šµAITKcine V2 takes your creative projects to the next level with a vastly expanded range of art styles and capabilities. In this version, the model has been trained on 10,000 images from movies, incorporating a variety of model merges to enhance its versatility. Whether you're aiming for cinematic realism, digital art, anime, or illustrations, AITKcine V2 delivers outstanding results across all mediums.
Optimal Settings:
These settings are my personal preference, and I encourage you to test the model with the settings you prefer to find what works best for your specific needs. the model works on all samplers. i have been messing around with DEIS lately and this modle delivers incredibly coherent result with the settings below.
Sampler: New DEIS sampler with DDIM uniform scheduler
Clip Skip: 3
CFG: 3.5
Steps: 20-40
Special Features:
No Trigger Words: While there are no specific trigger words, you need to clearly state the medium you want (e.g., cinematic, photography, anime, illustration).
Negative Prompts: Typically requires minimal negative prompts. Start with none and gradually add what you don't want to see. The deep^negative embedding is a good starting point.
High-Resolution Tips: For maximum detail, use highres fix on A1111 (denoising range: 0.4 to 0.6). On ComfyUI, upscale latent by 1.5 and run it into another sampler with a denoise range of 0.5 to 0.65.
Additional Notes:
Word Economy: The model doesn't respond well to "word salads" and unnecessary terms. Keep prompts clear and concise for best results.
Literal Prompts: The model responds well to literal prompts, meaning that describing a scene in detail with natural language is better than a succession of keywords. There might be a great balance between these two ways of prompting.
Realism Settings: For realism, the recommended settings (DEIS sampler and DDIM uniform) work exceptionally well.
šµAITKcine V2: Expanding Your Artistic Horizons
šµAITKcine V2 takes your creative projects to the next level with a vastly expanded range of art styles and capabilities. In this version, the model has been trained on 10,000 images from movies, incorporating a variety of model merges to enhance its versatility. Whether you're aiming for cinematic realism, digital art, anime, or illustrations, AITKcine V2 delivers outstanding results across all mediums.
Optimal Settings:
These settings are my personal preference, and I encourage you to test the model with the settings you prefer to find what works best for your specific needs. the model works on all samplers. i have been messing around with DEIS lately and this modle delivers incredibly coherent result with the settings below.
Sampler: New DEIS sampler with DDIM uniform scheduler
Clip Skip: 3
CFG: 3.5
Steps: 20-40
Special Features:
No Trigger Words: While there are no specific trigger words, you need to clearly state the medium you want (e.g., cinematic, photography, anime, illustration).
Negative Prompts: Typically requires minimal negative prompts. Start with none and gradually add what you don't want to see. The deep^negative embedding is a good starting point.
High-Resolution Tips: For maximum detail, use highres fix on A1111 (denoising range: 0.4 to 0.6). On ComfyUI, upscale latent by 1.5 and run it into another sampler with a denoise range of 0.5 to 0.65.
Additional Notes:
Word Economy: The model doesn't respond well to "word salads" and unnecessary terms. Keep prompts clear and concise for best results.
Literal Prompts: The model responds well to literal prompts, meaning that describing a scene in detail with natural language is better than a succession of keywords. There might be a great balance between these two ways of prompting.
Realism Settings: For realism, the recommended settings (DEIS sampler and DDIM uniform) work exceptionally well.