[ Stable Diffusion web UI ] Introduction of SD1.5 Recommended Checkpoint Model
![[ Stable Diffusion web UI ] Introduction of SD1.5 Recommended Checkpoint Model featured Image](/_next/image?url=https%3A%2F%2Fdca.data-hub-center.com%2Fcontent%2Fuploads%2F2024%2F05%2Feyecatch-sd15-recommended-checkpoint-models-en.jpg&w=3840&q=80)
This article introduces recommended SD1.5 checkpoint models for use with Stable Diffusion web UI. While SDXL is all the rage, there are still plenty of models that can produce high-quality images by using ControlNet’s Tiling technology, for example. Let’s take a look at illustrative, 2.5D, and photorealistic checkpoint models along with reference images.
Please refer to this article for instructions on how to install the model.
Illustrative checkpoint models
First, we will introduce the SD1.5 illustration-based checkpoint models. As you can see from the sample images, each model defines the basis for the illustration style; you can use techniques such as Lora to adjust the illustration style, but it is easier to generate images without Lora to produce images with less noise.
Dark Sushi Mix 2.25D
- Creator: 🔗Aitasai
- Download: 1.99 GB
- Upload: 2023/6/11
- Data Format: Pruned Model fp16
Recommended setting
VAE: vae-ft-mse-840000-ema
Highres. Fix: onThis is “Dark Sushi Mix Oroshita Sushi Mix – 2.25D” by Aitasai, which we introduced in our previous article. According to the author, 2.25D means “between 2D and 2.5D,” which makes the model a little more realistic. The detailed depiction of women and backgrounds in the model is a recommended checkpoint model.
ShiratakiMix
- Creator: 🔗Vsukiyaki
- Download: 2.13 GB
- Upload: 2023/4/7
- Data Format: fp16
Recommended setting
Sampler: DPM++ SDE Karras
Steps: 20 ~ 60
CFG scale: 7.5
Hires steps: 20
Hires Denoising strength: 0.55
Hires upscaler: Latent, R-ESRGAN 4x+ Anime6B
Clip skip: 2
Textual Inversion: easynegative(easynegative:1.0),(worst quality,low quality:1.2),(bad anatomy:1.4),(realistic:1.1),nose,lips,adult,fat,sad, (inaccurate limb:1.2),extra digit,fewer digits,six fingers,(monochrome:0.95)This is “ShiratakiMix” by Vsukiyaki, which we also introduced in our previous article. The model is very colorful and detailed. The file includes a flat model, a model with a built-in VAE, and a fixed model that has been further restored.
Counterfeit-V3.0
- Creator: 🔗rqdwdw
- Download: 3.95 GB (fp16) / 5.28 GB (fp32)
- Upload: 2023/4/29
- Data Format: Pruned Model fp16/fp32
Recommended setting
Recommend using EasyNegative
“Counterfeit-V3.0” by rqdwdw is a popular checkpoint model in CIBITAI and has been downloaded more than 300,000 times. There are two download files, fp32 and fp16, so download the one that suits your environment. If you are not sure, download the Fp16 version.
MeinaMix
- Creator: 🔗Meina
- Download: 1.99GB
- Upload: 2023/7/17
- Data Format: Pruned Model fp16
Recommended setting
Sampler: DPM++ SDE Karras: 20 ~ 30
Sampler: DPM++ 2M Karras: 20 ~ 60
Sampler: Euler a: 40 ~ 60
CFG Scale: 4 ~ 11
Resolutions: 512x768 Portrait
Resolutions: 768x512 Landscape
Hires steps: 10
Hires upscaler: R-ESRGAN / 4x+Anime6b
Hires Denoising strength: 0.3 ~ 0.6
Clip Skip: 2(worst quality, low quality:1.4), (zombie, sketch, interlocked fingers, comic)(worst quality:1.6, low quality:1.6), (zombie, sketch, interlocked fingers, comic)Meina’s “MeinaMix” is another checkpoint model that has been downloaded over 300,000 times on cibitai. The model is trained to produce high quality images with simple pronto. It is recommended to enable K samplers quantization in the settings. (It is located in the “Stable Diffusion” field in the settings.)
Aniflatmix
- Creator: 🔗CyberAIchemist
- Download: 1.99 GB
- Upload: 2023/4/9
- Data Format: Pruned Model fp16
Recommended setting
VAE: Orangemix, Anything V4.5, NAI
Sampler: DPM++ 2M Karras
Sampling steps: 20
Negative embedding: EasyNegative、badhandv4Recommend using badhandv4
This is CyberAIchemist’s “Aniflatmix” flat paint model. ligne claire gives it a “ligne claire technique” style. You can also use lineart or monochrome to create a cartoon style. If you do not want a monochrome style, you can put it in the negative prompt.
Milky Wonderland
- Creator: 🔗technobyte
- Download: 1.99 GB
- Upload: 2024/4/16
- Data Format: Pruned Model fp16
Recommended setting
There was no specific description, but We will introduce the negative embedding used in the sample.
“Milky Wonderland” by technobyte features illustrations with a soft touch. The background and hand drawings are beautifully rendered.
NeverWinter Mix
- Creator: 🔗neverwinter13
- Download: 1.99 GB
- Upload: 2024/4/9
- Data Format: Pruned Model fp16
Recommended setting
Sampler: DPM++ 2M Karras, Euler Ancestral(Euler A)
Sampling steps: 20 ~ 28
CFG Scale: 5 ~ 7“NeverWinter Mix” by neverwinter13 is a little more realistic, but We thought We had not even mentioned 2.5D, so here it is. The amount of writing produces a beautiful picture.
Kakarot 2.5D ChiChi
- Creator: 🔗vay_kakarot
- Download: 2.14 GB
- Upload: 2024/4/21
- Data Format: Pruned Model fp16
Recommended setting
Sampler: DPM++ 3M SDE Karras, DPM++ 2M Karras, Euler a Karras
Steps: 20 or 80,
CFG scale: 7
Hires steps: 20
Hires Denoising strength: 0.4
Hires upscaler: 4x-UltraSharp
Clip skip: 2
VAE: vae-ft-mse-840000-ema-pruned
Textual Inversion: EasyNegative, badhandv4, bad-picture-chill-75v, bad_prompt, negative_hand-neg ng_deepnegative_v1_75t BadNegAnatomyV1-neg“Kakarot 2.5D ChiChi” by vay_kakarot can generate women with originality. This model has 2.5D in the title, but “Kakarot 2.8D,” which will be introduced later, has more of a 2.5D feel, so We will introduce it here.
2.5D Illustrative checkpoint models
Next, we introduce the 2.5D Illustrative checkpoint models of SD1.5. The term “2.5D” refers to the intermediate between 2D and 3D, and produces more realistic images than pure 2D.
AniVerse
- Creator: 🔗Samael1976
- Download: 1.99 GB
- Upload: 2024/4/4
- Data Format: Pruned Model fp16
Recommended setting
Sampler: DPM++ 2M SDE Karras
CFG Scale: 3 -> Steps: 15
CFG Scale: 4 -> Steps: 20
CFG Scale: 5 -> Steps: 25
CFG Scale: 6 -> Steps: 30
Hires upscaler: R-ESRGAN 4x + Anime6B
Hires upscale: 1.5
Hires steps: 20~30
Hires Denoising strength: 0.4
Clip skip: 2
Textual Inversion: negative_hand-neg, bad-hands-5(worst quality, low quality), negative_hand-neg, bad-hands-5, naked, nude, braless, cross, sepia, black&white, B&W, painting, drawing, illustrationAdetailer recommended
Samael1976’s “AniVerse” generates high quality 2.5D type images. According to the author, it is better to de-emphasize the Negative Prompt for this model. He also recommends using the Adetailer extension.
RealCartoon3D
- Creator: 🔗7whitefire7
- Download: 1.99 GB
- Upload: 2024/3/16
- Data Format: Pruned Model fp16
Recommended setting
Sampler: DPM++ SDE Karra, DPM++ 2M Karras
Steps: 40
CFG scale: 7
Hires steps: 20
Hires Denoising strength: 0.35
Hires upscaler: R-ESRGAN 4x+, 4k-UltraSharp
Clip skip: 2
Textual Inversion: easynegative, badhandv4easynegative,(badhandv4),(bad quality:1.3),(worst quality:1.3),watermark,(blurry),5-funny-looking-fingers“RealCartoon3D” by 7whitefire7 is a popular model with over 150,000 downloads on Citibai. It is updated frequently. The full model is available for download, but even the pruned model produces a good enough quality output.
CamelliaMIx_2.5D
- Creator: 🔗Mods13
- Download: 1.99 GB
- Upload: 2024/9/14
- Data Format: Pruned Model fp16
Recommended setting
Sampler: DPM++2M Karras, DPM++ SDE Karras, DPM++ 3M SDE Karras
Steps:
CFG scale: 6 ~ 11
Hires steps: 13
Hires upscaler: 4x-UltreaSharp
Clip skip: 2
VAE: ft-mse-840000, orangemix.vae.pt
Textual Inversion: negative_hand-neg, verybadimagenegative (worst quality, low quality:1.3)Mods13’s “CamelliaMIx_2.5D” can generate relatively more Asian faces, and recommends using negative_hand-neg.
Dark Sushi 2.5D
- Creator: 🔗Aitasai
- Download: 2.24 GB
- Upload: 2023/8/16
- Data Format: Pruned Model fp16
Recommended setting
Hires. Fix: on
VAE: vae-ft-mse-840000-emaThis is a 2.5D version of “Dark Sushi 2.5D” by Aitasai. Personally, We don’t think it is 2.5D, but since it is said to be 2.5D, We will describe it here.
Featureless 2.5D Mix
- Creator: 🔗pawapawa
- Download: 2.14 GB
- Upload: 2023/10/28
- Data Format: Full Model fp16
pawapawa’s “Featureless 2.5D Mix” is a mix of 2.5D series models. The data is only full models. The quality of the generated images is also very good.
Kakarot 2.8D
- Creator: 🔗vay_kakarot
- Download: 2.14 GB
- Upload: 2024/4/21
- Data Format: Pruned Model fp16
Recommended setting
Sampler: DPM++ 3M SDE Karras, DPM++ 2M Karras, Euler a Karras
Steps: 20 or 80,
CFG scale: 7
Hires steps: 20
Hires Denoising strength: 0.4
Hires upscaler: 4x-UltraSharp
Clip skip: 2
VAE: vae-ft-mse-840000-ema-pruned
Textual Inversion: EasyNegative, badhandv4, bad-picture-chill-75v, bad_prompt, negative_hand-neg ng_deepnegative_v1_75t BadNegAnatomyV1-neg“Kakarot 2.8D” by vay_kakarot is a more realistic model of the illustration-based “Kakarot 2.5D”.
Photorealistic checkpoint models
Let’s move on to photorealistic checkpoint models. Depending on the checkpoint model, some models have a limited range of expression. Some models can generate illustration-style images using prompts, for example.
majicMIX realistic
- Creator: 🔗Merjic
- Download: 1.99 GB
- Upload: 2023/10/5
- Data Format: Pruned Model fp16
Recommended setting
Sampler: Euler a, Euler
Steps: 20~40
Hires steps: 15+
Hires Denoising strength: 0.05~0.5
Hires upscale: 2+
Hires upscaler: ESRGAN 4x, 4x-UltraSharp, 8x_NMKD-Superscale_150000_G
Clip skip: 2Merjic’s “majicMIX realistic” is a live-action type model, so fantasy-like expressions will be difficult to achieve. He specializes in generating Asian people.
DreamShaper
- Creator: 🔗Lykon
- Download: 1.99 GB
- Upload: 2023/7/30
- Data Format: Pruned Model fp16
Lykon’s “DreamShaper” was created as an alternative to MidJourney. It has a wide range of expression, including fantasy expression. Depending on the pronto, illustration and 2.5D expressions are also possible.
XXMix_9realistic
- Creator: 🔗Zyx_xx
- Download: 2.24 GB
- Upload: 2023/1/23
- Data Format: Pruned Model fp16
Recommended setting
Sampler: DPM++ 2M Karras, DPM++ SDE Karras
Steps: 20 ~ 40
CFG scale: 6 ~ 8
Hires steps: 15
Hires Denoising strength: 0.2 ~ 0.5
Hires upscaler: 4x-UltraSharp
Clip skip: 2
Textual Inversion: “XXMix_9realistic” by Zyx_xx is a popular live-action model that has been downloaded more than 120,000 times on Citibai. It is capable of producing beautiful women with fantasy and other expressions.
asian_brm(beautyrealmix)
- Creator: 🔗topitopi
- Download: 1.99 GB
- Upload: 2024/4/20
- Data Format: Pruned Model fp16
Recommended setting
Sampler: DPM++ 2M Karras
Steps: 35
CFG scale: 6.5 ~ 7
Hires steps: 10
Hires Denoising strength: 0.3 ~ 0.4
Hires upscaler: 4x-UltraSharp
Clip skip: 2
VAE: vae-ft-mse-840000-ema-pruned
ADetailer: mediapipe_face_full“asian_brm(beautyrealmix)” by topitopi can create beautiful Asian people. It is photo-realistic, but the background may be more like an illustration.
Conclusion
This article has been written to introduce the SD1.5 model checkpoint model. As you explore each model, please note the recommended prompt parameters and licensing information. If you plan to use these models for purposes other than personal use, be aware of the license restrictions.
Thank you for reading to the end.
