Recently updated on 12 6 月, 2023 at 05:07 上午
以下作者原文：
🔸🌤️ Hello, this is Ether Real Mix 🌤️🔸
Ether Real Mix is a realistic to semi-realistic anime style model tuned with the flexibility for general purpose use (hopefully). It still has some shortcomings inherit to anime models such as biases in generating females and humans.
✨ Let’s continue to share and create! ✨
UPDATES:
* Ether Real Mix LUX – Version 2
Ether Real Mix LUX 2 is an alternative model that trades realism for vibrancy and style. The fantastically stylish ColorBombMix is added and merge weights were adjusted to compliment a more stylish interpretation of Ether Real.
* Ether Real Mix – Version 1.1
Version 1.1 slightly alters weights to help instances of de-saturated images. Expect a subtle improvement in saturation and vibrancy, with overall style and character proportions staying the same.
Please update if you feel your images were too de-saturated or if you’d like to make them more vibrant.
Do some faces feel weird? Please try adding “realistic” to your prompts to help them feel a bit more comfortable. Or try LUX 2 for a more stylized approach.
* Inpainting Models
1.1 and LUX 2 Inpainting models added. I’m not super familiar with inpainting models as I don’t inpaint much myself. Please try it if it suits your needs.
☄️ Generation Recommendations
Use whatever sampler, steps, cfg you prefer.
If you miss the more vibrant colors of “Ether Blu Mix”, try Ether Real Mix LUX, the VAE I recommend below, or increasing your CFG scale.
Most sample images were generated with:
Sampler: DPM++ 2M Karras
Steps: 28
CFG: 7 – 11
Clip Skip: 1
Higher CFG can lead to more vibrant results.
Negative Prompt Example:
“NSFW, (worst quality, low quality:1.3), watermark, signature”
The sample images were generated using Waifu Diffusion VAE kl-f8-anime2. Please use it if you wish to achieve a similar look.
They were further upscaled using AUTO1111 Hi-res fix.
Latent (nearest-exact)
Upscale by: 2
Denoising strength: 0.55
Are your results changing too much after Hi-res fix? Please try 4x-foolhardy-Remarci.
🖊️ Prompting Advice
Prompt however you like. The model was merged with both Danbooru-style tag prompting as well as natural language in mind. You’re free to mix and match prompting styles, the model should understand it.
Do some faces feel weird? Please try adding “realistic” to your prompts to help them feel a bit more comfortable. Or try LUX 2 for a more stylized approach.
I personally like to use short prompts mixing prose and danbooru-styled tags. Experiment and give the AI some freedom to interpret your ideas!
If you look at the prompts included with the sample images they demonstrate that even with a very loose and vague style of prompting you can achieve fun, high quality images.
General guidelines:
Prompt in the style of anime\danbooru-style models, “masterpiece, best quality, ….. ”
More anime style images
Prompt in the style of photography\non-anime models, “professional photograph of ….. ”
More realistic images, higher style variety
📙 Model Merging Info – v1.1
I have no training or real knowledge of machine learning. I merged the model using the information provided on WarriorMama777’s AOM model page. I very arbitrarily used the UBM weights given to merge and create AOM3 and its derivatives and modified the values with trial and error until I was visually happy with it.
I utilized the SuperMerger extension for AUTO1111 by hako-mikan.
▼ Merging
MODEL A: Counterfeit-V2.5_fp16.safetensors
MODEL B: AOM3.safetensors
Weight Sum merge – Use MBW
Alpha: 1
0,0.9,0.9,0.9,0.9,0.9,0,0,0,0,0.9,0.3,0.9,0.9,0.9,0.9,0.9,0.7,0.3,0.3,0.9,0.9,0.5,0.9,0.9,0.9
Model: CounterfeitV25-AOM3SumS2-09
▼
MODEL A: CounterfeitV25-AOM3SumS2-08
MODEL B: nyanMix_230303.safetensors
MODEL C: pastelmix-better-vae-fp16.safetensors
Weight Sum merge – Use MBW
Alpha: 1
0,0.3,0.3,0.3,0.3,0.3,0,0,0,0,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0,0
Beta: 1
0,0.4,0.4,0.4,0.4,0.4,0,0,0,0,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0.4,0,0
Model: CounterfeitV25-AOM3SumS2-09-Nyan2303SumS2-03-PastelSumS2-05
▼
MODEL A: CounterfeitV25-AOM3SumS2-09-Nyan2303SumS2-03-PastelSumS2-04
MODEL B: ayonimix_V5.safetensors
MODEL C: deliberate_v2.ckpt
Weight Sum merge – Use MBW
Alpha: 1
0,0.8,0.8,0.8,0.8,0.8,0,0,0,0,0.8,0.2,0.8,0.8,0.8,0.8,0.8,0.6,0.2,0.2,0.8,0.8,0.4,0.8,0.8,0.8
Beta: 1
0,0.5,0.5,0.5,0.5,0.5,0,0,0,0,0.5,0.2,0.5,0.5,0.5,0.5,0.5,0.3,0.2,0.2,0.5,0.5,0.2,0.5,0.5,0.5
Final Model: EtherRealMix-v1.1
Models Used:
AOM3.safetensors
D124FC18F0232D7F0A2A70358CDB1288AF9E1EE8596200F50F0936BE59514F6D
Counterfeit-V2.5_fp16.safetensors
A074B8864E31B8681E40DB3DFDE0005DF7B5309FD2A2F592A2CAEE59E4591CAE
deliberate_v2.ckpt
9ABA26ABDFCD46073E0A1D42027A3A3BCC969F562D58A03637BF0A0DED6586C9
pastelmix-better-vae-fp16.safetensors
D01A68AE76F97506363F387F5F28BB564AD9E20924844FD5945E600B72D39E79
ayonimix_V5.safetensors
B46B04432B070FD88D69161B62C43EDCD4340DFC27FFB2E5FE7EDA64950C7E34
nyanMix_230303.safetensors
94885C27E5066745E863CEE2047D06CBC29EC2F4D31ABBDD71BB1F15882AD155
🧡 Credits
Thank you to WarriorMama777 for providing AOM and other various mixes as well as detailing your workflow and inspiring me to try mixing my own models.
Thank you to hako-mikan for creating the SuperMerger extension allowing a quicker workflow for merging.
Thank you to gsdf, rqdwdw for creating Counterfeit.
Thank you to XpucT for creating Deliberate.
Thank you to andite for Pastel-Mix.
Thank you to Nyanbre for Nyan Mix.
Thank you to Ayoni for AyoniMix.
Thank you to devdevice545734 for ColorBombMix.
Thank you to AUTO1111 for creating the web UI everyone uses.
Thank you to StabilityAI for starting everything with Stable Diffusion.
Thank you to the entire SD community for continuing to openly share and create.
原文链接：https://cj.suclub.top/3164.html，转载请注明出处。
请先 ！