Shakker-Labs

This repository contains an unified ControlNet for FLUX.1-dev model released by Shakker Labs. We provide an online demo. A FP8 quantized version provided by community can be found in ABDALLALSWAITI/FLUX.1-dev-ControlNet-Union-Pro-2.0-fp8. Keynotes In comparison with Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro, - Remove mode embedding, has smaller model size. - Improve on canny and pose, better control and aesthetics. - Add support for soft edge. Remove support for tile. Model Cards - This ControlNet consists of 6 double blocks and 0 single block. Mode embedding is removed. - We train the model from scratch for 300k steps using a dataset of 20M high-quality general and human images. We train at 512x512 resolution in BFloat16, batch size = 128, learning rate = 2e-5, the guidance is uniformly sampled from [1, 7]. We set the text drop ratio to 0.20. - This model supports multiple control modes, including canny, soft edge, depth, pose, gray. You can use it just as a normal ControlNet. - This model can be jointly used with other ControlNets. Recommended Parameters You can adjust controlnetconditioningscale and controlguidanceend for stronger control and better detail preservation. For better stability, we highly suggest to use detailed prompt, for some cases, multi-conditions help. - Canny: use cv2.Canny, controlnetconditioningscale=0.7, controlguidanceend=0.8. - Soft Edge: use AnylineDetector, controlnetconditioningscale=0.7, controlguidanceend=0.8. - Depth: use depth-anything, controlnetconditioningscale=0.8, controlguidanceend=0.8. - Pose: use DWPose, controlnetconditioningscale=0.9, controlguidanceend=0.65. - Gray: use cv2.cvtColor, controlnetconditioningscale=0.9, controlguidanceend=0.8. Resources - InstantX/FLUX.1-dev-IP-Adapter - InstantX/FLUX.1-dev-Controlnet-Canny - Shakker-Labs/FLUX.1-dev-ControlNet-Depth - Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro Acknowledgements This model is developed by Shakker Labs. The original idea is inspired by xinsir/controlnet-union-sdxl-1.0. All copyright reserved. Citation If you find this project useful in your research, please cite us via

AWPortrait-QW is based on the Qwen-Image. It's trained using a training set that better reflects Chinese facial features and aesthetics. It covers a wide range of genres, including indoor and outdoor portraits, fashion, and studio portraits, ensuring strong generalization. Compared to the original Qwen-Image, AWPortrait-QW delivers a more detailed and realistic rendering of skin texture. No trigger words are requireds. LoRA recommends a weight of 1.0. Acknowledgements This model is trained by our copyrighted users DynamicWang. We release this model under permissions.

This is a LoRA (Logo-Design) trained on FLUX.1-dev by CJim on Shakker AI. You should use `wablogo, logo, Minimalist` as trigger words. The recommended scale is `0.8` in diffusers. Usage suggestion - Dual Combination: something and something, e.g., cat and coffee, - Font Combination: a shape plus a letter, e.g., a book with the word "M," or The fingerprint pattern consists of the letters "hp," - Text Below Graphic: Below the graphic is the word "coffee," directly using with the word "XX" is also feasible You can also download this model at Shakker AI, where we provide an online interface to generate images. Acknowledgements This model is trained by our copyrighted users CJim. We release this model under permissions. The model follows flux-1-dev-non-commercial-license.

This is a LoRA (Vector Journey) trained on FLUX.1-dev for blended realistic illustration by Muertu, the front character is in illustration style, while the background is realistic. You should use `artistic style blends reality and illustration elements.` to trigger the image generation. The recommended scale is `0.8` to `1.0` in diffusers. You can also download this model at Shakker AI, where we provide an online interface to generate images. Acknowledgements This model is trained by our copyrighted users Muertu. We release this model under permissions. The model follows flux-1-dev-non-commercial-license.

The new version of AWPortraitCN2, as an upgraded version of AWPortraitCN, delves deeper into and explores a broader spectrum of Eastern aesthetics. In terms of character portrayal, it now encompasses a wider range of facial data across all age groups. It also responds more effectively to themes such as cuisine, architecture, environments, and a variety of traditional ethnic costumes. Whether you’re drawn to minimalist, poetic landscapes or the vibrant, everyday life of streets and neighborhoods, this update allows for a richer and more intuitive creative journey. .container { display: flex; gap: 10px; / 图片之间的间隔 / } Trigger words No trigger words are requireds. LoRA recommends a weight of 0.9-1. Acknowledgements This model is trained by our copyrighted users DynamicWang. We release this model under permissions. The model follows flux-1-dev-non-commercial-license.

This is a sketch style LoRA trained on FLUX.1-Kontext-dev by yao. Trigger words You should use `convert real photos into sketches`, `sketch` to trigger the image generation. Acknowledgements This model is trained by our copyrighted users yao. We release this model under permissions. The model follows flux-1-dev-non-commercial-license.

This is a flat cartoon style LoRA trained on FLUX.1-Kontext-dev by Mango. You should use `Convert to a flat cartoon style while keeping the subject unchanged` to trigger the image generation. Acknowledgements This model is trained by our copyrighted users Mango. We release this model under permissions. The model follows flux-1-dev-non-commercial-license.

This is an illustration style style LoRA trained on FLUX.1-Kontext-dev by zooey. You should use `Turned into a plush felt toy` to trigger the image generation. Acknowledgements This model is trained by our copyrighted users zooey. We release this model under permissions. The model follows flux-1-dev-non-commercial-license.

This is an illustration style style LoRA trained on FLUX.1-Kontext-dev by 木木木亘. You should use `illustration style` to trigger the image generation. Acknowledgements This model is trained by our copyrighted users 木木木亘. We release this model under permissions. The model follows flux-1-dev-non-commercial-license.