Midjourney风格迁移至LoRA工作流

1- Create a 3D texture ball containing the needed style. 2- Put it as image references with iw 0.6. 3- Add prompt to create first grid, check if modification on image reference weight is needed; if good, continue. 4- Take best result from grid to add as sref, then create again until finding amazing results to split between image references and sref. 5- Play with sw + iw to enhance and modify prompt for better results. 6- If style is captured, take image as style reference + keep image reference the same, enhance outputs using s+c+w and other parameters. 7- Create final version of prompt with desired outputs, adding q 2 to enhance final output. 8- Take reference and create more images, check outputs one by one to ensure all are good for dataset. 9- Enhance final outputs using upscalers for a good start point. 10- Filter results to ensure they match the style. 11- Put final set on fal training model, wait 3 minutes to finalize data uploading, check results by prompting. 12- Share new LoRA with remix to share with community.