Try size: 768 x 1280 or 1280 x 1280
Weights: 0.7~1.0
Pictures from ZImageTurbo and ZImageBase
Note: Some skating moves cannot be easily simulated. It is recommended to try converting images to videos, which may yield better results. A suitable total number of steps is approximately 25, which ensures that mistakes are reduced when the hands and legs meet. The girls' faces will most likely be facing right; after a few tries, they will face left./ 有些溜冰動作不是能輕易模擬出來的，建議嘗試從圖像轉影片，可能效果會更好。適合的總步數大約是25步，能確保手部和腿部交會時，減少失誤。女孩們的臉會高機率面向右方，多試幾次就會面向左方了。(ZImageDeTurbo)
Note: The effect of the girls' aerial movements has been enhanced, but the overall aesthetic appeal is not as good as the ZImageDeTurbo version. If you have enough hard drive space, you might consider downloading both. / 對於女孩們的騰空動作效果做了一些強化，但整體美感度不如ZImageDeTurbo版本。假如硬碟空間還足夠，可以考慮同時下載。 (ZImageTurbo)
Note: Please use natural language and have Gemini write it. There's no way to do the high-difficulty moves; it's mainly about the clothes. Training with "base" was just an experiment, and I didn't feel that it would overwhelmingly replace "turbo", so I do not recommend this version. / 請使用自然語言，並交給Gemini撰寫。高難度動作沒辦法，主要是針對衣服。採用"base"訓練只是一個嘗試，我並沒有感受到"turbo"被它取代的壓倒性，所以我不推薦這個版本。(ZImageBase)
Description
FAQ
Comments (9)
Z-Image Base is pretty good with figure skating posing out of the box without a lora. Might be the better model to finetune figure skating on!
I'll give it a try; training for this is really tiring...
Moreover, natural language is more suitable.
I've tried... "Illustrious" remains the preferred choice for poseture training, while The "base" model offers variety but doesn't provide particularly significant poseture reinforcement. The "base" model cannot perform the "biellmann spin" and "sit spin" action.
@ghostpaint Really? Here is my prompt. Spin works fine for me. https://civitai.com/images/122222851
Key is to describe the pose. ZIB is very prompt adherent. My advice is to take a sit spin image and let gemini caption it for you so you have a good worded description. It should work fine.
@deitychaser Front view and full body Photograph of a pale skin, blonde hair and blue eyes female very petite figure skater performing a high split leap on an ice field. She leaned forward and arched her back. Her hind leg is extended high above her head while her fore leg supports her balance on the ice. Her arms are raised, with her both hands above her head holding her high up raised leg. She has her back arched and her head close to her leg with her head almost touching her raised leg. Her body and her raised leg are parallel to each other. She looked up at the camera. The high ceiling and the audience in the stands were clearly visible.
https://civitai.com/images/122239713
(Thank you. Your prompts from a rear-view perspective have a higher success rate. However, the forward view was rather weak, so I made some changes. And your negative prompts will make the screen turn yellow. It might be because I'm using "Stable Diffusion WebUI Forge - Neo")
@ghostpaint Oh, its because I used a lora for blue hue but it didnt affect the posing. It was a little too blue so I excluded the blue theme for a more neutral look. You can ignore my negative prompt basically. Yeah, side and rear view work good. I think I had some front views too.
How did you do the cool videos btw? Did you train a lora for ice skating on WAN or can wan do this by itself?
@deitychaser They were all made on Pixai Art. https://pixai.art/artwork/1982566737528903229?utm_source=copy_web
If you want me to recreate those difficult poses, I'll have to train separately; it's just a costume. / 如果要你們想還原高難度姿勢，我就必須另外訓練，這只是衣服而已。