1-Click INSTALL OVI THE FIRST OPEN-SOURCE SORA 2 WebUI!
Added 2025-10-04 17:49:20 +0000 UTCEDIT 18 October: Just reuploaded a fixed runpod installer since the last one didn't work anymore because of an update, now it works fine, redownload the new INSTALL_OVI_RUNPOD.sh installer.
Edit 05 october: local installer has been updated and all bugs have been ironed out (hopefully) + I also added support for FP8 model that takes around 16gb of vram to run, so try it out
Hey everyone! I've created a 1-click installer for “The FIRST Open-Source Sora 2 Model" Webui, featuring the brand-new OVI model. This one is a absolute first in the history of open-source AI models because we finally have a model that is able to generate a video with AUDIO either from scratch with a simple text prompt or with an Image to convert it into a video with audio, total game-changer!
You can check out the video right here: https://youtu.be/2SvPfkXs3Nk
The installer of course automates the entire install process for the webui!
Edit 05 october: local installer has been updated and all bugs have been ironed out (hopefully) + I also added support for FP8 model that takes around 16gb of vram to run, so try it out
Hey everyone! I've created a 1-click installer for “The FIRST Open-Source Sora 2 Model" Webui, featuring the brand-new OVI model. This one is a absolute first in the history of open-source AI models because we finally have a model that is able to generate a video with AUDIO either from scratch with a simple text prompt or with an Image to convert it into a video with audio, total game-changer!
You can check out the video right here: https://youtu.be/2SvPfkXs3Nk
The installer of course automates the entire install process for the webui!
Now as of right now, it's still a "full" model because it's not quantized so to run it locally you need around 32gb of VRAM...yeah...so unless you got a 5090 lying around, your only option is runpod, at least as of right now.
LOCAL INSTALL!!! RECOMMENDED TO HAVE A 5090 OTHERWISE USE RUNPOD!!!
Download the INSTALL_OVI.bat
Run the bat file
If it asks for the cpu offload option, press Y (unless for some reason you got a huge 48gb+ vram gpu at home, in that case press N) press enter
Wait for the install to finish
The webui will launch, next time use the bat file launcher to start the webui and there you go😎
IF YOU ARE USING RUNPOD (RECOMMENDED FOR ALL):
Create an account if you haven't already: Runpod
Click on Pod (on the left side) then click deploy
Choose a GPU with at least 32gb of VRAM (5090 is best and cheap), chose the pytorch 2.8.0, then edit the template and choose 100gb for both the container and volume disk, add the 3000 port and then deploy on demand
Go to my pods, wait for everything to finish and then click "connect", then "Connect to HTTP SERVICE port 8888" and drag and drop the INSTALL_OVI_RUNPOD.sh file on the left side of the UI then click on the "Terminal" icon on the right side on the UI
Copy and paste these two lines then press enter:
chmod +x INSTALL_OVI_RUNPOD.sh
./INSTALL_OVI_RUNPOD.sh
Wait for everything to be installed
Go to the port 3000 and there you go 😎
As always, supporting me on Patreon allows me to keep creating helpful resources like this for the community. Thank you for your support - now go have some fun😉!
Comments
I changed to the 720x720_5s in the inference_fusion but and it passed, gradio opens but then on generation, runs through "i assume" iterations but it is definitely running longer than it should. loading model... enable_cpu_offload=True, use_image_gen=False, fp8=True, qint8=False for gradio demo Score model (Fusion) all parameters:11660753108 Removing weight norm... Successfully loaded fusion checkpoint from ./ckpts\Ovi\model_fp8_e4m3fn.safetensors loaded model * Running on local URL: http://127.0.0.1:7891 * To create a public link, set `share=True` in `launch()`. Pure T2V mode: calculated video latent size: 32 x 62 18it [51:13, 172.28s/it]
Jesse Moore
2025-12-18 17:30:43 +0000 UTCFirst, thank you very much for all the work you do, amazing. I am having an issue getting this up and running. Running everything on RTX 5090 32gb. Install seems to work like a charm but once it starts the service, i get the following: loading model... enable_cpu_offload=True, use_image_gen=False, fp8=True, qint8=False for gradio demo Score model (Fusion) all parameters:11660753108 Removing weight norm... Traceback (most recent call last): Ovi\gradio_app.py", line 53, in ovi_engine = OviFusionEngine() File "D:\OviComfyUI Temp\Ovi\ovi\ovi_fusion_engine.py", line 100, in __init__ assert model_name == "720x720_5s", "FP8 quantization is only supported for 720x720_5s model currently." AssertionError: FP8 quantization is only supported for 720x720_5s model currently.
Jesse Moore
2025-12-18 14:50:38 +0000 UTCYeah that just means you ran out of memory for the video you tried to generate...check your parameters or rent a bigger gpu
Aitrepreneur
2025-12-15 02:38:14 +0000 UTCI’m reporting an issue with the RunPod deployment using your provided .sh install script. I spent around 1.5 hours deploying a pod with an RTX 5090, exactly as you recommended. I used the script you provide, which is supposed to have everything preconfigured. The installation completes successfully and the UI starts, but when I try to generate a video it consistently fails with a CUDA Out Of Memory error. Here is the exact error I get during video generation: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 324.00 MiB. GPU 0 has a total capacity of 31.37 GiB of which 81.81 MiB is free. Including non-PyTorch memory, this process has 31.28 GiB memory in use. Of the allocated memory 30.37 GiB is allocated by PyTorch, and 319.87 MiB is reserved by PyTorch but unallocated.
Jorge Luis
2025-12-15 02:33:41 +0000 UTCAlso, hypothetically... If I were to atempt a Render with a 4090 how long would a 5s render take to fully bake? (asking for a friend!)
Vato Vegas
2025-11-26 21:49:05 +0000 UTCthe FP8-OVI Launcher doesn't "launch" the Web-UI for me... anyone else come across that issue with a resolve?
Vato Vegas
2025-11-26 21:47:01 +0000 UTCDoes anyone know what Loras work with this? Wan 2.1 or Wan 2.2? I don't see the typical High and Low categories required for Wan2.2 Loras, even though it says Wan2.2 Animate which is why I am asking.
Eyes_Wide_Shut
2025-11-03 10:42:44 +0000 UTCinside the OVI directory tree is a dir full of examples. It's the one called Example_prompts
Tinkering
2025-10-26 07:24:53 +0000 UTCYou mentioned sample prompts, can you point me in that direction?
Emilien-Edmond D'aramitz (LeonceTheLion)
2025-10-24 08:29:25 +0000 UTCStupid question maybe (definitely). But it took hours to install on jupiter last night and by the time it did I needed to sleep so I just closed the pod down on runpod. Opened it up now but I dont know how to start it again in jupiter. Pressing 3000 does nothing. Hope that makes sense
Ana Herrera
2025-10-23 15:51:58 +0000 UTCaside from having to install python 3.10 this was the first installer that ran and finished flawlessly for me
Jollyboy
2025-10-18 17:47:09 +0000 UTCHello. Is there any simple way to change the compression bitrate of the output video, because it seems to be overly compressed? I wish I could understand python programming so that I could improve this demo program...
Bradford
2025-10-16 22:12:12 +0000 UTCI tried running it on runpod but it fails at the install. I tried with a 6000 PRO and 5090 but same results. Below is where the installer gets stuck. Even after 10 hours it stays on this stage Building wheels for collected packages: flash_attn DEPRECATION: Building 'flash_attn' using the legacy setup.py bdist_wheel mechanism, which will be removed in a future version. pip 25.3 will enforce this behaviour change. A possible replacement is to use the standardized build interface by setting the `--use-pep517` option, (possibly combined with `--no-build-isolation`), or adding a `pyproject.toml` file to the source tree of 'flash_attn'. Discussion can be found at https://github.com/pypa/pip/issues/6334 Building wheel for flash_attn (setup.py) ... /
Väinämöinen
2025-10-16 06:58:01 +0000 UTCthanks for trying tho.
theRCB
2025-10-16 02:43:08 +0000 UTCdoesn't work. wasted money waiting on this for 30 mins.
theRCB
2025-10-16 02:42:38 +0000 UTCEvery single time I follow these tutorials there are dozens of missing dependencies that refuse to download or no longer exist ;o;
Hunnie
2025-10-16 02:20:31 +0000 UTCHi. I installed this and runs smooth on the 5090 video card of my pc. Does this only add audio to a picture transformed in a video? I ma looking to transform pictures in videos with commands/actions to assign to the picture. What should I use for that, please?
MC
2025-10-16 01:39:25 +0000 UTCYour Lora training videos are the best on YouTube. Are there any plans to release a Qwen Lora tutorial anytime soon?
James
2025-10-12 22:38:45 +0000 UTCFIXED, the issue as mentioned above. The gradio_app.py file has changed, so we now need to specify the port. Replace the last line of the file , open INSTALL_OVI_RUNPOD.sh—so the last line must look like "python3 gradio_app.py ${CPU_FLAG} --cpu_offload --server_port 3000 --server_name 0.0.0.0"
Manuel Ramos
2025-10-12 13:35:34 +0000 UTCHi, I encountered the same problem. My assumption is that the Python file has changed. Here’s how I got it working: 1. Install everything using the 1-click installer. 2. When you see “[ * ] Launching Ovi on 0.0.0.0:3000 ...”, stop the running script using Ctrl+C. 3. Navigate to the project folder: cd /workspace/Ovi/ 4. Activate the virtual environment: source .venv/bin/activate 5. Run the app with CPU offload enabled: python gradio_app.py --cpu_offload --server_port 3000 --server_name 0.0.0.0 After that, it should be accessible.
Ievgenii
2025-10-12 10:48:45 +0000 UTCWhen your Ovi script finishes, it says: "Running on local URL: http//127.0.0.1:7891" but I can't connect. When I try to connect to 3000 via the RunPod console I get a Bad Gateway for https://tm5f7qf6tfokqw-3000.proxy.runpod.net/ indicating some problem with RunPod (my browser and Cloudflare working).
Star Hawk
2025-10-12 10:03:23 +0000 UTCI'm trying to use the RunPod setup but it's not connecting to port 3000. When I do a "curl localhost:3000" I get fail to connect but when I do curl on port 7891 I get a bunch of html that looks like it would be the Gradio interface. I get the same on different computers, browsers and wifi/ethernet connections. I'm guessing I'm making some simple mistake but idk. Thanks in advance for help and I really appreciate all your amazing work and efforts!
Star Hawk
2025-10-12 09:51:12 +0000 UTCGetting this with OVI Installer and i have 5090 ... [*] Installing PyTorch 2.7.0 + cu128 Looking in indexes: https://download.pytorch.org/whl/cu128, https://pypi.org/simple WARNING: Retrying (Retry(total=4, connect=None, read=None, redirect=None, status=None)) after connection broken by 'ConnectTimeoutError(, 'Connection to download.pytorch.org timed out. (connect timeout=15)')': /whl/cu128/torch/ WARNING: Retrying (Retry(total=3, connect=None, read=None, redirect=None, status=None)) after connection broken by 'ConnectTimeoutError(, 'Connection to download.pytorch.org timed out. (connect timeout=15)')': /whl/cu128/torch/ WARNING: Retrying (Retry(total=2, connect=None, read=None, redirect=None, status=None)) after connection broken by 'ConnectTimeoutError(, 'Connection to download.pytorch.org timed out. (connect timeout=15)')': /whl/cu128/torch/ WARNING: Retrying (Retry(total=1, connect=None, read=None, redirect=None, status=None)) after connection broken by 'ConnectTimeoutError(, 'Connection to download.pytorch.org timed out. (connect timeout=15)')': /whl/cu128/torch/ WARNING: Retrying (Retry(total=0, connect=None, read=None, redirect=None, status=None)) after connection broken by 'ConnectTimeoutError(, 'Connection to download.pytorch.org timed out. (connect timeout=15)')': /whl/cu128/torch/ ERROR: Could not find a version that satisfies the requirement torch==2.7.0+cu128 (from versions: 1.11.0, 1.12.0, 1.12.1, 1.13.0, 1.13.1, 2.0.0, 2.0.1, 2.1.0, 2.1.1, 2.1.2, 2.2.0, 2.2.1, 2.2.2, 2.3.0, 2.3.1, 2.4.0, 2.4.1, 2.5.0, 2.5.1, 2.6.0, 2.7.0, 2.7.1, 2.8.0) ERROR: No matching distribution found for torch==2.7.0+cu128 ERROR: PyTorch install failed.
mustispustis
2025-10-11 22:33:21 +0000 UTCdid you know why he keep runing it on the local URL: http://127.0.0.1:7891 on runpod? so i can't open it with the port : 3000. and he say : To create a public link, set `share=True` in `launch()`.
Erwan Cozic
2025-10-11 21:16:47 +0000 UTCdidnt he say near the beginning of the vid that you need a 5090
Jeffrey
2025-10-10 20:44:25 +0000 UTCI get the following Error when I try and install the FP8 version. The full version will install but doesn't seem to work on my 4090. ===================================================== OVI - INSTALLER ===================================================== REQUIREMENTS Windows 10/11 64-bit NVIDIA GPU (Ampere / Lovelace / Blackwell) NVIDIA driver ~560+ (CUDA 12.x) Python 3.10 64-bit in PATH (needed for FA wheels) FlashAttention policy: Blackwell (RTX 50xx or RTX 6000 Pro or GB/B series): marcorez8 wheel Non-Blackwell (RTX 30/40 or RTX 6000 Ada): petermg wheel If FA fails, we enable PyTorch SDPA fallback so the UI still runs. ===================================================== Press Enter to continue. Detected GPU: NVIDIA GeForce RTX 4090 Enable CPU offload to reduce VRAM use - press Y or N: Y Is your GPU Blackwell (RTX 50xx or RTX 6000 Pro) - press Y or N: N Download FP8 quantized weights and run with --fp8 - press Y or N: Y ===== Starting install with CPU_FLAG: --cpu_offload (Blackwell=0) (FP8=1) ===== [*] Cloning Ovi Cloning into 'Ovi'... remote: Enumerating objects: 647, done. remote: Counting objects: 100% (174/174), done. remote: Compressing objects: 100% (91/91), done. remote: Total 647 (delta 119), reused 121 (delta 82), pack-reused 473 (from 1) Receiving objects: 100% (647/647), 84.65 MiB | 45.41 MiB/s, done. Resolving deltas: 100% (299/299), done. [*] Applying Windows fix for io_utils.py [OK] Patched ovi\utils\io_utils.py [*] Creating virtual environment with Python 3.10 [*] Upgrading pip, setuptools, wheel Requirement already satisfied: pip in c:\ovi\venv\lib\site-packages (23.0.1) Collecting pip Using cached pip-25.2-py3-none-any.whl (1.8 MB) Requirement already satisfied: setuptools in c:\ovi\venv\lib\site-packages (65.5.0) Collecting setuptools Using cached setuptools-80.9.0-py3-none-any.whl (1.2 MB) Collecting wheel Using cached wheel-0.45.1-py3-none-any.whl (72 kB) Installing collected packages: wheel, setuptools, pip Attempting uninstall: setuptools Found existing installation: setuptools 65.5.0 Uninstalling setuptools-65.5.0: Successfully uninstalled setuptools-65.5.0 Attempting uninstall: pip Found existing installation: pip 23.0.1 ERROR: Exception: Traceback (most recent call last): File "C:\Ovi\venv\lib\site-packages\pip\_internal\cli\base_command.py", line 160, in exc_logging_wrapper status = run_func(*args) File "C:\Ovi\venv\lib\site-packages\pip\_internal\cli\req_command.py", line 247, in wrapper return func(self, options, args) File "C:\Ovi\venv\lib\site-packages\pip\_internal\commands\install.py", line 507, in run installed = install_given_reqs( File "C:\Ovi\venv\lib\site-packages\pip\_internal\req\__init__.py", line 68, in install_given_reqs uninstalled_pathset = requirement.uninstall(auto_confirm=True) File "C:\Ovi\venv\lib\site-packages\pip\_internal\req\req_install.py", line 663, in uninstall uninstalled_pathset = UninstallPathSet.from_dist(dist) File "C:\Ovi\venv\lib\site-packages\pip\_internal\req\req_uninstall.py", line 527, in from_dist paths_to_remove.add(path) File "C:\Ovi\venv\lib\site-packages\pip\_internal\req\req_uninstall.py", line 329, in add path = os.path.join(normalize_path(head), os.path.normcase(tail)) UnboundLocalError: local variable 'tail' referenced before assignment ERROR: pip upgrade failed. Press any key to continue . . .
Lance Fanger
2025-10-10 20:08:50 +0000 UTCJust so I know I'm not going crazy, no comfyui workflow?
James Clow
2025-10-09 20:45:30 +0000 UTCThis, runs great on my 3090, most painless installer ive found for Ovi
Sense Departed
2025-10-09 19:48:13 +0000 UTCIs there any way to increase the lenhgt of the video?
Reinis Salmiņš
2025-10-09 15:07:52 +0000 UTCIm literally running it on a 24 gb 3090 with no issue at all, LW. Just generated a bunch of videos. Just takes 5 min or so per video, not bad at all. Would be nice if we could do longer than 5 second videos, but hey, its a start and a good one at that! Thanks for Aitrepereneur for yet another amazing one click install that works flawlessly!
Paul Mazurkiewicz
2025-10-09 02:36:24 +0000 UTCLooks fun. I'll keep an eye out for the version which will run locally on a 24gb 3090 :)
LW
2025-10-08 09:14:37 +0000 UTCYou mentioned the chatgpt prompt to prime the text part but I could not locate it, perhaps I missed? Thanks
VINCENT ALONZO
2025-10-07 19:07:26 +0000 UTCi have an issues with the runpod set up i keep getting this error Directory not found: "workspace" whenever it gets to pytorch
Peter Greene
2025-10-07 17:26:32 +0000 UTCto stop the public link, (and why would ANYONE want a public link) thusly invoking the Frpc thing, just edit the "gradio_app.py" file with note pad, go to last line, change share=True to share=False.
Brad
2025-10-07 15:18:01 +0000 UTCis ovi's max resolution still 992?
Kenny
2025-10-07 02:32:57 +0000 UTCweird...dm me
Aitrepreneur
2025-10-06 23:13:40 +0000 UTCThe FP8 version doesn't seem to want to launch. The CMD window just closes. If I install the Full version it will launch and run on my 4090... kinda.
Sena
2025-10-06 22:59:27 +0000 UTCnope that's the limit of the model for now but they are working on a longer video model
Aitrepreneur
2025-10-06 22:20:51 +0000 UTCThat’s not a virus. Gradio webui uses frpc to create public gradio links. So it's normal
Aitrepreneur
2025-10-06 22:20:18 +0000 UTCyour bat is downloading a hacktool according to malwarebytes https://gyazo.com/ae3fd89770b0188804b324cb96fdb707
silent chaos
2025-10-06 19:40:12 +0000 UTCMoving that file into utilities did it, thanks. my 5090 is happy again. any way to go past 5 seconds?
Brad
2025-10-06 19:07:06 +0000 UTCJust downloaded the latest installer and can confirm it works on my 5090.
Speedy2023
2025-10-06 04:41:32 +0000 UTCCan confirm that this is working on rtx pro 6000
ale mart
2025-10-05 23:25:15 +0000 UTCtmpbnhcxtfn.wav it's funny it says permission denied, I checked, it is not even there, so the "correct" error should have read, File not found. how about a little help :)
Brad
2025-10-05 22:14:14 +0000 UTCThis means you aren't using the latest installer
Aitrepreneur
2025-10-05 21:56:25 +0000 UTCError during video generation: [Errno 13] Permission denied: 'C:\\Users\\User\\AppData\\Local\\Temp\\tmpdu_n6zvg.wav'
Brad
2025-10-05 21:05:30 +0000 UTCyes. but you can also dl them from here: https://huggingface.co/Aitrepreneur/test/resolve/main/LAUNCHER-OVI.bat?download=true and the fp8 launcher: https://huggingface.co/Aitrepreneur/test/resolve/main/LAUNCHER-FP8-OVI.bat?download=true
Aitrepreneur
2025-10-05 20:40:37 +0000 UTCFixed with this: https://www.python.org/downloads/release/python-31011/ Download the "Windows installer (64-bit)" version and check the "Add to path"
uniqornate
2025-10-05 20:19:14 +0000 UTCIsn't the INSTALL.OVI+FP8.bat should automatically install the right python version via .venv ? ERROR: Python 3.10 (64-bit) not found. Please install Python 3.10 (64-bit) and ensure either the "py" launcher or python3.10.exe is available. Press any key to continue . . .
uniqornate
2025-10-05 20:14:38 +0000 UTCdoes this generate a "launch.bat"?
Brad
2025-10-05 19:53:44 +0000 UTCI just reuploaded the new installer try that one out
Aitrepreneur
2025-10-05 19:43:00 +0000 UTCI just reuploaded it
Aitrepreneur
2025-10-05 19:42:47 +0000 UTChi mate, where is the bat file ? or sh can run on windows too ?
AN
2025-10-05 18:47:29 +0000 UTCwill it include the "launch.bat" file?
Brad
2025-10-05 18:34:25 +0000 UTCgive me a few minutes, I'll upload it back it
Aitrepreneur
2025-10-05 18:30:46 +0000 UTCWhere can i find the local install file?
Kenny
2025-10-05 18:29:32 +0000 UTCI added support for the fp8 model and fixed the access the Temporary Wav file error as well
Aitrepreneur
2025-10-05 18:01:12 +0000 UTCno worries, I already just fixed the installer, it works now. I also added support for the fp8 model and fixed the access the Temporary Wav file (which is a repository issue, the guys still haven't fixed it for some reason) so just use the new installer or if you have everything already you can just dl this file: https://we.tl/t-YSokRXRqpm put that file in the Ovi\ovi\utils folder and replace the other file you can make a copy of the other just in case
Aitrepreneur
2025-10-05 18:00:25 +0000 UTCHonestly, I am going to just give up for a while. I even downloaded the fp8 model manually but i still get an error that permission is denied and it can't access the Temporary Wav file. i should just wait a few days for all of the bugs to be worked out.
Bradford
2025-10-05 17:36:10 +0000 UTCTriton needs to be installed in your virtual environment. F:\Ovi\venv\Scripts\activate (venv) F:\Ovi\python.exe -m pip install -U --pre triton-windows
David
2025-10-05 17:33:57 +0000 UTCsadly cannot do a SS but no launch.bat file in the OVI folder
Brad
2025-10-05 16:49:07 +0000 UTCgo into the OVI folder and there is a file called LAUNCHER-OVI.bat you can double click on it or copy it and paste a shortcut on your desktop, I believe.
Bradford
2025-10-05 16:32:13 +0000 UTCit still closes once done and does not launch, does it put a "launch.bat" or similar somewhere?
Brad
2025-10-05 16:27:55 +0000 UTCHello. It looks like an FP8 option has been added to the OVI software, but your installation script does not have an option to download the FP8 model.
Bradford
2025-10-05 16:06:24 +0000 UTCIs there any way to give more quality to the outputs? My generated videos are terrible in quality..
Esteban
2025-10-05 15:54:36 +0000 UTCyou got an older version of the script anyway
Aitrepreneur
2025-10-05 15:09:26 +0000 UTCI posted the bat into Chatgpt is suggested the following: What actually happens after you press Y The Y/N is read with set /p (you must press Enter after Y). Then the script immediately hits prechecks that can abort before any download starts: Line 57: where git >nul 2>nul || (echo [ERROR] Git not found... & pause & exit /b 1) Line 58: where python >nul 2>nul || (echo [ERROR] Python 3.10 (64-bi...in PATH... & pause & exit /b 1) If either Git or Python 3.10 (64-bit) isn’t on PATH, it prints the error, waits for a key (pause), then exits. Double-clicking the BAT makes this look like it “closes instantly” and “never starts downloading.” (Verified from your file at those exact lines.) After that, the first actual “download the program” step is the git clone: Lines 63–66: if Ovi\download_weights.py doesn’t exist, it does git clone ... || (echo [ERROR] Clone failed. & pause & exit /b 1) So the usual reasons it “fails right after Y”: Git not installed / not on PATH → exits at line 57. Python not on PATH (or wrong major/minor; the wheels are for 3.10) → exits at line 58. Norton blocks git, curl, or powershell Invoke-WebRequest → clone or later downloads fail and the window closes after the pause. You pressed Y but didn’t press Enter (the prompt at line 46 requires Enter). That leaves %CPU_FLAG% empty (harmless), and you go straight into the failing precheck above. Make it stay open and fix the failure A) Keep the window open (so you can read the error) Replace the last block with a guaranteed pause (so success or fail, it won’t vanish):
Brad
2025-10-05 14:53:47 +0000 UTCdm me the full error log
Aitrepreneur
2025-10-05 14:28:23 +0000 UTCStill errors getting " [*] Installing FlashAttention for BLACKWELL... Collecting flash-attn==2.7.4.post1+cu128.torch270 ERROR: HTTP error 404 while getting https://huggingface.co/marcorez8/flash-attn-windows-blackwell/resolve/main/flash_attn-2.7.4.post1+cu128.torch270-cp310-cp310-win_amd64.whl ERROR: Could not install requirement flash-attn==2.7.4.post1+cu128.torch270 from https://huggingface.co/marcorez8/flash-attn-windows-blackwell/resolve/main/flash_attn-2.7.4.post1+cu128.torch270-cp310-cp310-win_amd64.whl because of HTTP error 404 Client Error: Not Found for url: https://huggingface.co/marcorez8/flash-attn-windows-blackwell/resolve/main/flash_attn-2.7.4.post1+cu128.torch270-cp310-cp310-win_amd64.whl for URL https://huggingface.co/marcorez8/flash-attn-windows-blackwell/resolve/main/flash_attn-2.7.4.post1+cu128.torch270-cp310-cp310-win_amd64.whl Collecting flash-attn==2.7.4.post1 ERROR: HTTP error 404 while getting https://huggingface.co/marcorez8/flash-attn-windows-blackwell/resolve/main/flash_attn-2.7.4.post1-cp310-cp310-win_amd64.whl ERROR: Could not install requirement flash-attn==2.7.4.post1 from https://huggingface.co/marcorez8/flash-attn-windows-blackwell/resolve/main/flash_attn-2.7.4.post1-cp310-cp310-win_amd64.whl because of HTTP error 404 Client Error: Not Found for url: https://huggingface.co/marcorez8/flash-attn-windows-blackwell/resolve/main/flash_attn-2.7.4.post1-cp310-cp310-win_amd64.whl for URL https://huggingface.co/marcorez8/flash-attn-windows-blackwell/resolve/main/flash_attn-2.7.4.post1-cp310-cp310-win_amd64.whl << was unexpected at this time."
MadHamster
2025-10-05 14:25:44 +0000 UTCAlright...I uploaded a new version, dl that one. Please dm me to le me know if that one worked, I need people with 5090 to test it for me since I can't do it myself right now
Aitrepreneur
2025-10-05 14:20:22 +0000 UTCAlright...I uploaded a new version, dl that one. Please dm me to le me know if that one worked, I need people with 5090 to test it for me since I can't do it myself right now
Aitrepreneur
2025-10-05 14:20:18 +0000 UTCAlright...I uploaded a new version, dl that one. Please dm me to le me know if that one worked, I need people with 5090 to test it for me since I can't do it myself right now
Aitrepreneur
2025-10-05 14:20:15 +0000 UTCbatch file exits with Detected GPU: NVIDIA GeForce RTX 5090 Enable CPU offload to reduce VRAM use [y/N]? y ===== Starting install with CPU_FLAG: --cpu_offload ===== not was unexpected at this time.
Alex Pearce
2025-10-05 13:32:07 +0000 UTCI have the 5090, I select Y to let it offload to cpu as you suggest and the cmd window closes.
Brad
2025-10-05 13:24:34 +0000 UTCIn every case, I get the following error when generating: Pure T2V mode: calculated video latent size: 32 x 62 0it [00:00, ?it/s] ERROR:root:Traceback (most recent call last): File "C:\Users\Norbi\Downloads\Ovi\ovi\ovi_fusion_engine.py", line 235, in generate pred_vid_pos, pred_audio_pos = self.model( File "C:\Users\Norbi\Downloads\Ovi\venv\lib\site-packages\torch\nn\modules\module.py", line 1751, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "C:\Users\Norbi\Downloads\Ovi\venv\lib\site-packages\torch\nn\modules\module.py", line 1762, in _call_impl return forward_call(*args, **kwargs) File "C:\Users\Norbi\Downloads\Ovi\ovi\modules\fusion.py", line 294, in forward vid, audio = gradient_checkpointing( File "C:\Users\Norbi\Downloads\Ovi\ovi\modules\model.py", line 21, in gradient_checkpointing return module(*args, **kwargs) File "C:\Users\Norbi\Downloads\Ovi\ovi\modules\fusion.py", line 187, in single_fusion_block_forward audio_y = audio_block.self_attn( File "C:\Users\Norbi\Downloads\Ovi\venv\lib\site-packages\torch\nn\modules\module.py", line 1751, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "C:\Users\Norbi\Downloads\Ovi\venv\lib\site-packages\torch\nn\modules\module.py", line 1762, in _call_impl return forward_call(*args, **kwargs) File "C:\Users\Norbi\Downloads\Ovi\ovi\modules\model.py", line 253, in forward x = flash_attention( File "C:\Users\Norbi\Downloads\Ovi\ovi\modules\attention.py", line 118, in flash_attention assert FLASH_ATTN_2_AVAILABLE AssertionError Error during video generation: cannot unpack non-iterable NoneType object
Adam White
2025-10-05 13:21:21 +0000 UTCwhen you restart you can simply go inside the ovi folder, open a terminal and inside type source .venv/bin/activate python3 gradio_app.py --cpu_offload you don't have to relaunch the installer
Aitrepreneur
2025-10-05 12:20:26 +0000 UTCyes I know...I just reupdated the installer again and using some other third party flash attention wheels for blackwell gpus, hopefully this time it works.. make sure you got python 3.10 and cuda 12.8 installed for it
Aitrepreneur
2025-10-05 12:19:42 +0000 UTCThanks for the installer. Worked fine until I tried running it. It can't generate because of some errors with flashattention. I have a 5090 and used the one-click installer.
troulertrian
2025-10-05 12:10:16 +0000 UTCQwen + Wan combo, I've made a video about it. You need a model that can follow your prompt precisely, this way you can train precise loras and get a variety of styles
Aitrepreneur
2025-10-05 11:50:27 +0000 UTCyou can stop the pod then when you restart you can simply go inside the ovi folder, open a terminal and inside type source .venv/bin/activate python3 gradio_app.py --cpu_offload
Aitrepreneur
2025-10-05 11:49:19 +0000 UTCI installed on RunPod with a network storage attached. First run was fine, but after rebooting the server and re-running the install file to initialise port 3000 I get the following error. /workspace/Ovi/.venv/lib/python3.12/site-packages/torch/cuda/__init__.py:182: UserWarning: CUDA initialization: CUDA driver initialization failed, you might not have a CUDA gpu. (Triggered internally at /pytorch/c10/cuda/CUDAFunctions.cpp:109.) return torch._C._cuda_getDeviceCount() > 0 /workspace/Ovi/ovi/modules/model.py:37: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead. @amp.autocast(enabled=False) /workspace/Ovi/ovi/modules/model.py:47: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead. @amp.autocast(enabled=False) /workspace/Ovi/ovi/modules/model.py:72: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead. @amp.autocast(enabled=False) /workspace/Ovi/ovi/modules/model.py:102: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead. @amp.autocast(enabled=False) Traceback (most recent call last): File "/workspace/Ovi/gradio_app.py", line 4, in from ovi.ovi_fusion_engine import OviFusionEngine, DEFAULT_CONFIG File "/workspace/Ovi/ovi/ovi_fusion_engine.py", line 13, in from ovi.utils.model_loading_utils import init_fusion_score_model_ovi, init_text_model, init_mmaudio_vae, init_wan_vae_2_2, load_fusion_checkpoint File "/workspace/Ovi/ovi/utils/model_loading_utils.py", line 6, in from ovi.modules.fusion import FusionModel File "/workspace/Ovi/ovi/modules/__init__.py", line 3, in from .t5 import T5Decoder, T5Encoder, T5EncoderModel, T5Model File "/workspace/Ovi/ovi/modules/t5.py", line 472, in class T5EncoderModel: File "/workspace/Ovi/ovi/modules/t5.py", line 478, in T5EncoderModel device=torch.cuda.current_device(), ^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/Ovi/.venv/lib/python3.12/site-packages/torch/cuda/__init__.py", line 1071, in current_device _lazy_init() File "/workspace/Ovi/.venv/lib/python3.12/site-packages/torch/cuda/__init__.py", line 412, in _lazy_init torch._C._cuda_init() RuntimeError: CUDA driver initialization failed, you might not have a CUDA gpu.
Ryan Peplow
2025-10-05 11:46:00 +0000 UTCEasy to say when I don't have access to it right now
Aitrepreneur
2025-10-05 11:45:57 +0000 UTCmultiple users with local 5090 report the same error, please put your 5090 in and test before posting installers.
Ryoga Hoshino
2025-10-05 11:21:42 +0000 UTCHi, i am new to the patreon. Nice to meet you all. I wanted to make a question on the OVI runpod installation: if we just use the runpod without adding a Network Storage, we will have to install everything again everytime we want to come back and generate more videos, right? Or is there something i didnt get right?
Alex G New Media Journalism
2025-10-05 11:17:30 +0000 UTCplease share your favorite model . i try other models regularly but have not found something as beautifully and complete as flux.
rubz
2025-10-05 06:33:51 +0000 UTCI updated the installer, dl it again
Aitrepreneur
2025-10-05 01:11:50 +0000 UTCyeah but Flux has been obsolete for quite some time now...
Aitrepreneur
2025-10-05 01:11:29 +0000 UTCyes, almost ready basically
Aitrepreneur
2025-10-05 01:11:07 +0000 UTCI just tried to use the flag "--use_image_gen" but it doesn't work. Apparently it is trying to access a non-public model which throws a fatal error. Therefore, I will take an image from a video I generated using Wan22.
Bradford
2025-10-05 00:18:24 +0000 UTC5090 card and it is at 2546.9 seconds and counting, still no image, 64gig of regular ram too. this is a secondary computer but I am calling it.
Brad
2025-10-04 23:17:32 +0000 UTCthis is great! but Flux ultimate workflow kinda would like an update :)
rubz
2025-10-04 22:47:07 +0000 UTCit seems the installer crashes for me. I get this error before it auto crashes. ERROR: flash_attn-2.7.4.post1+cu128.torch270-cp310-cp310-win_amd64.whl is not a supported wheel on this platform.
Virtamouse
2025-10-04 22:34:45 +0000 UTCAny plans for a WAN animate video?
Mr Morphin Time
2025-10-04 22:22:36 +0000 UTCanime is not great no unfortunately, but I know they are working on a better model already, we might soon have wan 2.5 as well (I can't wait for that one tbh)
Aitrepreneur
2025-10-04 21:24:48 +0000 UTCit's in the other post: https://www.patreon.com/posts/140447944
Aitrepreneur
2025-10-04 21:23:30 +0000 UTCthis is not a comfyui installer, it's its own webui that requires 32gb of vram to run for now, highly suggest watching the video if you haven't already
Aitrepreneur
2025-10-04 21:23:12 +0000 UTCCan it do anime or is it still being developed? Sora 2 is useless now because of copyright unless you want to see Sam Altman 8 thousand times a day
Jim Settle
2025-10-04 21:12:05 +0000 UTCHi thanks for the great video. I can not find the prompt template you mentioned anywhere, can you help?
Gavin
2025-10-04 20:56:52 +0000 UTCCouple questions if you have a sec... Is this the same as last time where the installer only works with portable? Do you have one for regular ComfyUI also? Should work well with a Blackwell, right? Would I want to tell it to offload with that much VRAM? Thanks!!
Brian Monarch
2025-10-04 19:56:18 +0000 UTCit needs a connection to dl the model but after that it shouldn't
Aitrepreneur
2025-10-04 19:29:40 +0000 UTCYou just need to be patient and wait, it takes a few minutes after seeing the initial url to actually start completely
Aitrepreneur
2025-10-04 19:29:12 +0000 UTCOf course
Aitrepreneur
2025-10-04 19:28:30 +0000 UTCYes it is
Aitrepreneur
2025-10-04 19:28:24 +0000 UTCI'll check it out
Aitrepreneur
2025-10-04 19:28:17 +0000 UTCdoes this require internet connection if I am using a 5090 card? it runs but does not make a video.
Brad
2025-10-04 19:09:48 +0000 UTCNo matter how many times I try installation with Runpod, it always only start on 0.0.0.0:3000 and not also on xxxxxxx.gradio.live, so when I open port 3000 I just get message that port is not ready :(
Jack Kennedy
2025-10-04 19:03:19 +0000 UTCOne thing to note is that this is creating a publicly accessible link if anyone has the string. While the link is convoluted and unlikely to be accessed from anyone on the outside, if anyone wants to disable it they need to edit their 'gradio_app.py' file and scroll to the bottom to where it says "demo.launch(share=True)" and change "True" to "False" Thank you based aitrepreneur
J
2025-10-04 18:49:58 +0000 UTCThis is awesome! Will you update us when the quantized version comes out? I'd love to run this on my 5070ti
guildofkings
2025-10-04 18:37:35 +0000 UTCI will be waiting for Quantized with my 4070, but this is exciting
Rick Doman
2025-10-04 18:35:45 +0000 UTCHow about this model? https://huggingface.co/wavespeed/Ovi-e4m3_e4m3_dynamic_per_tensor
nurtu
2025-10-04 18:15:02 +0000 UTCIt should yeah
Aitrepreneur
2025-10-04 17:59:26 +0000 UTCHey there, thank you for all the hard work. I have a RTX 6000 ADA. it should work, correct?
scifimovie2019
2025-10-04 17:56:22 +0000 UTC
