This is my experiment in creating an SD1.5 merge style model for i2i.I am satisfied with the current state, but I will update it if I feel anything is needed.or I might share ways to enhance this model rather than updating it directly.
■Since the model now has both anime and real versions, the detailed explanations have been moved to each model’s tab.
Both are merged models created by selecting multiple high-quality models with minimal artifacts.
■With the three models—asian, real, and anime—now available, it could be fun to adjust their mix to find your ideal style.
●asian 0.5 + real 0.5 might yield a more mixed, half-and-half look.
●asian 0.5 + anime 0.5 might produce a cute, 2.5D-style appearance.
Feel free to experiment with different ratios.
■Since this is just a merge, it shares a common SD1.5 limitation where NSFW tags may not be fully understood or followed.
I have decided to manage the concept-enhancing LoRA separately.
https://civarchive.com/models/1253884/sd15loralab
Of course, it can be used on its own, It is designed for i2i processing the models below.
https://civarchive.com/models/505948/pixart-sigma-1024px512px-animetune
■Depending on the situation, this extension may also improve colors and contrast.
https://github.com/Haoming02/sd-webui-diffusion-cg
https://github.com/Haoming02/comfyui-diffusion-cg
■Using external tools for level adjustment is also a good option.
Reducing gamma slightly while enhancing whites can improve contrast even further.
Using these should help achieve color rendering closer to that of SDXL.
■Surprisingly, generating at 768px or 1024px sometimes works fine.If you want more stability, merging with Sotemix could help.But since most LoRAs are trained at 512px, high resolutions can break the output.So it’s safer to use highres.fix or kohya_deep_shrink when using LoRAs.
Personally, I prefer i2i upscaling over highres.fix, as it tends to produce fewer artifacts.
Description
■This is a 1024px merged model for SD1.5.
The concept is a model capable of generating flexible yet stable quality. It combines the rich tag knowledge and high-resolution capabilities of NovelAI_v2 with the stable style and quality of SD1.5 anime models.
Pruned Model fp16 (13.24 GB):Shared Merge Materials
Training Data (23.77 KB):comfyui workflow
Pruned Model bf16 (570.31 MB):High-Resolution Doras Only.
■This model can be used immediately by anyone with the same user experience as a standard SD1.5 model. There is no need to swap the VAE; it uses the novelai_v2_vae, ensuring vivid results. For tags, you can use the same ones as previous models. Additionally, NovelAI_v2 tags should also work. Regarding NovelAI_v2, I have prepared tag rules and other information on my page, so please check it out.
https://civitai.com/models/1772131/novelaidiffusionv2
■Please set Clip Skip: 2.
■The base resolution is 1024px.
It can generate stably up to 1024x1536. 1152x1728 is also possible, though slightly unstable. Increasing the resolution will improve composition and make fine details sharper.
■With i2i, you can use almost limitless resolutions; 2048x3072 provides a good balance of speed and quality. Using i2i at this resolution allows for the generation of highly detailed images.
■I have also provided an inference workflow for your reference.
The "tipo workflow" is especially recommended, as it can automatically generate tags, making inference easy.
■To create this merge, I selected and merged high-quality models, and also created components like a "Dora" that enables 1024px resolution for SD1.5 models and stabilizes the quality and resolution of NovelAI_v2.
I have prepared materials for both anime and realistic styles. This model is built from a combination of these elements. My merge model aims for a somewhat balanced style, merging anime and realistic models at approximately an 8:2 ratio.
■I am sharing all of these merge materials as well.
If you are interested in high-resolution SD1.5 merging, please try it yourself.
There are countless SD1.5 models, so the combinations are infinite. You can likely achieve various customizations, such as creating a more stable merge than mine, enabling even higher resolutions, or specializing in a specific style. You could also create high-resolution versions based on your own favorite models. If you create a great merged model, please consider uploading it!
■If you combine only the anime-style materials, the result will be even more anime-like. I am not skilled at training realistic-style materials, so their quality isn't very high, but depending on how they are used, you might be able to create a high-resolution realistic model. Combining the anime and realistic materials can also produce a semi-realistic style.
■I have also provided my merge workflow, so please use it as a reference to see how I applied the "Dora" and performed the merge. Of course, I am sure there are better merging methods. Depending on the technique, an even better model might be created.
■If you have any questions about the materials or inference, I will do my best to provide as much information as possible, so please feel free to ask.