|
--- |
|
license: creativeml-openrail-m |
|
library_name: diffusers |
|
tags: |
|
- stable-diffusion |
|
pipeline_tag: text-to-image |
|
--- |
|
# X-mix |
|
**Civitai**: [X-mix | Stable Diffusion Checkpoint | Civitai](https://civitai.com/models/13069/x-mix) |
|
|
|
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 |
|
``` |
|
|
|
%2C%20white%20t.png) |
|
|
|
``` |
|
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 |
|
``` |
|
|
|
%2C%20solo%2C%20Flowery%20meadow%2C%20cloudy%20sky%2C%20aqua%20eyes%2C%20white%20pantyhose%2C%20blonde%20hair%2C.png) |
|
|
|
``` |
|
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 路 Hugging Face](https://huggingface.co/embed/EasyNegative) |
|
|
|
- **bad-artist:** [nick-x-hacker/bad-artist 路 Hugging Face](https://huggingface.co/nick-x-hacker/bad-artist) |
|
|
|
- **bad_prompt_version2:** [embed/bad_prompt 路 Hugging Face](https://huggingface.co/embed/bad_prompt) |
|
|
|
- **Deep Negative V1.x:** [Deep Negative V1.x | Stable Diffusion TextualInversion | Civitai](https://civitai.com/models/4629/deep-negative-v1x) |
|
|
|
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. |
