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.
■1024px_model
●Additionally, due to merging with NAI v2 and Dora fine-tuning, the 1024px model has largely solved the problems associated with the 512px model. It features high resolution and much better tag adherence.Therefore, none of the drawbacks typically pointed out regarding SD1.5 apply to this specific model.
Personally, I see it as a game-changer that significantly extends the limits of SD1.5. I strongly suggest giving it a try.
●That being said, even at 1024px, fine details like the eyes are often insufficient. I highly recommend upscaling further using i2i.
●kohya_deep_shrink is also effective for t2i, so it might be a good idea to try using it.
Doing so can sometimes reduce the breakdown of backgrounds and fingers, leading to more stable results.
■512px_model
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.
■Please feel free to ask if you have any questions!
日本語での質問も大丈夫ですので気軽にお声がけください!
Description
■I trained Dora on a dataset of 400,000 images.
This one is intended for Cascade 3.6B.
■Training Data (52.58 KB):It also includes a README with my inference workflow, OneTrainer training settings, and tips. Looking through it should give you a deeper understanding of Cascade.
■This Dora is primarily designed to be applied to the Cascade 3.6B base model, but I believe it can also be used with models that have been fine-tuned by others.
■The Dora has been trained at multiple resolutions: 768, 1024, 1280, and 1536px.
Because of this, it allows for very flexible inference. In my workflow, it enables image generation at 3200×4800px.
■Achieving this resolution was possible thanks to the extensions, models, and workflows created by ClownsharkBatwing. He is a dedicated explorer of Cascade, and without his work, I wouldn’t have been able to complete this training. I want to take this opportunity to express my gratitude.
Of course, this Dora can also be used with the standard Cascade workflow without any extensions. However, I personally prefer ClownsharkBatwing’s approach, as I find it more intuitive.
■Beyond that, there are many other contributors to Cascade. Thanks to their efforts, I am still able to continue experimenting with inference and training. I deeply appreciate everyone who has shared their remarkable results in exploring Cascade’s potential—such as sote, R35ON4NCE, ultrapixel, resonance_lite_r1, and others.
■This Dora was trained on Danbooru tags.
For inference, I recommend using tipo for automatic tag generation, as it tends to produce stable and high-quality results.
Although character tags are included in the training, they were not heavily emphasized. So please consider it more like: popular characters may be possible to generate, but it’s not guaranteed.
Pruned Model bf16 (6.2 GB):This is a DoRA backup from an intermediate stage.
In some cases, the intermediate version might produce better results, so feel free to give it a try.
Training Data (52.58 KB):Inference workflow, training settings, and tips