDarren LaurieL_A_XNebulaeFranciscoWANGyouhuangztxzhyDracusernian__gao233DUOKai WongRequiredforsomereasonspawnerpeoscrhawawwitterativeNama MTalanMagpieBKM Desu花火流光tairitsujiang1232222kspawner青苇Showcase ImagesItterativeRyushoPanchovixTalanSilvelterDracHardwareChenkin and Heathcliff for providing compute.
