demoA Story of Detours and Focus.

My journey, guided by my AI technical advisor, 'dot robot', began with an ambitious goal: to create the ultimate anime model by merging various models I love. However, after numerous unsatisfactory merging experiments, I realized that simple blending couldn't forge the unique, soulful identity I was aiming for.

Ultimately, I made a crucial strategic shift: from a 'Fusion Artist' to a 'Specialist Trainer'. I decided to stop pursuing breadth and instead seek depth. I set my sights on a model I believed had excellent potential—Hassaku v1.3 Style A—and performed a fine-tuning on it.

For anyone interested, the entire thought process and project settings are documented in this log: Dot Anime Project Core Settings - v1.1 Update

Core Philosophy: Focus and Refinement

The core of Dot Anime is the fine-tuning of a solid foundation. My goals are:

1. Inherit and Enhance: To fully inherit the excellent qualities and anime style of the base model.

2. Inject a Soul: To imprint my desired 'manga feel, crayon texture, and cel-shading style' more precisely through fine-tuning, creating Dot Anime's unique artistic signature.

3. Stylistic Plasticity: The model retains a high degree of flexibility. You can guide it towards different artistic styles with precise prompts.

Recommended Settings

• Steps: 25-40 (I personally prefer 35 steps for richer detail)

• CFG Scale: 3-5 (A lower CFG can yield a softer, more painterly style)

• Sampler: dpmpp_2m

• Scheduler: sgm_uniform

• Resolution: 1024x1024 or other aspect ratios with an equivalent pixel count is recommended.

Prompting Advice

Note: This version may occasionally generate watermarks. To prevent this, it is highly recommended to include watermark in your negative prompt.

For best results, it's recommended to use basic quality tags.

• Recommended Positive: masterpiece, best quality, highres

• Recommended Negative: (worst quality, low quality, normal quality:1.4), blurry, watermark

Future Plans

This is just the first step. In future versions, I will conduct deeper training based on community feedback and explore more possibilities.

Prompting Advice

Note: This version may occasionally generate watermarks. To prevent this, it is highly recommended to include watermark in your negative prompt.

For best results, it's recommended to use basic quality tags.

• Recommended Positive: masterpiece, best quality, highres

• Recommended Negative: (worst quality, low quality, normal quality:1.4), blurry, watermark

Hires. Fix / Upscaling Advice

Upscaler: I recommend using 4x-Ultrasharp

• Denoising strength: Set this to a range of 0.1 - 0.25.

• Key Tip: A lower denoising strength is crucial for preserving and stabilizing the original line art when upscaling. This helps to avoid blurring the image or causing unwanted stylistic shifts.

License & Use

This model is fine-tuned from Hassaku and therefore fully inherits its license agreement.

All Dot Anime models use the Fair AI Public License 1.0-SD （https://freedevproject.org/faipl-1.0-sd/）.

Thank you to every creator who downloads and tries Dot Anime. I look forward to seeing your amazing creations