This is a lora with wan2.2 special effects
The prompt can be written like this: "xuanzhuan. The video shows a child model, wearing a green Polo shirt, beige cargo pants, a belt, a top and sports shoes, placed on the ground." This children's model has undergone a 360-degree rotation movement."
I uploaded my model and trial application on runninghub. You can try them out on the website first or download them directly for a trial.
This website address is a trial one
https://www.runninghub.cn/ai-detail/1955834690592804866/?inviteCode=rh-v1059
Global friends can use the following links to trial in advance: https://www.runninghub.ai/ai-detail/1956168462395736066?inviteCode=rh-v1059
Fan benefits
If you haven't registered for this website yet, you can get 1000RH coins and generate many pictures and videos for free by registering for the first time through my link! With 4090-24G of video memory, you only need 0.2RH coins per second. Additionally, you get 100RH coins for daily login. Enjoy playing with 4090! The platform also features 4090-48GB of video memory, allowing you to experience the charm of super power!
💖 show your support! 💖
If you like this model, don't forget to give it a thumbs up and leave a comment - your feedback helps improve the project! In addition, please also give the talented creators who use this model to release their works the appreciation they deserve to support them.
这是一款wan2.2特效lora
提示词可以这样写“xuanzhuan,视频中展示了一个儿童模型,穿着绿色Polo 衫、米色工装裤一条腰带、一顶子和运动鞋,被放置在地面。这个儿童模型进行了360度旋转动作。”
我在runninghub上上传了自己的模型和试用应用,恁可以先在网站上试用,也可以直接下载试用。
这个网址是试用网址
https://www.runninghub.cn/ai-detail/1955834690592804866/?inviteCode=rh-v1059
全球通的朋友可以使用下面的链接提前试用:https://www.runninghub.ai/ai-detail/1956168462395736066?inviteCode=rh-v1059
粉丝福利:
如果你还没有注册过这个网址,通过我的链接首次注册可以领1000RH币可以免费生成好多图片视频哦!4090-24G显存,每秒钟只需要0.2RH币,另外每日登录送100RH币,畅玩4090!平台还有4090-48G显存让你体验超能魅力!
💖展现您的支持! 💖
如果您喜欢这个模型,别忘了点赞和评论——您的反馈有助于改进项目!此外,也请给予那些使用此模型发布作品的才华横溢的创作者应得的赞赏,以支持他们。
Description
FAQ
Comments (18)
Works great on Wan 2.2 I2V, just put it on the high-noise.
yes,just put it on the high-noise.
I can't make it work on a person's portrait with a cut view just below shoulders. (I didn't test yet with other). Does it have to be full view model? it worked on the other 2.1 version but can't make it work with this one.
Prompt used: "xuanzhuan. The video shows a woman model, wearing a white wool sweater, placed on the ground. This woman's model has undergone a 360-degree rotation movement."
Could you tell exactly what is the trigger word(s)?
The trigger word is: xuanzhuan
I tested it and found that I couldn't achieve a good 360-degree rotation for close-up shots. This was my oversight. Please follow me first and look forward to my next version update. Thank you for your support.
You could just pad it and crop after, you might have to do little stuff like that, if the subject is partially off frame.
I was able to achieve this without a lora. Here is the prompt I used "The woman maintains a serious expression. She turns her body to the left, she then turns to face the back, then she turns to face the right. Finally she turns to the front." Done with 81 frames.
slaad0 For portraits?
rsflipson2 yep
I’ve published two rotating videos that can’t be achieved with standard example prompts. Feel free to refer to my prompt structure—I’m simply describing the process by using only the first and last frames.
@yutangAI are your done?
For more of an "orbit" camera rotation where the background rotates along with the subject, you can use a prompt like "The [subject] and the background has undergone a 360-degree rotation movement. As the camera moves, the scene behind the initial camera placement is revealed as the camera orbits around the [subject]".
Perhaps you need to describe it this way to handle any complex scene with subject rotation. Since wan2.2 can understand shot cuts, you should frame your image-to-video prompt by specifying the first and last frames:
(xuanzhuan.)The camera starts at the cat’s eye level, centered on the cat, then gradually tilts left and begins a 180° orbit. It moves from a frontal view to a full rear view, finally closing in on the tail for a tight shot of the cat’s back.
Thanks to Wan2.2 it's finally possible to get proper 360 turns. Sadly I very often get generations where the subjects turns very slowly and thus never reaches even a quarter of the full turn. But pushing a bit with FLF2V it's possible to get there.
Is this meant for the high or low noise pass? Or is it just for 5B models? Or do you not know? I feel like a lot of people training for 2.2 just don't know how to use 2.2.
I think High is usually associate with movement so I would try it on high
@parogide I'll try that. I'm noticing that 2.1 movement loras can be forced into working (somewhat) by using a high strength in the high pass, and full strength in low pass. Non hi-lo loras that effect the appearance of specific things usually work the opposite; only used in low pass at much lower strengths (0.5 or less).
Congratulations on developing and sharing your LoRA! The quality is great. I am also facing a problem making the character rotate a full 360 degrees; it only rotates 180 degrees. It is possible that this is because I am using Q3_K_M.gguf, as it runs best on my RTX 3060 TI 8GB. My goal is to create pose frames for consistent characters. If it were possible to reverse the rotation direction (to the right), it would solve my problem, as this way it would be possible to extract pose frames from all sides of the character.
I too often get generations where the subjects turns very slowly and thus never reaches even a quarter of the full turn. Even with a generation time 16 seconds, I only get about 30 degrees of a full 360 orbiting to the right, then it goes back 30 degrees, if there is a background. When there is only a single object in the image (no background), it seems to complete the full 360.
Details
Files
Available On (3 platforms)
Same model published on other platforms. May have additional downloads or version variants.