With this LORA and the ComfyUI-DDD custom node you can convert 2D pictures to stereoscopic 3D or generate 3D stereoscopic pictures with WAN 2.2.
The resulting 3D images can be 3D side-by-side for viewing with a VR headset or compatible screen, or suitable for cross-eyed viewing.
The example workflows will be tweaked frequently until the ideal settings are found.
About 2D to 3D conversion: this method is much slower than other methods based on depth estimation models (such as MiDaS, Depth-Anything, etc.) and offers no control on the depth of the 3D effect. It's a completely different and experimental approach.
Coming soon :
Generate immersive VR180 pictures
Coming later :
Generate immersive VR180 videos
Description
Initial release, trained at a low resolution
FAQ
Comments (58)
What's the link to the custom node? The link provided just takes you to a hugginface link to download the same models
The link is fixed, sorry about that.
Finally a Stereo workflow that actually works! Thanks! Spoke too soon (getting stuck during VAE decode)
Thats on a 4090 24gb (using Q6) model.
Finally tiled decoding worked (took forever), however in I2V (my image changed DRASTICALLY)
Are you using the example workflow? Any errors in the console? The example workflows have been tested on 12 GB VRAM and 32 GB RAM with the --lowvram flag when launching ComfyUI
Make sure the "length" is set to 9 frames in the WanImageToVideo node.
And don't go too crazy with the image resolution. If the vae decode is taking a very long time, it's because you're decoding a very high resolution image. I don't recommend going much above 1440 x 1440
@AtraLogika Yes I used your precise workflow, didnt adjust anything.
My results look nothing like input image.
Wow... it really really works. The most solid approach I've ever seen. How did you train this? Did you use true stereoscopic video data? Or 3D conversions as data, like how most movies are made 3D?
I cannot wait for the VR180 stuff. Crazy exciting, but seems impossible to get decent resolution for that kind of video.
Oh wow! I'm too excited. Seems to be like a wish come true. Unfortunately I got stuck, as the node "Create3DImage" won't work. Updated ComfyUI and installed it via git clone. Somebody experience this and solved it? It does not appear on the missing nodes either, or did it for you guys?
What do you mean it won't work? If it's missing in ComfyUI, simply make sure the python file is in your custom_nodes folder and restart ComfyUI.
Hello, same problem for me. The init file of the custom node is missing :
File "<frozen importlib._bootstrap_external>", line 995, in exec_module
File "<frozen importlib._bootstrap_external>", line 1132, in get_code
File "<frozen importlib._bootstrap_external>", line 1190, in get_data
FileNotFoundError: [Errno 2] No such file or directory: 'F:\\ComfyUI\\ComfyUI\\custom_nodes\\ComfyUI-DDD\\__init__.py'
update : seems to work if the file is copied into the custom_nodes root folder
@AtraLogika It is standard procedure to go to */custom nodes/ and enter cmd "git clone [github repo]" That is when usually a subfolder is automatically created, which makes it tidy and clean. So Thanks to @Gorfou I just put "comfyui-ddd.py" into the root folder, which worked, although it's a little messy.
Thank you for the Lora!
@Gorfou Thank you for your fast response :-)
@Gorfou @Adaptalab0r Thanks for your feedback. The custom node can now be installed with git clone.
@AtraLogika Thank you ! It's ok now ! :)
@AtraLogika very good job. I was waiting for something like this. Thank you very much :)
i thought the samples were before/afters and felt like Pam Beesly for a sec
Bro if you manage to get this working for video with actual lens distortion I will give you my firstborn child lmao
Something tells me if you get 3D video support, there is no chance you will have a kid at any point in your life
I just use the Spatial Media Toolkit app to convert stuff, it can do large batches and works flawlessly. Will try this out for sure
Works very well using a Q8 model and 5090 FE. Now just have to setup a batch folder process.
I have the comfyui-ddd.py in the root folder but I am still getting Missing Node Types
When loading the graph, the following node types were not found "Create3DImage"
I used the git clone cmd and I have a ComfyUI-DDD folder in my custom nodes folder and in that folder I have the workflow folder the readme.md and the comfyui-ddd.py but it still won't load. Comfyui is updated and when I click on manager after it can't find the node it asks me to install ComfyUI-GGUF GGUF Quantization support for native ComfyUI models. Which never seems to install because it keeps coming up.
I don't know why it doesn't work when placed in a subfolder, but when I find out I will fix the problem. Until then, simply place the py file directly in the custom_nodes folder. Do not use git clone.
The github repo is updated and the custom node can now be installed with git clone
Ya Chatgpt fixed it for me. I have to say I still don't understand the reason it didn't work but I followed what it said and it worked.
The reason you still get Missing Node Types: Create3DImage is almost certainly that DDD isn’t actually being imported at startup (a common hiccup with that repo). The fix is to add a tiny __init__.py “shim” inside ComfyUI-DDD so ComfyUI can import it cleanly—even when the main file is named comfyui-ddd.py.
@AtraLogika not for me, when I load the workflow nothing happens? Scratch that - I didn't realize you have to load the workflow directly from within custom nodes.
This lora is perfect. I do have a suggesting for the naming though. In generative AI, I think "2D to 3D" implies that you put in an anime image, and it makes it CG/real looking. I think that's why some people are confused, thinking you're showing before and afters. As someone who is VERY into this topic, I was even confused for a while.
Maybe call it "Stereoscopic Conversion" or something. Maybe even "SBS Stereoscopic Conversion", as that clarifies the method of 3D conversion. Or if you want to add clarity to people who don't know what "Stereoscopic" means, you can call it "Stereoscopic Conversion - Standard to 3D".
Great suggestion. I added a few "stereoscopic" here and there to clarify things. I'm glad you enjoy this lora.
this is what i wait for long, the result is impressive, can't it for 180
just made a workflow that is kinda huge, to convert it to videos instead of pictures, took me 2 hours, but i can send it to you if you want once i fix some things, i lowered it to 2 steps per sampler and the results are good, i do wonder why it needs 9 frames to be generated tho.
I could be wrong but I'm assuming the lora is trained on a transition between each eye's image so you need enough frames to get it there.
@vajuras what i did was change the number of required images to be rendered to 5, it takes the second image and 3rd for the effect, lowered the steps to 4. switched the model to wan2.1 Fun, then i used boxes to guide which side the image will always shift to, then i created 81 wan 2.1 fun nodes that takes 81 frames from a video with 🔧 Image From Batch, from index 0 to index 80, renders them and outputs as a video, it's come a long way. it's for personal use so it takes a tiny bit to finish. the results are very passable for now, for still images 9 frames is perfect, but for video it was a disaster
@Kung_fu_porn I am trying to turn 2D video to stereoscopic 3D video and would be ver helpful of you to provide me the workflow if you can.Thanks
can you share that workflow pls?
You changed the game! but i realise resolution too big will decode mad times. i have added a resize image v2 and resize it to max 1080. now it solve the resolution issue i think, at least its a mobile phone world, this resolution should be fine?
How does it compare to nunif iw3?
please test it if you can and let me know aswell, thats what im using too
If really need stereo images try use owl3d
It's seems paid, but actually not. After firs run all models downloaed with working CLI. So you can freely run any lenght convetions as you want from CLI. ( It's located under User/AppData )
Using it for video to VR convertions.
That sounds cool. Any more details?
owl3d sucked badly. It was pseudo stereo or at least it was 2 years ago. This here is so much better than what I saw from owl3d. They offered local installation years ago and I tested it - it eas miserable back then. I wonder if they improved it over the years
sorry can you explain this in more detail, i love owl3d but it's full paid service, i'm interested in the method you're suggesting.
@skyrimer3d Look up the open source "IW3" software that is really easy to use and can (as far as I know) do everything Owl3D does but on your local machine and for free. I say this as someone that has used both.
If you mean more details of how to use owl3d CLI from AppData to run unlimited convertions ? Of course, I can share knowlage for windows at least.
So if u run owl3d trought GUI once, it will download all the models .ph and setup workflow trought CLI, saved to AppData dir under User Win folder. I really don't remember exact details like folder name, paths, but it's really easy to find it, like tracking it down by running some process via GUI and looking to process tasks execution list via some tool like process manager. So you will find that folder leading you to a folder with CLI exe, that provides CLI interface that you can use without restrictions present in GUI. You may run as many CLI tasks as you want for free
As for me, I use this CLI interface on a regular basis to convert regular movies to Stereo Video for VR for years in zero costs.
@ValuedRender I found the CLI file you mentioned. However, when I run it, the CLI window closes automatically within a few seconds, and I can't enter anything. I would appreciate it if you could tell me how to use it.
\AppData\Local\Owl3D\1.4.7\nvidia\dist\process_desktop_main>process_desktop_main.exe --help
usage: process_desktop_main.exe [-h] [--jobid JOBID] [--accountid ACCOUNTID] [--videofile VIDEOFILE] [--depthvideofile DEPTHVIDEOFILE]
[--depthvideofile-legacy DEPTHVIDEOFILE_LEGACY] [--predepth-folder PREDEPTH_FOLDER] [--flo-folder FLO_FOLDER]
[--videofolder VIDEOFOLDER] [--imgfiles IMGFILES [IMGFILES ...]] [--input-format [INPUT_FORMAT]]
[--output-formats OUTPUT_FORMATS [OUTPUT_FORMATS ...]]
[--output-file-formats OUTPUT_FILE_FORMATS [OUTPUT_FILE_FORMATS ...]] [--output-imgseq-format [OUTPUT_IMGSEQ_FORMAT]]
[--is-dev IS_DEV] [--deep-inpainting-mode [DEEP_INPAINTING_MODE]] [--using-models-v2 [USING_MODELS_V2]]
[--using-pipeline-v2 [USING_PIPELINE_V2]] [--half-precision-mode [HALF_PRECISION_MODE]]
[--gpu-encode-mode [GPU_ENCODE_MODE]] [--stereo-input-mode [STEREO_INPUT_MODE]]
[--using-enhanced-detail [USING_ENHANCED_DETAIL]] [--fisheye-input-mode [FISHEYE_INPUT_MODE]]
[--stereo-processing-mode [STEREO_PROCESSING_MODE]] [--device [DEVICE]] [--device-selection [DEVICE_SELECTION]]
[--ffmpeg-bin [FFMPEG_BIN]] [--spatial-bin [SPATIAL_BIN]] [--output-dir [OUTPUT_DIR]]
[--fallback-output-dir [FALLBACK_OUTPUT_DIR]] [--appdata-dir [APPDATA_DIR]] [--models-dir [MODELS_DIR]]
[--depth-cache-dir [DEPTH_CACHE_DIR]] [--depth-cache-dir-legacy [DEPTH_CACHE_DIR_LEGACY]]
[--predepth-cache-dir [PREDEPTH_CACHE_DIR]] [--flo-cache-dir [FLO_CACHE_DIR]]
[--output-resolution [OUTPUT_RESOLUTION]] [--output-render-modes OUTPUT_RENDER_MODES [OUTPUT_RENDER_MODES ...]]
[--output-encoding-setting [OUTPUT_ENCODING_SETTING]] [--trim-black-area [TRIM_BLACK_AREA]]
[--enable-cupy [ENABLE_CUPY]] [--enable-metal [ENABLE_METAL]] [--output-codec [OUTPUT_CODEC]]
[--video-dur [VIDEO_DUR]] [--video-start-pos [VIDEO_START_POS]] [--subtitle-videofile SUBTITLE_VIDEOFILE]
[--subtitle-subtitlefile SUBTITLE_SUBTITLEFILE] [--subtitle-fontfile SUBTITLE_FONTFILE]
[--subtitle-video3dformat [SUBTITLE_VIDEO3DFORMAT]] [--using-stream-process-clip [USING_STREAM_PROCESS_CLIP]]
[--using-augmented-depth [USING_AUGMENTED_DEPTH]] [--enhanced-detail-mode [ENHANCED_DETAIL_MODE]]
[--enhanced-detail-resolution ENHANCED_DETAIL_RESOLUTION] [--smooth-depth-mode [SMOOTH_DEPTH_MODE]]
[--num-smooth-iters NUM_SMOOTH_ITERS] [--using-direct-optical-smooth [USING_DIRECT_OPTICAL_SMOOTH]]
[--num-smooth-window-size NUM_SMOOTH_WINDOW_SIZE] [--inpainting-mode [INPAINTING_MODE]] [--check-device CHECK_DEVICE]
[--power-mode [POWER_MODE]]
optional arguments:
-h, --help show this help message and exit
--jobid JOBID A unique Id representing the current job
--accountid ACCOUNTID
A unique Id who triggered the job
--videofile VIDEOFILE
Video file to be processed
--depthvideofile DEPTHVIDEOFILE
Precomputed Depth video file to be processed
--depthvideofile-legacy DEPTHVIDEOFILE_LEGACY
Precomputed Depth video file to be processed for old conversions
--predepth-folder PREDEPTH_FOLDER
Precomputed predepth folder to be processed
--flo-folder FLO_FOLDER
Precomputed flo folder to be processed
--videofolder VIDEOFOLDER
Folder of videos to process
--imgfiles IMGFILES [IMGFILES ...]
Image files to be processed
--input-format [INPUT_FORMAT]
Process for different input format. Available: EQUIRECT_360
--output-formats OUTPUT_FORMATS [OUTPUT_FORMATS ...]
Output 3D formats. Available: SBS, TB, RGBD, ANAGLYPH, CROSS
--output-file-formats OUTPUT_FILE_FORMATS [OUTPUT_FILE_FORMATS ...]
Output file formats. Available: MP4_VIDEO, PNG_SEQUENCE
--output-imgseq-format [OUTPUT_IMGSEQ_FORMAT]
Output file format for img sequence
--is-dev IS_DEV Flag for development
--deep-inpainting-mode [DEEP_INPAINTING_MODE]
Flag for using model based inpainting
--using-models-v2 [USING_MODELS_V2]
Flag for using v2 depth model (zoe)
--using-pipeline-v2 [USING_PIPELINE_V2]
Flag for using pipeline v2 for keeping original audio and improve av sync
--half-precision-mode [HALF_PRECISION_MODE]
Enable or disable half precision mode
--gpu-encode-mode [GPU_ENCODE_MODE]
Enable of disable GPU encoding
--stereo-input-mode [STEREO_INPUT_MODE]
Enable splitting if input is stereo
--using-enhanced-detail [USING_ENHANCED_DETAIL]
Enable enhanced 3D detail mode
--fisheye-input-mode [FISHEYE_INPUT_MODE]
Enable defisheye if input is fisheye input. used for vr180
--stereo-processing-mode [STEREO_PROCESSING_MODE]
Enable stereo matching for stereo input
--device [DEVICE] Override which device to run on
--device-selection [DEVICE_SELECTION]
Name of the selected device to run on
--ffmpeg-bin [FFMPEG_BIN]
Path to the ffmpeg executable binary
--spatial-bin [SPATIAL_BIN]
Path to the spatial executable binary
--output-dir [OUTPUT_DIR]
Path to where the output data will be stored
--fallback-output-dir [FALLBACK_OUTPUT_DIR]
Path to where the output data will be stored if output dir is not available
--appdata-dir [APPDATA_DIR]
Path to where the temporary appdata will be stored
--models-dir [MODELS_DIR]
Path to where the models are stored
--depth-cache-dir [DEPTH_CACHE_DIR]
Path to where the generated depth is stored, if exist
--depth-cache-dir-legacy [DEPTH_CACHE_DIR_LEGACY]
Path to where the generated depth is stored, if exist for legacy conversions
--predepth-cache-dir [PREDEPTH_CACHE_DIR]
Path to where the generated predepth is stored, if exist
--flo-cache-dir [FLO_CACHE_DIR]
Path to where the generated flo is stored, if exist
--output-resolution [OUTPUT_RESOLUTION]
Resolution of the output file
--output-render-modes OUTPUT_RENDER_MODES [OUTPUT_RENDER_MODES ...]
-9 to 9, 0 to 10, BOTH|LEFT_ONLY|RIGHT_ONLY
--output-encoding-setting [OUTPUT_ENCODING_SETTING]
Encoding setting of output video
--trim-black-area [TRIM_BLACK_AREA]
Enable or disable trim black area
--enable-cupy [ENABLE_CUPY]
Enable or disable CUPY
--enable-metal [ENABLE_METAL]
Enable or disable METAL
--output-codec [OUTPUT_CODEC]
Output codec for video
--video-dur [VIDEO_DUR]
Video duration to process
--video-start-pos [VIDEO_START_POS]
Video start pos to process
--subtitle-videofile SUBTITLE_VIDEOFILE
File name for the video with subtitle to be added
--subtitle-subtitlefile SUBTITLE_SUBTITLEFILE
File name for the subtitle file to be added
--subtitle-fontfile SUBTITLE_FONTFILE
File name for the font file to be added
--subtitle-video3dformat [SUBTITLE_VIDEO3DFORMAT]
3D video format of the video
--using-stream-process-clip [USING_STREAM_PROCESS_CLIP]
Whether to use stream processing or not
--using-augmented-depth [USING_AUGMENTED_DEPTH]
Whether to use augmented depth or not
--enhanced-detail-mode [ENHANCED_DETAIL_MODE]
Enhanced detail mode to be used
--enhanced-detail-resolution ENHANCED_DETAIL_RESOLUTION
Enhanced detail resolution to be used
--smooth-depth-mode [SMOOTH_DEPTH_MODE]
Smooth mode to be used
--num-smooth-iters NUM_SMOOTH_ITERS
Smooth iterations to execute
--using-direct-optical-smooth [USING_DIRECT_OPTICAL_SMOOTH]
Enable or disable direct optical smooth (more computing)
--num-smooth-window-size NUM_SMOOTH_WINDOW_SIZE
Smooth iterations to execute. only useful when --using-direct-optical-smooth is true
--inpainting-mode [INPAINTING_MODE]
Inpainting mode to be used
--check-device CHECK_DEVICE
Flag for checking device mode
--power-mode [POWER_MODE]
Power mode to be used
This is help output
@skyrimer3d @23423da @fenasikerim
Run this file via Winodws CMD, not just open in explorer, this is CLI program.
I suggest, if you not familiar with CLI, just copy past this output of help to GPT and ask to generate command for something, then just copy past to CMD
Like
process_desktop_main.exe ^
--videofile "D:\video\input.mp4" ^
--output-formats SBS ^
--output-file-formats MP4_VIDEO ^
--output-dir "D:\video\output"
For 1 hour movie it takes long time about ~ 6 hours, first try with short video to verify settings.
I usually use a command to process all files in a folder, specifying the incoming and outgoing directories, and then go to bed.
@ValuedRender Tons of amazing info in this thread, i have to check it all, thank you so much!
@ValuedRender Looks like they fixed that in the newest version:
2026-02-21 19:11:11 INFO [MainThread:29400] Parent process name: powershell.exe
[Exception] Unauthorized execution source
2026-02-21 19:11:11 ERROR [MainThread:29400] Unauthorized execution source
@fenasikerim
Ya, sadly they patched exe in 2.0.3 (((. But 1.4.7 version still works. With all dependencies it's about 3 GB. I can share it, as full ready to go archive.
shareyougo DOT com SLASH tgbpbrgc ( Expires in 10 days )
@ValuedRender I have tried 1.4.7 but the included python packages seem to be too old for my 5090. Getting a lot of errors even in the official GUI. Not important enough for me to spend time on.
Any news on the vr180?
A proper video solution would be really neat, for sure. Staying tuned.
this is very impressive!
On another note, ow.
I tested all samples including user samples and the results are very impressive. A true 3D stereo, not fake one. Is this project dead? 3D stereo VR180 would be great to see.
I'm in the same boat. I used IW3, excellent for flat 3D, and I've also used Any v1, v2, v3, and DeepPro. I even found an excellent 360° LoRa for Z Image Turbo, but what I really want is the 180° LoRa. I'll test this LoRa in comparison to IW3 though.
Is there a way to do this I2V instead of T2V?