桂花酱 created with SeaArt AI

SDXL (Stable Diffusion XL) and other SDXL-based models (e.g., Juggernaut XL, and many other high-quality Realistic/Fantasy models):

Advantages: Generally very powerful, capable of understanding complex prompts and generating high-quality, detailed images. Most of these models effectively support the use of weights (keyword:value) and recognize artist names for style guidance.
Note: Optimal parameters like CFG Scale may vary slightly among specific models within this group.

High-Quality Realistic Models (often based on SD1.5 or similar architectures):

Advantages: Many models in SeaArt's library fall into this category (e.g., Deliberate, Realistic Vision). They usually understand weights and artist names.
Note: May require more detailed prompts compared to SDXL to achieve similar complexity.

Characteristics: This model might be specially fine-tuned for creating videos or images with a cinematic style.
IMPORTANT NOTE: Based on experience and feedback, this model may not understand or may interpret prompt weights differently from the traditional SD1.5/SDXL manner. Instead of weights, detailed descriptions and keyword repetition might be more effective. The optimal CFG Scale might also be much lower (e.g., in the 1.5 - 4.0 range).

Flux Models (including SeaArt Infinity) and models based on new architectures like SD3.5 (if available on SeaArt):

Characteristics: These are newer model architectures with potentially significantly different prompt processing.
CRITICAL NOTES:

The use of weights (keyword:value) often has no effect with these models.
Artist names in prompts are also often not recognized by these models for style guidance.
CFG Scale often has no effect or is optimized at a fixed value (typically 1.0 for Flux models). Changing this parameter in the interface may not affect the outcome.
To emphasize an element when using these models, you need to focus on detailed descriptions with words, repeating important keywords, or using other advanced prompting techniques specific to that model type.