Trained with OSTRIS AI-Toolkit @ runpod.io
Flux Kontext:
Comfyui Workflow:
Transform any input image into immersive 360-degree panoramic views with this specialized LoRA model.
Model Description
This Kontext LoRA has been specifically trained to convert regular input images into full 360-degree panoramic representations. The model leverages advanced training techniques to understand spatial relationships and generate contextually appropriate wraparound imagery.
Training Details
Dataset: Custom before/after image pairs showcasing the transformation from standard images to 360-degree panoramas
Training Framework: Ostris AI Toolkit
Model Type: Kontext LoRA for Flux
Purpose: Image-to-360° panorama conversion
Usage Notes
Base Quality: The raw output without upscaling may appear less refined
Recommended Enhancement: Use tiled diffusion upscaling for significantly improved results
Workflow: A complete ComfyUI workflow for creating seamless 360-degree images will be uploaded soon
Applications
Perfect for creating immersive content, virtual environments, VR experiences, and panoramic artwork from standard input images.
----------------------------------------------------------------------------------------------------
Flux Version:
The LoRA can create 360-degree panoramic images, and there is a high probability that the textures at the edges will be seamless.
However, as is often the case with AI-generated images, it doesn't always work perfectly.
But it works ;)
Here’s a tip for creating the images:
Choose a 2:1 format, such as 1536x768 pixels, for the first sampler. Then, use the Ultimate Upscaler to do a 2x or 4x upscaling. This should help maintain quality while generating 360-degree panoramic images.
The 'SCHNELL' version is trained with Rank 64 and SCHNELL as a Base Model, making it a bit more resource-efficient. 'Dev' offers the best image quality with Rank 128, but if speed is the priority, 'SCHNELL' is ideal.
Description
The 'SCHNELL' version is trained with Rank 64 and SCHNELL as a Base Model, making it a bit more resource-efficient. 'Dev' offers the best image quality, but if speed is the priority, 'SCHNELL' is ideal.
FAQ
Comments (76)
VERY cool
What the hell bro... AI 360 pictures. That's insane...
Thank you :)
Ai Vr soooOooon!
Ai2Vr soooOoon!
Thanks for sharing, i want to render 360 images to see in VR
Pretty cool LoRA. I suspect that if you shift a copy of all your training data by 50% (i.e. taking the first half of each image and tacking it on the end) and train the model on both, the model would probably learn to both better seal the seams, and learn to better understand "halved" main centered subjects.
Even if it doesn't improve the seams directly, it should allow for easier automatic restoration via inpainting.
About schnell - What kind of lora training algorithm was used? I tried tuning schnell using simpletuner but it doesnt seem to produce good results
I used Ostris AI-Toolkit
This is quite clever. Side note, I've wondered if we'd ever see a lora do Anaglyph 3D. Possible?
Theoretically, that's possible, but I'm not sure if it works in practice. I also don't have my glasses here to test it right now.
this one works for me in comfyui. https://github.com/Dobidop/ComfyStereo
@jjacked Well that's a nice find! Thanks!
Very cool! Can I ask what software you use to view the 360 image in your example video?
No Software I use a web viewer. Just google 360degree viewer.
@denrakeiw thanks for this! I found that Panorado is good software for viewing the images on Windows. Also I experimented with animating the images into videos, I used SadTalker to animate the face. You can use PotPlayer to play the 3D pano-video.
@denrakeiw @futuristic I built a 360 viewer that works offline inside of ComfyUI here: https://github.com/ProGamerGov/ComfyUI_preview360panorama
@progamergov Thats so cool, I will add this node to my workflow soon. Thanks
@denrakeiw Awesome! I also have this set of nodes to make working with and editing 360s even easier: https://github.com/ProGamerGov/ComfyUI_pytorch360convert
@progamergov That's awesome too—it opens up completely new possibilities for upscaling. I just suspect that the seams will be visible when slicing the image like that. I’ll have to come up with a solution for that.
@progamergov I update the workflow :)
https://civitai.com/models/1144605/seamless-360-flux-comfyui-workflow
thx
@denrakeiw You can also use the roll and rotation nodes combined with the crop/paste ones for in-painting
@denrakeiw Super cool that you are still refining this workflow! I love to see it.
What do you use to view the 3d images?
I use a Web Viewer.
Panorado is great for Windows. If you use animation tools to make these into videos, then PotPlayer is good for playing them.
so nice. thank you!!
cool
你使用了多少张图片来训练他们呢How many pictures did you use to train them?
60 Images
@denrakeiw 感谢，我应该如何搜索类似的图片呢，我想自己训练一个类似的试试看，因为我想要的风格和你的有一些区别，有什么技巧可以分享一下吗Thanks. How can I search for similar pictures? I want to train a similar one by myself and try it out. Because the style I want is a bit different from yours, is there any tip you can share?
@tunan996 u can find some here:
https://hdri-haven.com/
@denrakeiw I hope to find some anime types. However, thank you very much.
@tunan996 Maby u can make 360 degree Screensots in Anime VR Games
Very impressed with what I've seen so far. Looking to improve the definition in the faces. Saw you mention upscaling, which models do you suggest?
Ultimate upscaler with FLux + LoRa
@denrakeiw That's the workflow and custom node, right? I meant the upscaling model, like UltraSharp or Gigapixel.
Most of what I've seen are 4x models, but even the original render is running my GPU fan.
@unclehater Yes, 4x UltraSharp for Backgrounds or Objects
I don't know. I click your preview images, they don't move around, or do the 360 thing., and most of the images on the page from users are all flat images, no 360. If I can't move the image on the demo/download page, why would I download it? Either this is bullshit, or this website is shit and can't even correctly display what it promotes. Lame and a waste of time.
If you don’t have a 360-degree screen, a mini Hyperbowl, or a VR headset, you’ll either need to heat up your monitor until the plastic becomes bendable so you can shape it into a 360-degree display, or you can load the image into a 360-degree viewer and view it in full panorama on a flat screen.
Here’s a viewer you can use: Renderstuff 360 Panorama Viewer.
360全景近处的物品应该缺乏立体感，会出现很大的感觉，不知道有什么软件可以变成左右眼格式那就爽了
Awesome lora, but it's still fascinating how many 3-walled and 2-walled banana-rooms it creates. Not your fault though. AI works in mysterious ways :)
Yes, the dataset was quite limited. Nvidia, together with Shutterstock, has now trained a model that doesn't have these errors—but they also have more resources than I do. ;)
@denrakeiw Yeah... It sucks having less resourses than Nvidia and Shutterstock... I feel ya 😊
Just noticed a problem with the lora (or with me?). I can make it work using GGUF checkpoint, but not with any other Flux-checkpoints
Please update the iteration as soon as possible, use more pictures and more training time, and strive for a high-quality product, thank you.
Thanks for the tips from someone who doesn't have a single model on Civitai...
I don't have the same resources available as NVIDIA and Shutterstock....
https://build.nvidia.com/shutterstock/edify-360-hdri
Please update your comment. It comes off as rude.
any particular wf that works well for you? TIA
Yes, I have a workflow that can generate 360-degree images without the seams. However, I still need to refine it a bit before I can upload it.
honestly this lora is mindblowing!
Thank you :)
Looking forward to an update, better model, thanks
Check out this workflow, with this it's working better ;)
https://www.seaart.ai/de/workFlowAppDetail/ctcagd5e878c738m1i60
For some reason I am not getting a full spherical panorama, it appears to be clipping the top and bottom. Any ideas?
Try it with this workflow:
https://www.seaart.ai/de/workFlowAppDetail/ctcagd5e878c738m1i60
@denrakeiw Where can I download the workflow to run locally on my Mac?
@denrakeiw this is for ComfyUI how about forge?
@mazen3ds583 seamless it's just working with comfyui. With forge maby it works but with seams.
期待更新，更好的模型，谢谢
期待更新，更好的模型，谢谢
期待更新，更好的模型，谢谢
can this work with forge
I never used Forge but it should work. But the best Option is to use it in comfyui with the 360 Workflow.
@mcrodose just tested the LoRA with FLUX, works flawlessly!
期待更新，用更多图片，训练更好效果的模型，感谢啦，辛苦啦，爱你么么哒！
非常感谢！看看我什么时候能收集到足够的图片，到时候可以用这个数据集训练一个 DoRA 版本。
Dear friend,
I hope this message finds you well. I wanted to kindly inquire about the progress of your updates, as I’ve been eagerly anticipating them. To further enhance the quality and effectiveness, I would sincerely suggest incorporating a larger dataset for training—perhaps no fewer than 10,000 images—and dedicating at least 1,000 GPU hours to the process. I believe this approach could significantly improve the results.
Thank you so much for your hard work and dedication. I truly appreciate your efforts, and I’m grateful for your commitment to this project.
Warm regards,
Your friend
look at the new version
this is so Good, please keep update
Thank you ^^Next Update with the next amazing Base Model
Friends, looking forward to your update, you are so great!