X-mix

Civitai: X-mix | Stable Diffusion Checkpoint | Civitai

X-mix is a merging model used to generate anime images. My English is not very good, so there may be some parts of this article that are unclear.

V2.0

V2.0 is a merged model based on V1.0. This model supports nsfw.

Difference from V1.0

• The performance of V2.0 is not better than that of V1.0, but the generated images now exhibit a different artistic style.
• V2.0 offers better support for nsfw than V1.0, but the drawback is that even when you do not intend to generate an nsfw image, there is still a possibility of generating one. If you are more interested in the sfw model, I will provide a detailed explanation in the recipe section.
• In my opinion, V2.0 is not as user-friendly as V1.0, and it appears to be more challenging to generate an excellent image.

Recommended Settings

• Sampler: DPM++ SDE Karras (sfw), DDIM (nsfw)
• Steps: 20 (DDIM may require more steps)
• CFG Scale: 5
• Hires upscale: Latent (bicubic antialiased), Latent (nearest-exact), Denoising strength: 0.4~0.7
• vae: NAI.vae
• Clip skip: 2
• ENSD: 31337
• Eta: 0.67

Example

masterpiece, best quality, ultra-detailed, illustration, portrait, 1girl
Negative prompt: EasyNegative, photograph by bad-artist, bad_prompt_version2, DeepNegative-V1-75T
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 4291846267, Size: 512x512, Model hash: 7bc4c05c90, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (nearest-exact), Eta: 0.67

Indoor, bright, 1Girl, gray hair, amber eyes, smile, black dress, barefoot, sitting posture,
Negative prompt: EasyNegative, by bad-artist, bad_prompt_version2
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 2118045521, Size: 600x400, Model hash: 7961a4960e, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (bicubic antialiased), Eta: 0.67

landscape, in spring, cherry blossoms, cloudy sky, 1girl, solo, long blue hair, smirk, pink eyes, (school uniform:1.05), white thighhighs,
Negative prompt: EasyNegative, by bad-artist, bad_prompt_version2, DeepNegative-V1-75T
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 3093571233, Size: 400x600, Model hash: 7961a4960e, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (bicubic antialiased), Eta: 0.67

1girl, on bed, wet, see-through shirt, thighhighs, cleavage, collarbone, full body,
Negative prompt: EasyNegative, photograph by bad-artist, bad_prompt_version2, DeepNegative-V1-75T
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 986400693, Size: 512x512, Model hash: 7961a4960e, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (bicubic antialiased), Eta: 0.67

Alice \(Alice in wonderland\), solo, Flowery meadow, cloudy sky, aqua eyes, white pantyhose, blonde hair,
Negative prompt: EasyNegative, photograph by bad-artist, bad_prompt_version2, DeepNegative-V1-75T
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 273840053, Size: 512x512, Model hash: 7961a4960e, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (bicubic antialiased), Eta: 0.67

masterpiece, best quality, ultra-detailed, illustration, portrait, hakurei reimu, 1girl, throne room, dimly lit
Negative prompt: EasyNegative, by bad-artist, bad_prompt_version2, DeepNegative-V1-75T
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 2212365348, Size: 512x512, Model hash: 7961a4960e, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (nearest-exact), Eta: 0.67

masterpiece, best quality, ultra-detailed, illustration, 1girl, witch hat, purple eyes, blonde hair, wielding a purple staff blasting purple energy, purple beam, purple effects, dragons, chaos
Negative prompt: EasyNegative, photograph by bad-artist, bad_prompt_version2, DeepNegative-V1-75T
Steps: 20, Sampler: DDIM, CFG scale: 5, Seed: 293615512, Size: 512x512, Model hash: 7961a4960e, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (nearest-exact)

1girl, solo, black skirt, blue eyes, electric guitar, guitar, headphones, holding, holding plectrum, instrument, long hair, , music, one side up, pink hair, playing guitar, pleated skirt, black shirt, indoors
Negative prompt: EasyNegative, photograph by bad-artist, bad_prompt_version2, DeepNegative-V1-75T
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 3442031040, Size: 512x512, Model hash: 7961a4960e, Denoising strength: 0.6, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (nearest-exact), Eta: 0.67

Recipe

Step 1: animefull-latest (model) + pastelmix-lora (lora) + ligneClaireStyleCogecha (lora) = pastel-Cogecha

You can try replacing animefull-latest with Anything-V3.0 or your preferred model. However, I cannot confirm if this will yield better results and it requires you to experiment with it on your own.

Step 2: MBW: Chilloutmix + X-mix-V1.0

Model A	Model B	base_alpha	Weight	Merge Name
Chilloutmix	X-mix-V1.0	1	1,1,1,1,1,1,1,1,0,0,0,0,1,0,0,0,0,1,1,1,1,1,1,1,1	X-mix-V2.0-base

This is the step of the sfw version. The steps for the nsfw version are as follows: I merged several LoRAs into Chilloutmix to obtain Chilloutmix-nsfw. Then I merged Chilloutmix-nsfw and X-mix-V1.0 to get X-mix-V2.0-nsfwBase1. Finally, I merged several LoRAs into X-mix-V2.0-nsfwBase1 to get X-mix-V2.0-nsfwBase2.

LoRAs related to real people should be merged into Chilloutmix or other photo-realistic models that you like, while LoRAs related to anime should be merged into X-mix-V2.0-base. Which LoRAs to use depends on your preference.

Step 3: MBW: pastel-Cogecha + X-mix-V2.0-base

Model A	Model B	base_alpha	Weight	Merge Name
pastel-Cogecha	X-mix-V2.0-base	0	1,1,1,1,1,0.3,0,0,0,1,0.1,1,1,1,1,1,0,1,0,1,1,0.2,1,1,1	X-mix-V2.0-sfw

In fact, I never tried to obtain the sfw version because I didn't plan on using it from the beginning. So this process is for reference only, and I am not sure about the actual effect of the sfw model.

V1.0

I have forgotten the recipe for X-mix-V1.0, as too many models were used for merging. This model supports nsfw, but the effect may not be very good.

Recommended Settings

• Sampler: DPM++ SDE Karras
• Steps: 20
• CFG Scale: 5
• Hires upscaler: Latent (bicubic antialiased), Denoising strength: 0.5~0.6
• vae: NAI.vae
• Clip skip: 2
• ENSD: 31337
• Eta: 0.67

Examples

masterpiece, best quality, ultra-detailed, illustration, portrait, 1girl
Negative prompt: EasyNegative, by bad-artist, bad_prompt_version2
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 1906918205, Size: 512x512, Model hash: 7bc4c05c90, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (bicubic antialiased), Eta: 0.67

Indoor, bright, 1girl, gray hair, amber eyes, smile, black dress, barefoot, sitting posture,
Negative prompt: EasyNegative, by bad-artist, bad_prompt_version2
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 2118045521, Size: 600x400, Model hash: 7bc4c05c90, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (bicubic antialiased), Eta: 0.67

landscape, in spring, cherry blossoms, cloudy sky, 1girl, solo, long blue hair, smirk, pink eyes, (school uniform:1.05), white thighhighs,
Negative prompt: EasyNegative, by bad-artist, bad_prompt_version2
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 3093571233, Size: 400x600, Model hash: 7bc4c05c90, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (bicubic antialiased), Eta: 0.67

1girl, on bed, wet, see-through shirt, thighhighs, cleavage, collarbone, full body,
Negative prompt: EasyNegative, photograph by bad-artist, bad_prompt_version2
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 1666118295, Size: 512x512, Model hash: 7bc4c05c90, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (bicubic antialiased), Eta: 0.67

Alice \(Alice in wonderland\), solo, Flowery meadow, cloudy sky, aqua eyes, white pantyhose, blonde hair,
Negative prompt: EasyNegative, sketch by bad-artist
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 807449917, Size: 512x512, Model hash: 7bc4c05c90, Denoising strength: 0.5, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (nearest-exact), Eta: 0.67

masterpiece, best quality, ultra-detailed, illustration, portrait, hakurei reimu, 1girl, throne room, dimly lit
Negative prompt: EasyNegative, by bad-artist, bad_prompt_version2
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 116927034, Size: 512x512, Model hash: 7bc4c05c90, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (bicubic antialiased), Eta: 0.67

masterpiece, best quality, ultra-detailed, illustration, 1girl, witch hat, purple eyes, blonde hair, wielding a purple staff blasting purple energy, purple beam, purple effects, dragons, chaos
Negative prompt: EasyNegative, photograph by bad-artist, bad_prompt_version2
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 1705759664, Size: 512x512, Model hash: 7bc4c05c90, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (bicubic antialiased), Eta: 0.67

1girl, solo, black skirt, blue eyes, electric guitar, guitar, headphones, holding, holding plectrum, instrument, long hair, , music, one side up, pink hair, playing guitar, pleated skirt, black shirt, indoors
Negative prompt: EasyNegative, by bad-artist, bad_prompt_version2
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 5, Seed: 2548407675, Size: 512x512, Model hash: 7bc4c05c90, Denoising strength: 0.55, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires upscaler: Latent (bicubic antialiased), Eta: 0.67

Embedding

If you need the embedding used in examples, click them:

• EasyNegative: embed/EasyNegative · Hugging Face

• bad-artist: nick-x-hacker/bad-artist · Hugging Face

• bad_prompt_version2: embed/bad_prompt · Hugging Face

• Deep Negative V1.x: Deep Negative V1.x | Stable Diffusion TextualInversion | Civitai

You can consider whether to use them according to your preferences.

More

1. Since my prompts are usually brief, I'm not sure if this model will be able to meet all of your requirements if you need to use a large number of prompts.
2. Using low resolution is not recommended for generating pictures.
3. I did my best, but the hands are not perfect.
4. The above settings may not necessarily be perfect.
5. Due to my computer's performance, it's difficult for me to comprehensively test this model. I'm looking forward to your feedback.