Hey everyone! This is a camera movement Lora for wan2.2 that creates an orbit shot, moving from the character's front to their back.
💻Online One-Click Generator & Workflow
I've set up an online ➡️ workflow for you so you can easily try out other different effects.
🎁Bonus: If you're signing up for the first time, you can get 1,000 free RH Coins by using my link and the invite code ➡️rh-v1182. Plus, you'll get another 100 RH Coins for logging in daily.
🚀Here’s how to use it in ComfyUI:
1. This is a High_Noise model. It should only be used in the high-noise channel, not the low-noise channel.
2. A Lora weight of 1.0 is recommended.
3. The sampler's boundary is set to 0.9. This is the signal-to-noise ratio for the MOE sampler in I2V cases, so you don't need to change it. For the sampler and scheduler, I recommend uni_pc+beta or res_2s+bong_tangent.
4. A local decoding module is provided. Go to https://github.com/liangtongt/TT-tools and download TT_lmg_Loc.exe to any folder on your computer. Download the grayscale image from the online workflow and save it to the same folder. Then, just drag the grayscale image and drop it onto the TT_lmg_Loc.exe file to decode it. You'll get the video that was generated online, and the locally decoded version won't have a watermark.
✍For the positive prompt, use this structure:
ymq,[Shot type, Composition],surround camera movement, the camera circles 360° from the front to the back of the subject,[Description of the main appearance],[The environmental description of the subject in the image],The camera pans from her front to her back,[Transform scene description],[Image tone and atmosphere description]
Prompt Example:ymq, medium shot, central composition, surround camera movement, the camera circles 360° from the front to the back of the subject, showing a classical woman with her hair coiled into a bun and a graceful figure. She is wearing a light cyan halter-neck tight dress with ink-wash bamboo patterns. She stands still on a stone bridge. The camera pans from her front to her back, surrounded by stone Bridges and buildings from the water towns of Jiangnan. The tone of the picture is cold and clear, presenting a hazy and poetic atmosphere of the misty rain in Jiangnan.
It's important to note that for the [Transform scene description] part of the prompt, the description should be related to the original background but not exactly the same. Otherwise, wan2.2 might just turn the character around from front to back instead of creating an orbiting camera shot. For example, if the original character is standing in a Chinese-style courtyard with flowers and trees, the transformed scene could be described as "surrounded by gardens and Chinese architecture." By facing different scenes, the wan2.2 model can smoothly switch the environment, which is what achieves the orbiting camera effect from the character's front to their back.
💖I'm E.K., an engineer who can't draw. I'm passionate about creating more fun and practical AI models.
If you like this model, please don't forget to drop a like, save it, and give me a follow! I'd also love to see what you create with this Lora, so feel free to share your work in the comments. I can't wait to see everyone's creativity!
If you run into any issues or have any suggestions, please leave a comment and let's chat. Can't wait to bring these models to life with you!
Description
FAQ
Comments (16)
Will give it a shot and report back, have been using a similar one but struggle to do an arc beyond about 90 degrees and it swings back.
Okay, thanks for the feedback! For what it's worth, four of the five videos shown on the main Lora page were generated using images that weren't in the dataset. I haven't run into the issue you're describing where the orbit can't go past 90 degrees, but a situation that can sometimes happen is the character turns around on the spot instead of the camera orbiting around them.
There is a 360 available but it doesnt work well. Just run this once, then do first-last and stitch them together, voila! perfect 360
can please lora for a full 360 orbit.
I'm not planning on training a full 360-degree orbit Lora anytime soon.
I'm missing something in reading your description. What is "A local decoding module"? Is it necessary? And why do you start your prompt examples with "ymq"?
1. The "local decoding module" provides another way to save your video. It's not required, so I have it disabled by default. When you want to use it, just enable the module and run the workflow, and you'll get a grayscale image that you can use to save the video. This module allows you to bypass runninghub's content moderation system. You can use their powerful online machines to quickly generate videos with sensitive content and then download them as grayscale images to decode on your own computer. The video or image you get after decoding locally will not have a watermark.
2. "ymq" is just a trigger word. Like any other trigger word, its only job during the Lora training is to be associated with the specific content or concept that the Lora is supposed to learn.
OK, many thanks for that. The critical bit I was missing was that the local decoding model was to save videos created online. I just don't do that - I make do with my RTX 4080. I guessed "ymq" might be a trigger but I couldn't guess what it was short for.
Not working for me.
Excellent
For all the Loras that do the surround shot, this one works the best (by far). The other ones are very inconsistent(just don't work a lot of time) and too fast or too slow. Everyone seems to provide a complicated workflow that requires a 5090 to run. This one worked with a normal workflow and the suggested promts. Thank you so much for creating this.
Thanks for your support! I've had some unsuccessful cases while testing this Lora myself, but as long as the prompt is used correctly, the success rate for getting the orbit shot effect is still quite high.
Why do you use the same trigger word for all your LoRAs?
For me, this trigger word is kind of a symbol, but really, it's just a trigger word for the Lora.
hi! nice work! any trick to control the direction of the spin?
Hey there! You can find the prompt structure guide right on the Lora's webpage. The key takeaway is that you need to describe a scene that's related to the original background but not exactly the same. This helps the model understand that as it rotates the character, the background needs to shift along with the rotation.