File size: 5,463 Bytes
b0cc2da 0841660 b0cc2da c855e43 b0cc2da 0841660 c855e43 29a3ce3 c855e43 29a3ce3 c855e43 29a3ce3 c855e43 29a3ce3 c855e43 29a3ce3 c855e43 29a3ce3 c855e43 29a3ce3 c855e43 29a3ce3 c855e43 29a3ce3 c855e43 29a3ce3 c855e43 29a3ce3 c855e43 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 |
---
license: creativeml-openrail-m
---
# Eleet Model: an anime style stable diffusion model
> This model is also available on: [Eleet Model - Civitai](https://civitai.com/models/77807).
Eleet model is a block-weighted merged stable diffusion model aiming at generating good quality 2D anime style images that satisfy my personal taste and, hopefully, your taste.
ELEET is an Internet slang for the word 'elite', which dates back to 80s/90s. Its modified spelling in a digital form is 31337, which has been the most common Eta Noise Seed Delta (ENSD) value in stable diffusion probably since the NAI era. The model is named in honor of this tradition.
- [Suggested settings](#suggested-settings)
- [Samples](#samples)
- [Merge ideas](#merge-ideas)
- [License](#license)
## Suggested settings
For users who know little about stable diffusion settings, I recommend:
* **Prompt**: Start with `masterpiece, best quality, aesthetic`. Personally I also like to add photography phrases such as `cinematic lighting, professional shadow`, etc.
* **Negative prompt**:
> (worst quality, low quality:1.4), lowres, bad anatomy, (blurry), (text, logo, watermark, signature, username)
* **Sampler**: DPM++ 2M Karras
* **CFG Scale**: 6\~9
* **Steps**: 16\~30
* Highres Fix (optional): Latent sampler; 0.6~0.7 Denoising strength; 16 Highres steps.
* **Clip skip**: 2
* No external VAE needed
## Samples
Here are 4 samples of the latest Eleet model version.
*Sample 1* (Highres Fix from 512x800):
```
masterpiece, best quality, aesthetic, 1girl, solo, black eyes, green hair, low-tied long hair, school uniform, sitting, wariza, thighhighs, black thighhighs, from above, looking at viewer, (cityscape), cinematic lighting, professional shadow
```
Other common settings through all samples:
```
Negative prompt: (worst quality, low quality:1.4), lowres, bad anatomy, (child, loli), (blurry), (text, logo, watermark, signature, username)
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Denoising strength: 0.6, Clip skip: 2, Hires upscale: 1.25, Hires steps: 16, Hires upscaler: Latent
```
*Sample 2* (Highres Fix from 512x800):
```
masterpiece, best quality, aesthetic, 1girl, solo, red eyes, one eye closed, brown hair, long hair, grin, arms up, crop top, denim shorts, midriff, navel, athletic body, earings, (cowboy shot), looking at viewer, (waterfall), sunny, cinematic lighting, professional shadow
```
*Sample 3* (Highres Fix from 832x512):
```
masterpiece, best quality, aesthetic, no humans, scenery, sky, cloud, outdoors, mountain, water, sunset, cloudy sky, lake, river, landscape, reflection, tree, nature, mountainous horizon, blue sky, evening, professional shadow, award winner photo
```
*Sample 4* (Highres Fix from 800x512):
```
masterpiece, best quality, aesthetic, forest, grass, blue sky, cloud, outdoors, no humans, nature, scenery, railroad tracks, sunlight, sunbeam, lens flare, cinematic lighting, professional shadow
```
Additionally, here are samples of the previous model versions. Click to expand:
<details>
<summary>Eleet v1.0 samples</summary>
*Sample 1*: Scenery. txt2img+highres, (640x384) x1.5.
```
masterpiece, best quality, aesthetic, highres RAW photo, landscape photography, wide shot, from below, scenery, sunrise, blue sky, clouds, lake, reflection, sun, trees, floating leaves, ripples, foreground interest, depth of field, cinematic lighting, asymmetric composition, professional shadows, sharp focus, lens flare
```
*Sample 2*: Anime girl. txt2img, 576x832.
```
masterpiece, best quality, aesthetic, 1girl, :d, blue eyes, fox ears, gold hair, twin braids, crop top, off-shoulder jacket, blue pleated skirt, thighhighs, garter belt, skindentation, breasts, bare shoulders, midriff, (cowboy shot), looking at viewer, waterfall, mountains
```
*Sample 3*: Scenery. txt2img, 832x576.
```
masterpiece, best quality, aesthetic, highres RAW photo, cool color tone, wide shot, scenery, snow, blue sky, cityscape, skyline, buildings, rooftop, roads, cinematic lighting, professional shadows, sharp focus
```
*Sample 4*: Anime girl. txt2img+highres, (384x640) x1.5.
```
masterpiece, best quality, 1girl, solo, angry, brown eyes, blue hair, bangs, sidelocks, half updo, clenched hand, parted lips, [bodysuit|armored dress], long sleeves, gloves, armor, looking to the side, clenched teeth, (cowboy shot), night, full moon, forest, underexposure, professional lighting
```
</details>
## Merge ideas
The weights for merging Eleet model were optimized through an automatic procedure with scoring, but I didn't necessarily pick the best-scored one as the final version. Instead, I will evaluate a number of high-scored candidates and score their outputs manually by myself.
I conducted the evaluation mostly on anime girls topics (no doubt) but also considered the model performance on scenery images. I will consider:
* Prompt response
* Image aesthetic quality over 3 scenarios:
1. txt2img under `--lowvram` mode on a low-end GPU
2. txt2img under normal or `--medvram` mode on a better GPU
3. txt2img + highres fix on the previous GPU
* Image flaws (color shift, illogical drawing, etc.) over the above scenarios
## License
CreativeML OpenRAIL-M.
|