Dedede

🚀 DeDeDe - Anime-style Stable Diffusion Model

DeDeDe is a Stable Diffusion model adjusted to easily generate anime-style characters. It combines multiple models and is fine - tuned with a large number of images, offering high - quality and diverse anime - style image generation capabilities.

🚀 Quick Start

If you can read English, please refer here.

When using DeDeDe, we recommend the following Prompt/Negative Prompt:

• Prompt: best quality, masterpiece
• Negative Prompt: 3d, flat shading, flat color, retro style, 1980s, 1990s, 2000s, 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, inaccurate limb

✨ Features

• Anime - style Focus: Specialized in generating anime - style characters, with clear details and vivid colors.
• Multiple Model Merging: Merges multiple high - quality models, such as DreamLike Diffusion 1.0, Trinart Characters v2 Derrida, and DreamLike Photoreal 1.0, to enrich the generation style.
• Large - scale Tuning: Tuned with 30,000 images from various sources, including SD2.1, Novel AI, WD1.3/1.4, CoolJapan Diffusion 2.1, and Dreamlike Photoreal 2.0, to improve the generation quality.

💻 Usage Examples

Basic Usage

Here are some examples of using DeDeDe to generate images:

Example 1

(((best quality, masterpiece, 8k))), detailed anime style of 1girl sitting in room and reading book wearing school uniform and wavy detailed pink hair pink and detailed yellow eye yellow, smiling
Negative prompt: [[3d]], (((flat shading, flat color))), retro style, 1980s, 1990s, 2000s, 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, inaccurate limb
Steps: 25, Sampler: Euler a, CFG scale: 8, Seed: 2277801742, Size: 512x768, Model hash: 6d1729a039, Denoising strength: 0.75, Clip skip: 2, ENSD: 31337, Hires upscale: 1.5, Hires upscaler: Latent

Example 2

(((best quality, masterpiece, 8k))), detailed anime style of anime 1girl bust shot sitting and dipping in river and wetty wearing white transparent onepiece dress with detailed wavy pink hair pink and hetailed yellow eye yellow, water splash in gorgeous scene secret garden
Negative prompt: [[[3d]]], (((flat shading, flat color))), retro style, 1980s, 1990s, 2000s, 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, inaccurate limb
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 8, Seed: 1117106368, Size: 512x768, Model hash: 6d1729a039, Denoising strength: 0.7, Clip skip: 2, ENSD: 31337, Hires resize: 768x1152, Hires steps: 5, Hires upscaler: Latent

🔧 Technical Details

Model Merging and Training Steps

The following symbol strings next to the models are the hash values of the ckpt files displayed when selecting the model at the time of Automatic1111 Webui commit hash c98cb0f8ecc904666f47684e238dd022039ca16f.

1. Difference Merging of Dreamlike Diffusion 1.0 and Trinart Derrida | Interpolation Method | Primary Model | Secondary Model | Tertiary Model | Merge Name | | --- | --- | --- | --- | --- | | Add Difference @ 1.0 | DreamLike Diffusion 1.0(0aecbcfa2c) | TrinArt Characters v2 Derrida(42d3f359b0) | Stable Diffusion 1.4(fe4efff1e1) | DDD_pre1(d1ac03017b) |

2. Hierarchical Merging of IN00 - IN05 in DDD_pre1 with Dreamlike Photoreal 1.0 | Model: A | Model: B | Weight | Base alpha | Merge Name | | --- | --- | --- | --- | --- | | DDD_pre1(d1ac03017b) | Dreamlike Photoreal 1.0(f403e4e2a5) | 0.45,0.45,0.4,0.35,0.3,0.25,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0 | 0 | DDD_pre2(601ec74593) |

3. Fine - Tuning of DDD_pre2 DDD_pre2 is fine - tuned with self - prepared material images from the outputs of other Diffusion Models. The services/models used for preparation are SD2.1, Novel AI, WD1.3/1.4, CoolJapan Diffusion 2.1, and Dreamlike Photoreal 2.0. A total of about 30,000 images (including flipped ones) are used, with a learning rate of 5e - 6 and 60,000 steps of training. The generated model is named DDD_pre3(4709475652).

4. Weighted Average Merging of DDD_pre3 and DDD_pre2 | Interpolation Method | Primary Model | Secondary Model | Merge Name | | --- | --- | --- | --- | | Weighted Sum @ 0.5 | DDD_pre3(4709475652) | DDD_pre2(601ec74593) | DeDeDe(6d1729a039) |

Model Flavors

DeDeDe_ip2p_0.7_0.8.ckpt/DeDeDe_ip2p_0.7_1.0.ckpt

These models are obtained by extracting and adding task vectors from the Instruct pix2pix model. Their sizes are DeDeDe 0.8/Instruct Pix2Pix 0.7 and DeDeDe 1.0/Instruct Pix2Pix 0.7 respectively. The license inherited from Instruct Pix2Pix is as follows: Copyright (c) 2023 Ren Nakayama Released under the MIT license https://huggingface.co/nakayama/DeDeDe/blob/main/MIT - License

DeDeDe_controlnet_.pth/DeDeDe_webui_controlnet_.safetensors

These are models obtained by merging DeDeDe with the models used in ControlNet. Please refer to here for the specific method.

DeDeDeP

https://huggingface.co/nakayama/DeDeDeP DeDeDeP is edited based on DeDeDe using hierarchical merging with Dreamlike Photoreal 1.0. Compared with DeDeDe, its output is more realistic.

📄 License

This model is under the influence of Dreamlike Diffusion 1.0 / Dreamlike Photoreal 1.0, so the modified CreativeML OpenRAIL - M license of the above models applies. The following is the Japanese translation of the modified part translated by DeepL, but the English language takes precedence in interpretation.

• You cannot host or use this model or its derivatives on websites/apps/others where you earn or plan to earn income or donations. If you want to do so, please email contact@dreamlike.art.
• You can freely host the model card and files (without actual inference or fine - tuning) on commercial and non - commercial websites/apps/others. Please specify the full model name (Dreamlike Diffusion 1.0 / Dreamlike Photoreal 1.0) and include a link to the model card (https://huggingface.co/dreamlike - art/dreamlike - diffusion - 1.0 / https://huggingface.co/dreamlike - art/dreamlike - photoreal - 1.0/).
• You can freely host the model or its derivatives on completely non - commercial websites or apps (meaning no revenue or donations are earned). Please specify the full model name (Dreamlike Diffusion 1.0 / Dreamlike Photoreal 1.0) and attach a link to the model card (https://huggingface.co/dreamlike - art/dreamlike - diffusion - 1.0 / https://huggingface.co/dreamlike - art/dreamlike - photoreal - 1.0/).
• A team of up to 10 people can freely use the output of the model or its derivatives for commercial purposes.
• You cannot use this model to intentionally create or share illegal or harmful outputs or content.
• The author makes no claims to the rights of the outputs you generate. You can use them freely and are responsible for their use, which must not violate the provisions stipulated in the license.
• You can redistribute the weights. When redistributing, please note that you must share a copy of the modified CreativeML OpenRAIL - M with all users, including the same usage restrictions as in the license (please read the license fully and carefully). You can view the full license here: https://huggingface.co/nakayama/DeDeDe/blob/main/License.md

💡 Usage Tip

When uploading the generated works to SNS or other platforms, if there are features like tags, please add #DeDeDeArt. It will be great if you can do so as I'll check them out.