[edit:
13.05.2026: Update version 4.4 (see version description).
Small fixes to get back fast generations.
Attention:
If you struggle with node conflicts or you get errors while running the workflow, please have a look at my short Trouble Shooting Guide note in the wokflow first. Most importent is to update all components sucsessfully! ]
Special thanks to:
@ArcleinSK for investigation and solving the FLF issue, as well as forcing the First-Mid-Last Frame option and last but not least for charing fantastic knowlage.
@boinobin730 for initialising, forcing and supporting this project in all kinds of matter, like providing links, running tests, sharing knowlage and inspiring diskussions.
@Urabewe for publishing the original, perfectly running 12 GB VRAM LTX-2.3 workflows mainly used here in this workflow.
Features:
Simple to use all-In-One LTX-2 workflow with options for:
Text to Video
Image to Video
First/Last Frame to Video
Fisrt/Mid/Last Frame to Video
Video to Video
Text + Audio to Video
Image + Audio to Video
First/Last Frame + Audio to Video
First/Mid/Last Frame + Audio to Video
easy switching between all options,
all steps highly automated: no manual frame or width/hight calculations necessary,
easy to set inputs by predefined sliders and aspeckt ratio inputs (no risk to set wrong frame counts or wrong width/hight values),
completely automated resizing and cropping (if necessary) of your input images/videos.
brilliant audio generation (speech/sound) with LTX-2.3.
LTX-2.3 specifications:
Workflow version v4.3 consistently follows the LTX-2.3 specifications for 16:9/9:16 aspect ratios, including automatic width/hight calculations, as well as automatic input image/video resizing/cropping.
In addition you can simply choose now any other aspect ratios according to your needs while still getting the right values calculated for width/hight and automatic image/video resize/crop.
Requirements:
GPU with 12 GB VRAM (some users reported they got it running with 8 GB too),
32 GB VRAM,
Swap file size: 64 - 128 GB.
Speed and video length:
Runs very fast: 5 second (1280 x 864) Video: < 10 minutes.
Generation of long high quality videos in one run possible: 10 - 20 seconds without any issues,
Testrun: 30 second video (1024 x 704) tooks around 40 minutes without any OOM errors. Longer videos might be possible, but not tested yet.
Important:
This workflow is intended for advanced comfyui users who know how to install and operate the system and are able to resolve basic system errors themselves, like as node conflicts, or general system issues.
About this workflow:
This workflow is mainly based on the fantastic LTX-2.3 workflows of @Urabewe.
As far as I know, those were the first workflows running LTX-2 with 12 GB VRAM. All credits goes to the original creator.
My job was only to combine and organise the different workflows in a simple to use all-in-one design.
Description
Wrong frame calculation in v2v fixed.
FAQ
Comments (59)
what's the difference between 1.2 and 1.3?
First look at the published version information 😉 Fixed calculation.
Oops. I didn't play with v2v yet so I didn't notice. Thank you
@AIden_AIzawa Hi, I saw you used special audio syntax, like: \"audio prompt\" in your recent clip and you was able to address different speakers. This looks very interesting. Is there any ltx documentation about this??
Sorry I saw only now your question. Anyway, no. It's just copy and paste from the metadata that change the format. In the prompt I use only "
Sometimes audio is not optimal and mixes voices, change tone etc ... I usually ask to chatgpt or grok to be detailed in describing who is talking, lipsync or off screen dialogues. Based on my experience also specifying dialect helps to make the voice more realistic compared to plain english
@AIden_AIzawa Thank you. Ahh - ok, I thought you found some special syntax for ltx too. Cause I experimented with different audio engines some times ago and there are special "tags" to control audio, like "pause", "laughing", "whisper", etc.
I'm really amazed at how well changing the narrator works for you. I have to try this too 🙂
Good idea to use AI for helping with dialogs 👍 And yes, experimenting with dialects, voices ranges and emotions is a lot of fun. Even all kinds of languages works very well. I did some tests for fun 😅
I encountered an issue when I use your default models. The error shows :RuntimeError: Input type (struct c10::BFloat16) and bias type (struct c10::Half) should be the same. I asked ChatGPT and it said:
"In the preview phase, Latent is passed into the VAE decoder as BF16.
But:
Decoder. weight/bias is fp16
Conv2d does not allow:
bf16 input + fp16 bias
So it exploded."
And I noticed there's another VAE model Wan2.1_vae in your instruction, so I changed to this and finally it works. But I don't know the difference between these 2 vae models.
@ggcutefoot Sorry, I don`t got about wich vae models you really talk.
1. In the LTX workflow we have to use 2 vae models: audio + video. The linked models and the model loaders in the workflow are correct.
2. "Model Links" note -> "File save location": You are right - here in the description for the file locations I accidently mentioned a "third" vae model (wan2.1). This is wrong of course. Thank you for the hint. I will fix this typo error in the next version.
So it seems you have done some model mismatch in your first attempt.
But: "And I noticed there's another VAE model Wan2.1_vae in your instruction, so I changed to this and finally it works" This is definately "nonsens" 😉The workflow works only with the linked LTX models. Please check your loader nodes to be sure about.
lançou o LTX 2.3
Oh my - the early bird 😂 Yeah, tried allready to implement the new models, but this is all quite more tricky 🙄
@arkinson I was going to ask the same thing since I have been using your workflow for some time now.
@arkinson (༼ つ ◕_◕ ༽つ TAKE MY ENERGY ༼ つ ◕_◕ ༽つ)
@Rifler1 Yeah - got your magic energy from behind the mountains 🤣 Urabewe, the creator of the originall gguf workflows, did a fantastic job in the meantime. With his solutions it is quite easy to implement the necessary changes now. Except from accidentally struggling with a wrong model for some hours, I am on the way now. If all final testing works well I will publish it probably tomorrow...
@Agino @Rifler1 @LewdAnimeEmpire @boinobin730 LTX-2.3 workflow v1.0 is published now as "alpha". All options should work properly. As mentioned in the version desription, please let me know if you find any bugs/errors or other strange things.
I have not checked, if there allready new ltx-2.3 Loras out. If you find something interesting, please let me know. I did some quick test with the image2vid adapter Lora, but all I got was a more blurry output....
@arkinson I got blurry output on a i2v as well as a t2v. I have disabled the detailer lora. trying again. https://huggingface.co/Kijai/LTX2.3_comfy/tree/main/loras this is new. I am downloading it now.
@boinobin730 (This conversation started here)
I used one of urabewes prompts and it gave a brilliant output. So, right prompting seems to be a "must do". I have to learn this for myself.
What the hell are you testing? A woman with man`s voice? 🤣No, believe me, the AI is right! A woman has a female voice and a man has a male voice. Enough!😂🙂
Thank you for the hint with the negative prompt. I allways thought it is not needed with LTX, but I just saw the template workflows have it. No problem, I will add this soon. I even will have a look at the video preview....
@boinobin730 Thank you for the Lora link. Seems I have to update quickly 🙄
@boinobin730 Blurry output: There must be something wrong on your side, here it works. Wrong Lora weight, any wrong model???
I`m working on the workflow update:
- negative prompt,
- generation preview,
- new ltx-2.3 Lora.
Just tested for t2v. Works fine so far. Will try to update soon.
@arkinson ok. thanks. urabewes workflow works ok. exactly the same models, text encoder, clip, vae.... ok i will play with it.
@boinobin730 Wich option do you use - t2v?
@arkinson just the basics. t2v. at the moment. i am trying urabewes again and see if im going crazy.. urabewes was working ok yesterday.
@arkinson Fixed it. For everyone be sure to change your Latent Upscale model to
ltx-2.3-spatial-upscaler-x2-1.0.safetensors instead of the old version. 2.0 get it from here.
@boinobin730 Yes - or just click the link in the workflow 😉 I`m glad you found the issue.
Btw. workflow version v1.1 is out now. I believe we have to run tests with the strength of the new destilled Lora and the use of additional ltx-2 Loras. I only have tested the generall functionallity of all options yet.
@arkinson Great. thanks. link needs to be adjusted in the new workflow 1.1 for spatial upscaler, still refers to the old 2.0 upscaler page. But other than that, it should be good. I'm just running a I2V on the old ltx2.3 workflow. i will test out the new 1.1 Preview helps a great deal when trying for a generation and you see crap. You can just abort and try again.
@PopHorn1956 @arkinson I did a generation on the new workflow 1.1, but the processing time was 50 minutes for 10seconds. something has changed. 1.0 didn't take that long. Preview Vae is only thing I can think of. I am going to keep playing around with it.
@PopHorn1956 @arkinson I am not getting slow mo- just long gen times on outputs. the old model 1.0 is fine. its not the preview vae. As I disabled it and it was still really slow for me. I think its the negative prompt or the way its handled. Since I turned it off and it generated a 10second clip in 436seconds. I didn't have to change the LTXpreview value.. still left at 0.
@boinobin730 @PopHorn1956 Thank you for fast testing 👍
Wrong latent upscale model link: boinobin730 you are right. Just added a warning/note to the model description.
Slow motion/LTXVPreproces: PopHorn1956 did you some serious side-by-side tests under exact same conditions? Cause the value is only for image compression rate and should have no influence at the video motion.
Extreme long generation times: I had the same issue after adding the generation preview part, caused by higher vram usage > 12 gb. My first thought was, the reason must be the generation preview of course. But different tests have not confirmed this. Finally, after a complete window reboot, everything worked fine. Please, let us have some more tests here.
@PopHorn1956 To be honest, I did not run much tests for myself yet. So your inputs are very usefull. Generally I have the impression that I achieve poorer results with ltx-2.3. In quick pre tests I saw similar issues, like you desribed: slow/no motion at all, unsynchronised speech, etc. I even did a quick test with the old Lora Image2Vid-Adapter, but got blurry outputs. On the other hand, what worked out of the box, was a detailed prompt from the original urabewe workflows with a structured desription of style, scene, character movements and camera movements. I have not dived in yet, but I believe ltx-2.3 is much more prompt "sensibel" All my simple/short ltx-2 prompts seems not to work well....
@arkinson The workflow is working after I rerouted the negative directly to each connection. I suppose deleting the negative get and set and redoing it may fix it if you are seeing similar errors. I have been mostly testing I2v and Image Audio to video. I will test some T2V. and post examples here.
@PopHorn1956 Yes, unfortunetaly. But have a look at urabewes prompts first (links to his page/workflows you find here in the model description.
@boinobin730 Strange. I checked the set/get nodes for i2v and they should be correct. Maybe you had mismatched something while playing around?
@arkinson I may have, as I have been playing around with the easy prompt node for variety. its working ok now.
@boinobin730 Easy-Prompt node: Are there any news about the vision part frontier actually? Last thing I tried still with ltx-2 was manually adding a short image description in the vision prompt field (inside the easy prompt node). But the results were not impressive.
Blhll gave me a good instruction how he use LM Studio to generate i2v prompts. Very interesting, but it works not automatically oc course and you have allways to switch between LM Studio and comfyui....
@arkinson the person who created the easyprompt updated the nodes, so it has a bit more functionality. The vision model now really does influence the actual prompt, so you do get a prompt that is based off the image file. But there are a few issues going on which I don't really like. Whenever it prompted it would always say rear shot and the camera is walking behind the subject yet the initial image is a forward facing shot. The 2nd problem is that it will still fail to lip sync with the voice if I am trying to do a image + Audio to video. If the vision node is attached. I think what i will end up doing is scrapping the LTX part of a workflow and only use the Easy prompt section to create a suitable prompt based off the vision and then copy paste into your workflow. It will give me better results I think.
@arkinson Also FYI, there is now an UD version of LTX V2.3 https://unsloth.ai/docs/basics/unsloth-dynamic-2.0-ggufs
https://huggingface.co/unsloth/LTX-2.3-GGUF/tree/main
I am not sure how much better these models are but it will be interesting to test and find out.
@boinobin730 No, no no! Not new models again! 🤣 I just found a little bit time to try a few own generatins after the "hype" of the last workflow releases 🙂 Except a couple of users with node/update issues, anything seems to work fine now. Some guys are really crazy, running it with 8 gb vram on a laptop 🙄
Actually I´m experimenting with "straigt hand made prompts" for some ideas. Not the lazy way, but some fun too.
I will give easy-prompt a new try soon. You wrote, you will use easy-prompt allone. I am afraid this will not work. At the beginning I tried to run the easy-prompt node just linked with a string/prompt output node. No no chance. It is strange, but it seems, easy-prompt will see a ksampler at least...
@arkinson ok. I didn't know that. It's a shame because the vision node is disrupting something, it seems to poison the workflow, because when I bypass the vision node, i still get that problem with no lip sync and just the audio playing. I delete the entire vision node totally and the workflow still won't give me lip sync on a Image + Audio to video. Crazy stuff. I tested it so many times as I didn't believe it.
@boinobin730 Yes, at least the last vision node version I tested, probably had killed my comfyui installation. I'm not sure, of course, but it's quite likely. After that, I started my third comfyui installation. I think that's quite more than just a little buggy 🙄 Did you try an OS reboot and a fresh workflow reload/download allready?
@arkinson So, I didn't quite have to nuke the entire install. I just deleted the workflow and loaded up a fresh workflow that was "untouched" by the vision node. I closed comfyui and reopened it. I still don't understand how it manages to damage the workflow. I didn't want to spend ages on it. Easy prompt by itself is ok for T2V and it doesn't wreck anything as far as I can tell. I will get around to testing the new UD versions of LTXV2.3 sometime....
@boinobin730 One reason for a "destroyed" workflow could be that there was something still in browser cache. Reloading the workflow might have overwrite the old cache. Just a guess. Did you tried to manually add some short image description in easy-prompt? My own tests were not very succsessfull, unfortunately.
@arkinson So what was interesting, I added a preview as text node to the vision node, primarily so I can see what the vision node is pumping out to the easy prompt. Fair enough, it gave me output, but what was interesting is even when the vision node is bypassed, an output is still being passed in the form of some code. I didn't do screen capture. I will test more if I have more time to kill. Prompt Daddy creator constantly updates his nodes, so it might have changed again.
Civitai seems to be having buggy issues, can't see gallery. I have some fun I+A2v I created plus some I2V. I will post when it is fixed.
I'm slowly getting more into flux 2 klein inpaints and I2V. Man, what a difference. I can really rejuvenate the images I have initially crafted. The quality of the keyframe can really make or break an LTX video. I will post the example as well.
@boinobin730 Vision node: As I allready mentioned (maybe not understandable) that was the way I went too: displaying the output from the vision node wich is mostly very simple. In the next step I deleted the vision node and manually added simple image descriptions in the text fields of the easy-prompt node. I did both: adding it in the upper text/prompt field and adding it in the lower text field (that one, the vision node is connected normally). I did not many tests. The first method gave some interesting random outputs. With the second method (what should be the better (or designed) way) my impression was that the results were even more random. But my tests were all not seroius, cause ltx-2.3 came on the market 😉
I´m actually experimenting a lot with z-image. Newest models tending more and more to realistic outputs. Spending a lot of time to improve my skills for z-image Lora training with AI Toolkit🙄
@arkinson thanks for the detailed description. I haven't tried the vision again yet. I mucked around with the easy prompt. It is pot luck. So A lot of the time I have spent stopping it and rewriting it.
Z-image stuff? sounds interesting. let me know how you get on. I tried making a lora of my blonde girl a few months back. It was ok. The biggest problem for me is that Z-image is a bit like SDXL, it isn't setup to make extreme bodies like I like em (i.e. off the shelf -no lora). So she ends up being pretty normal looking. Which is fine if thats what you are after. Pony and Illustrious are my favourites still for getting a character. I finally found a really good flux klein easy to use inpaint workflow. for i2i. I found that SeedVr2 can be a bit too over the top plasticky at times. I wanted an easy to use workflow where I can do a bit of head swap fix up, inpaint with mask, remove moire/ interlacing. So Using the outputs of this I created some really nice LTX videos. See my new stuff on your gallery if Civitai is fixed. Check out the mad max one. The first one is the best one I+audio2Video. The keyframe is really good and came from a very poor initial image that looked very fake. Check out the blue sky version with her on the truck. You can see the shadow over the fur coat, her mask looks fake, the eyes look fake. After throwing that image through the Klein workflow, things look a lot better.
The other examples is of the my blonde girl with friends. I spent a long time on that initial AI image, as was from a stock photo as a depth map then put through a pony workflow. The initial image was rubbish, put it through Qwen, then put it through Klein about 5 times doing inpaints for head swaps, boob increases, better hands etc etc. I just wanted to see how good I could get it. In the end, with a decent keyframe, it looked really good in LTX and the sound came up ok too.
Still can't believe there are so many more cool things that I haven't tried, let alone master.
@arkinson I forgot to link the Klein 9b. workflow. https://civitai.com/models/2437681/basic-workflow-flux2-klein-i2i-v22-or-4-in-1-image-editing
He has a website https://sololo.xyz/models with a lot of models as well as some interesting thoughts in his articles. He speaks about Z-image. It may help you. https://civitai.com/articles/25634/training-notes-my-take-on-klein-and-z-image
@boinobin730 Seems civitai has still problems. I can`t see any new stuff from you - latest is 2 days old. Ok, let us wait a little bit.
Yes, I believe we have completely different work processes. For myself, I am much too lazy to spent much time on a single image. In the very beginning (under automatic1111) I experimented with in/outpainting, trying to spice up my own photos. But I saw, all this is a lot of work and my own results were mostly not really satisfying. So I keept myself away from i2i workflows. But of course, I see what others can do and possably I will give it a try again.
Z-image vs Pony vs SDXL: SDXL is still horrible with nsfw. Pony is perfect for it, but I never got really photorealistic outputs and most character Loras do not look like the original character. Z-image is sometimes very picky about prompting. Some ideas working really well, with others you have no chance. And sometimes it is very static - one prompt, allway nearly the same output. But with wildcards and the capebillities of Lora Manager you can get large variety too. Thats suites my way very well: experimenting/optimizing settings -> running a bunch of automatic generations -> picking up the best results 🙂
Actually I am experimenting with this model. It delivers nearly photorealistic results and works pretty well in combination with Loras.
@arkinson I think that's the fun part where we all have different approaches to how we use and get value out of Diffusion models and the open source nature of sharing workflows, ideas, in the end we get constant development. There is no gatekeeper, that prevents users from creating what they want. If this thing wasn't open source, I wouldn't have spent any time on it as I know we are always regulated and constrained. Typical example is AI music. It's a crying shame what those companies did to AI music generation. It really should be like what we have in the image / video field. Hopefully open source proliferates in the music field as well.
Thanks for the Z-image model link. I haven't ever used it and it looks interesting. I think Z-image allows users to make some of the most accurate stable characters though. I'm not into the celebrities (I find it a bit spooky doing stuff with images of real people ) but some of the models people have made are extremely accurate. So for character fidelity Z-image might be worth a look.
There is so much stuff. My hard disk is getting full, I either need to delete some models or get a bigger hdd. How big is your hard DISK? I thought 4Tb was enough now I am thinking at least 8 TB...
@boinobin730 Has civitai still issues? I can`t open galleries or have to try several times...
I didn`t know about AI music. What happened there? And I do not remember - was it you or someone else who told me, he used Stabillity Matrix for AI music?? Seems I am getting older 🙄 Generating some audio I just tried with comfyui in the video context.
For image generation my favorite is still Flux.1 dev. It delivers the best quality with 12 gb vram in tolerable time + endless Loras, even for any artstyle imaginable. I tried Flux klein too, but quality is not comparable with Flux.1 dev.
Just another automated work path I sometimes use: random input image -> florance prompt generation -> additional promt with wildcards -> adding some noise -> ksampler/upscaling/etc.. That’s basically a i2i workflow. In my case optimized for "mass" production. Over several generations you can get "spectacular" results - sometimes 🙂
And yes, we all need much more vram, ram, disk space and much more 😂
@arkinson Yes, Civitai galleries still down.
I don't think it was me who spoke to you about AI music. I was using Udio and Suno the free version but what happened was Udio got taken over by a music company (UMG), they effectively gutted the product such that you couldn't download what you created. A lot of users lost a lot of work as they didn't back up , they also destroyed the model so that users couldn't even make good music from it anymore. So effectively anyone with an advanced subscription lost value from it. Suno is a similar story. They were/are being sued and also hobbled their model. Suno was not as good as Udio apparently. You can still download songs you have created but the model is not as good as it was before( basically forces you to waste money on music generations ). I don't spend money on these products.
That's an interesting idea about using florence and then wildcarding it. I might give it ago. You would certainly get some wild stuff happening, I suppose its constant surprise entertainment.
Flux 1 dev I agree is a very good model. You can do a lot with it. I am slowly playing with Flux a bit more, especially for environments, outdoor shots, indoor shots, backgrounds, backdrops.
@boinobin730 Music AI: Ahh, I see. Unfortunately, that`s exactly the way so many good projects went down over the decades - not only for AI stuff.
Florence workflow: That`s a little example. Generally I try to use a bunch of start images with a similar style/theme/scenario and additional prompting and Loras wich drives in the desired direction. Sometimes it is a little bit tricky to find the right amount of adding noise: too much and it might run out of the road, too little and you might stick around your start images.
@arkinson looks good. You can pretty much let your imagination go wild, then once Flux gives you a nice output, use this as a base and keep going with similar prompting, to create a unique style.
@boinobin730 LTX workflow v2.0 is out (added First/Last Frame to Video + Audio) 🙂
@arkinson Wow. great stuff. I will test now
@boinobin730 Yes, some testing of the First/Last Frame part would be great. To be honest, I did just a few tests - so technically it should work. But my quick outputs were mostly bull**it.
@arkinson I haven't tested as yet. I've just been busy with real life stuff. But I have now received a notice from Civitai , that all Australian users are to be banned from the platform, starting from tomorrow. I will have to buy a vpn now, i hope it works. If you notice users suddenly dissapearing this is the reason. here is the notice.
As of March 16, 2026, users located in Australia will no longer be able to access Civitai.
We know this is deeply disappointing news, and we're tremendously sorry to our Australian community. You've been an important and valued part of our platform since inception, and we're heartbroken to have to take such a drastic step. This decision wasn't taken lightly, and we want to be transparent about why it's necessary.
The Age-Restricted Material Codes Are a Serious Compliance Burden
The Australian government, through the eSafety Commissioner, has registered nine Age-Restricted Material Codes under the Online Safety Act. Six of these codes came into effect on 9 March 2026 and apply to social media services, app distribution platforms, and designated internet services — including platforms like ours that host user-generated and AI-generated content.
These codes are not limited to Australian companies. They explicitly apply to any platform accessible to Australian users, regardless of where it is based. The moment an Australian user can access our site, we're on the hook.
The codes require:
Age verification measures such as facial age estimation, digital identity wallets, or photo identification checks — simple self-declaration of age is no longer considered sufficient
Preventing access to age-restricted material including AI-generated imagery without verified age assurance
Ongoing compliance with evolving regulatory guidance from the eSafety Commissioner
We're a small team. We simply don't have the legal budget or manpower to implement the required age verification systems. Attempting to comply without expert legal counsel would be reckless, and if we get it wrong...
The Penalties Are Real
Non-compliance with the Age-Restricted Material Codes is no joke. The eSafety Commissioner has the power to levy civil penalties of up to AUD $49.5 million per breach.
These penalties are not theoretical. The eSafety Commissioner actively monitors compliance and has already taken enforcement action against platforms that fail to meet their obligations. Major platforms like Pornhub have responded by blocking Australian users entirely rather than face the compliance burden.
We cannot risk this level of liability.
Why Not Just Comply?
We wish we could. But the truth is, these codes were written with massive tech firms in mind, not modest teams like ours. Implementing compliant age verification systems — facial age estimation, digital identity wallets, or photo ID checks — requires significant technical infrastructure, ongoing legal review, and deep regulatory expertise. To do this responsibly would require hundreds of hours and tens of thousands of dollars. It's not just a few policy changes — it requires capabilities we simply don't have the resources to build and maintain.
What Happens Next
At the deadline, Australian users will begin seeing a block message when visiting Civitai from within Australia.
To Our Australian Community
We're truly sorry. This situation is deeply frustrating, and a geoblock was our last possible option. We care about this space, and we care about our users.
We hope the regulatory climate changes in the future, and if and when it does, we'll reassess our ability to reopen access to Australia.
Until then, thank you for everything. This is goodbye — for now.
FAQ
Q: Can't you just implement age verification and keep the site running in Australia?
A: The age verification requirements under the Age-Restricted Material Codes are technically complex and expensive to implement. They require systems like facial age estimation or digital identity wallets — not a simple checkbox. For a team of our size, the cost of building, maintaining, and auditing these systems is prohibitive.
Q: I'm in Australia and I'm furious. What can I do?
A: We understand — we're upset too. If you're an Australian citizen, we encourage you to contact your local Member of Parliament and make your concerns known about how the Age-Restricted Material Codes affect access to legitimate art, technology, and online communities.
@boinobin730 Oh my - life becomes better day by day 🙄 Yes, use a vpn, that will definitely work - and be happy you are not in Russia.
@arkinson Ha. tell me about it. Why can't the government allow me to make my waifus in peace, God Damn IT!!!!
@boinobin730 Unforunately, goverments can do completely other things too 🙁
@arkinson yes, life is insane at the moment in some places in the world.