This model is a balanced merge of anime and realistic models designed to fully harness the potential of RouWei v0.7 Vpred. By effectively referencing each other's strengths, it aims to expand the range of expressive possibilities.

This model is still in an experimental stage.

It is based on a merge of RouWei v0.7 Vpred and Illustrious v1.0, but it did not pair well with the NoobAI model, which made resolving issues challenging.

Currently, the merging of realistic models is not producing the desired results. Increasing the strength may lead to inconsistencies in the generated images.

Additionally, please note that the accuracy of human body depictions is not a primary focus at this stage, so I appreciate your understanding on this matter.

Why was the old version removed?:

While organizing the workflow, it was discovered that v0.70 mistakenly included a temporary model used during the prototype stage. Unfortunately, this model is not authorized for public release, so I have decided to discontinue its distribution.

Base Model

RouWei v0.7 Vpred + (Illustrious v1.0 - Illustrious v0.1) x 0.5

Recommended Settings

The Vpred settings have been adjusted by adding the v_pred and ztsnr keys to the model. Therefore, it should work without any special configuration if you are using the latest versions of ComfyUI, forge, or reforge.

Resolution

The default resolution is as follows, but can be increased to 0.5-1.5 (TBC).

640x1536
768x1344
832x1216
896x1152
1024x1024
1152x896
1216x832
1344x768
1536x640

Sampler

euler, euler ancestral, euler cfg pp, euler ancestral cfg pp, dpm pp 2s ancestral

Scheduler

karas scheduler doesn't work.

nomal, sgm uniform

CFG: 3-5

Rescale CFG: 0.7

Steps: 20-35

Prompting

General Prompting

Basically, it is not strictly enforced, but as long as you follow the prompt rules of NoobAI and RouWei, there should be no issues.

<1girl/1boy/1other/female/male/...>, <character>, <series>, <artist(s)>, <general tags>, <styling tags>, <adjustment tags>, <quality tags>, <rating tags>

Quality Tags

Quality tags, if used excessively, may prevent the model from fully showcasing its characteristics.

Start with the following tags and adjust them as needed.

# Positive
masterpiece, best quality, highres
# Negative
worst quality, low quality, lowres

If the number of fingers appears inconsistent, you may try adding the following to the Negative Prompt, although it may only provide limited improvement.

extra digits, fewer digits

Styling Tags

This model can be adjusted to lean toward either a realistic or an anime style. However, applying excessively high weights (especially for realistic styles) tends to cause image degradation or collapse.

If you want to apply an anime style, please add the following to the Positive Prompt.

anime screencap, anime coloring

If you want to apply an realistic style, please add the following to the Positive Prompt.

realistic, photorealistic

If you do not want to use realistic backgrounds, please add the following to the Negative Prompt.

photo background

Adjustment Tags

If you want to adjust the brightness of the image, adding one of the following to the Negative Prompt may make it easier to adjust.

It is recommended to start with a lower weight for adjustment.

low brightness / high brightness

If you want to make the image more vibrant or softer, consider adding one of the following to the positive prompt.

hdr / sdr

If you want to make the color tones more vivid or softer, adding one of the following to the Positive Prompt should help achieve the desired effect.

sharp colors / soft colors

Artist Tags

Artist tags can be used in both the traditional (danbooru) style and the RowWei (by ...) format, but please note that there are differences in their strength and effect.

ARTIST or by ARTIST

In general, when using traditional tags, they tend to work in a way that does not interfere too much with other adjustment tags or the overall harmony, but the effect on the art style can be a little more subtle. Mixing several artists can produce interesting effects.

The RouWei style, on the other hand, tends to have a stronger influence on the art style, but it also seems to have a greater impact on elements such as the background and composition.

Rating Tags

A rough classification is as follows

sfw / nsfw

When subdivided in more detail, the classification is as follows

general / sensitive / questionable / explicit

Other Tags

RouWei has many interesting tags that can expand the range of expression. If you're interested, I recommend checking out this page.

Known Issues

Increasing the strength of realistic-related tags too much may cause inconsistencies in clothing and background elements.

Merge Tools

comfy-mecha

Merged Models

I would like to express my deepest respect and gratitude to all the creators who have put in so much time, passion and resources to create such amazing models. Thank you very much.

Base Models

Illustrious v1.0
Illustrious v0.1 (diff base model)
NoobAI-XL v1.1 (Epred) (diff base model)
Pony Diffusion V6 XL (diff base model)
SD XL (diff base model)

Anime Models

SDXL

Pony

ebara pony 3

2DN Pony-v2

NoobAI / Illustrious

Real Models

SDXL

Noob-AI / Illustrious

Note

Since ebara pony 3 is currently not being distributed and its licensing terms remain unclear, I have independently decided to use it based on the following assumptions, along with the fact that other merged models have been publicly released on Civitai.

The license matches that of the upstream model, Pony Diffusion V6, through its merge source, 4th tail
There are no strict restrictions, such as a prohibition on merging

Regardless of the model mentioned above, if there are any license violations due to my oversight, please submit a request, and I will remove the content as soon as possible.

Merge Recipe

The detailed workflow has been uploaded, so please refer to it for more information.

While not an exact representation, the following is a simplified version of the overall structure.

anime-res = rouwei_vpred + a-sdxl + a-pony + a-noob
real-res = rouwei_vpred + r-sdxl + r-noob
final-merge = (rouwei_vpred  + illustrious_v1.0) + (anime-res + real-res)

Licence

This model is licensed under the Illustrious License.

In addition, I take no responsibility for any problems arising from the use of this model or any problems related to the generated images. Please use it at your own discretion.