Copax TimeLess
Copax TimeLess optimizes image generation (1024x1360 resolution) by providing optimal settings and prompt engineering techniques to avoid common issues like anatomical errors and blurring.
English
English
This document summarizes the optimal configuration parameters for generating 1024x1360 resolution images, as well as the setting method of positive and negative sample prompt words. Designed to help users obtain high-quality image output and avoid problems such as anatomical errors, hand deformities, and blur.
Recommended settings:
Generation steps: 30
CFG scaling: 3.5~7 (early versions) or 5~7 (later versions)
Sampler: Euler a (early versions) or Euler a + ADetailer face_yolov8n/s.pt (later versions)
Positive sample prompt words:
score_9, score_8_up, score_7_up (indicates image quality with a score of 9, 8 or 7 or above)
source_real, real skin (indicates real character skin)
Negative sample prompt words:
worst quality, low quality (meaning the worst quality or low quality image)
worst detail, low detail (an image with poor detail)
source_furry, source_pony, source_cartoon, 2d (indicates furry, cartoon or two-dimensional images)
bad anatomy, bad hands (an image indicating anatomical errors or hand deformities)
text, error, missing fingers, fewer digits (indicates an image containing text, error, missing fingers, or fewer digits)
cropped, bad artist, outlines, jpeg artifacts (images that are cropped, poorly painted, have outlines or jpeg compression artifacts)
signature, watermark, username (indicates an image containing a signature, watermark, or username scrambled)
blurry, 3d (indicating a blurry or three-dimensional image)
Update history:
V14: Adjusted the overall color, composition and background, and fine-tuned the positive and negative sample prompt words.
V13: Mainly adjusted the color tone and data, the character's face has basically remained unchanged, and stabilized the CFG at 7.
V12: The recommended CFG scaling ratio is 5~7, and a 12 step version of the setting is also provided.
V11: ADetailer face_yolov8s.pt is introduced for detail optimization.
V10: Adjusted the prompt words for positive and negative samples to emphasize the skin texture of real people.
V2-V9: The recommended CFG scaling ratio is 7, and is optimized using ADetailer face_yolov8n/s.pt.
V1: Initial version, can use Euler a or DPM++ 2M Karras sampler.
Things to note:
There may be subtle differences between different versions, please choose the appropriate configuration according to your needs.
The role of the positive and negative sample prompt words is to guide the generator to generate images that match the description. Appropriate adjustments can improve the quality of the generation.
Always use high-definition images as input to generate more detailed images.
Check whether the network connection is stable, try using a proxy or mirror source; confirm whether you need to log in to your account or provide an API key. If the path or version is wrong, the download will fail.
Make sure you have installed the correct version of the framework, check the version of the dependent libraries required by the model, and update the relevant libraries or switch the supported framework version if necessary.
Use a local cache model to avoid repeated downloads; or switch to a lighter model and optimize the storage path and reading method.
Enable GPU or TPU acceleration, use batch data processing methods, or choose a lightweight model such as MobileNet to increase speed.
Try quantizing the model or using gradient checkpointing to reduce the memory requirements. You can also use distributed computing to spread the task across multiple devices.
Check whether the input data format is correct, whether the preprocessing method matching the model is in place, and if necessary, fine-tune the model to adapt to specific tasks.
