1. Complete re-merge, all merged models are "Checkpoint Trained" models, and no "CHECKPOINT MERGE" models are merged.There is no merging of any lora.
2. Model selection: high-quality models, high-resolution models, or models with a large amount of training data.
3. Better understand natural language and danbooru tags.
4.The model does not contain vae.
5.You can try to generate the image at 1024 resolution. At 1024 resolution, it will be easier to generate good fingers. To generate standing full body images, it is also recommended to generate highly 1024 resolution images.
6.The model does not include VAE. If VAE is not used, the image generated will be darker. We recommend "vae-ft-mse-840000-ema-pruned" (brighter color) or orangemix.vae (lighter color).
2024/10/06
V6:
1.完全的重新合并，合并的模型皆为”培训(Checkpoint Trained)”的模型，并免合并混和”(CHECKPOINT MERGE)”的模型。没有任何lora的合并。
2.模型挑选高品质模型、高解析度的模型、或是具有大量培训数据的模型。
3.更好的理解自然语言与danbooru标签。
4.模型不包含vae.
5.你可以尝试生成1024 解析度的图片。 在1024解析度时，将更容易生成良好手指。要生成站立的全身图片，也建议生成高度为1024解析度的图片.
6.模型不包括 VAE，如果没使用VAE生成的图片会较灰暗，推荐"vae-ft-mse-840000-ema-pruned"(颜色较鲜艳)或orangemix.vae (颜色较浅)。
--
1. Complete re-merge, all merged models are "Checkpoint Trained" models, and no "CHECKPOINT MERGE" models are merged.There is no merging of any lora.
2. Model selection: high-quality models, high-resolution models, or models with a large amount of training data.
3. Better understand natural language and danbooru tags.
4.The model does not contain vae.
5.You can try to generate the image at 1024 resolution. At 1024 resolution, it will be easier to generate good fingers. To generate standing full body images, it is also recommended to generate highly 1024 resolution images.
6.The model does not include VAE. If VAE is not used, the image generated will be darker. We recommend "vae-ft-mse-840000-ema-pruned" (brighter color) or orangemix.vae (lighter color).