3 Years of Service
As per title, I'm looking for NSFW AI Image to Video Generator that I can install on my pc and run. All those on the web only allow a few free uses before I have to pay and it's too expensive.
Looking for NSFW AI Image to Video Generator - what are the best and are there guides around to install locally on my pc?
- Thread starter futos
- Start date
- Replies 18
- Views 4K
-
- Tags
- ai animation
1 Year of Service
If by installing in your PC you mean running the model itself and not calling a cloud service, then you probably need an extremely beefy GPU if you want to have at least an acceptable quality.
1 Year of Service
I have used this before
You must be registered to see links
But it seems this one is more comprehensive:
You must be registered to see links
3 Years of Service
My cheapo 12GB VRAM GPU should be fine. Probably. Only one way to find out
Thank you! Will go look them up
1 Year of Service
Amazing guide right here. This helped me get started easily. I use ComfyUI now
3 Years of Service
Thanks a lot!
3 Years of Service
What are the video generation times like? Minutes or hours?
3 Years of Service
Minute per second @ 480p, but it depends on frames per second and resolution too.
Most I have tried were 480p or 720p and you would upscale it after from that which takes more time.
With 12GB ram you'll want to use gguf versions.
Something like this I think
You must be registered to see links
(see resources bottom right.)
3 Years of Service
Good to know so I can temper my expectations. Thank you
U
If you are using ComfyUI this
You must be registered to see links
is a very helpful starting point using WAN 2.1. The workflow should also work with the GGUF versions. In general I had better results generating in 480p then upscaling than directly generating in 720p.
3 Years of Service
1 Year of Service
Framepack is a very simple Image 2 Video tool. It can be installed Standalone or with Pinokio. It is great for dancing, wiggling around and simple animations with good consistency to the original image.
Under ComfyUI, WAN 2.2 is king. But diving in the world of WAN requires quite some time.
Under ComfyUI, WAN 2.2 is king. But diving in the world of WAN requires quite some time.
3 Years of Service
hmmm I will likely try this out too, especially if it's simpler. Thanks!
2 Years of Service
Finally upgraded from the 1080Ti build I put together in 2018, and was hoping to do some dabbling in running models locally. Had hoped to start with this guide, but the link gives me a 404 error, unfortunately - and wasn't able to find via searches (closest I got was this thread). Is the guide still available, and can anyone point me to it if so?
3 Years of Service
Do you guys recommend a Mac M4 Max 128GB over a RTX 3080 10GB (~5-6GB available) even though it's not Nvidia, and what's preferred between StableDifussion WebUI vs ComfyUI?
Thanks,
Thanks,
2 Years of Service
Hell, I ran comfyUI for a bit trying to make it work out but none of the models I found worked well. Joined a couple discord dedicated to Ai creation but never got lucky enough to find anyone willing to do more than say "Get this and this and you'll be fine" and I kept running into problems that Google and YouTube didn't really help with. Still looking for anyone who has the patience to help a guy make AI images lol. I have a 4070Ti and 64gb of ddr5 so I should be able to do fairly well. Any help would be appreciated
2 Years of Service
I never did find a guide, but I was able to use the free version of claude.ai to get me most of the way. I was careful in my wording, but ultimately had no issue getting it to point me straight to all the files I needed, getting the canvas/workflows up and running, etc. I just asked it to take all the back and forth along the way (I spent several days, partly just due to the time it took vs. my free time available, and partly due to the limited tokens in the free version), and turn it into a guide in pdf format - attached. The particulars assume my own hardware configuration (I also had claude help me find a capable new PC with an RTX 5080, earlier in the same chat), so ymmv.
I have not verified that it included all the steps that were required to get it up and running. I may have used the free version of venice.ai for some steps in between, on the days when I ran into the token limit with claude and wanted to keep going. But from a quick scan, it at least seems to cover the relevant sections from what I remember (not that "what I remember" is ever going to be reliable...). If you get stuck, try using your own AI of choice to get to the next step - you can probably even feed it this guide, and ask it what's missing. I was very pleasantly surprised by how helpful claude was, after using Copilot at work and getting a useful response maybe 10% of the time (obviously on different topics than this). Tell the AI *exactly* what errors you run into or what step you're confused by, give it screenshots where possible - just be sure to crop around any prompts that you wouldn't want being seen by the wrong people. I'm not familiar enough to vouch for claude's privacy protections, so I just avoided giving it particularly sensitive information to begin with.
All that being said, I'm not yet producing videos that I'm fully happy with, but I'm still poking around, tweaking prompts, trying different settings. Not sure if I need different models/workflows/whatever (can't rule out that claude was relying on outdated information, there could be newer/better stuff out there to help), or if it's just down to my prompts/negative prompts and needing to let it run for more "steps" to get a better result. It's working, but the results are a few years behind the times, in terms of how janky the movements can be or how monstrous the woman's face may end up. I've gotten distracted with all the games I couldn't play on my old hardware, so haven't dug all that far into what can be done to improve it. I'm hopeful it's not just a matter of the paid-for AI generators using hardware that's way outside any normal person's budget, to get better results - on which note, I can recommend venice.ai's generation capabilities, and it at least talks the talk on being privacy focused (again, not an expert, so don't take my or their word for it, to be safe), though it does still rely on mainstream models (Grok, WAN, etc.) that may balk at certain prompts for legal reasons. It's also not remotely cost-effective for generating videos. The 3k I spent on a PC to run things myself would probably pay for itself in about 6 months, at the rate I burned through ~$100 worth of venice tokens. But for sheer convenience, it's pretty solid.
Given that this guide was put together by a mainstream, commercial AI with all the legal restrictions that entails, this guide *should* be safe for any section of the forums - though the videos it can generate with the right prompts, would not be.
I have not verified that it included all the steps that were required to get it up and running. I may have used the free version of venice.ai for some steps in between, on the days when I ran into the token limit with claude and wanted to keep going. But from a quick scan, it at least seems to cover the relevant sections from what I remember (not that "what I remember" is ever going to be reliable...). If you get stuck, try using your own AI of choice to get to the next step - you can probably even feed it this guide, and ask it what's missing. I was very pleasantly surprised by how helpful claude was, after using Copilot at work and getting a useful response maybe 10% of the time (obviously on different topics than this). Tell the AI *exactly* what errors you run into or what step you're confused by, give it screenshots where possible - just be sure to crop around any prompts that you wouldn't want being seen by the wrong people. I'm not familiar enough to vouch for claude's privacy protections, so I just avoided giving it particularly sensitive information to begin with.
All that being said, I'm not yet producing videos that I'm fully happy with, but I'm still poking around, tweaking prompts, trying different settings. Not sure if I need different models/workflows/whatever (can't rule out that claude was relying on outdated information, there could be newer/better stuff out there to help), or if it's just down to my prompts/negative prompts and needing to let it run for more "steps" to get a better result. It's working, but the results are a few years behind the times, in terms of how janky the movements can be or how monstrous the woman's face may end up. I've gotten distracted with all the games I couldn't play on my old hardware, so haven't dug all that far into what can be done to improve it. I'm hopeful it's not just a matter of the paid-for AI generators using hardware that's way outside any normal person's budget, to get better results - on which note, I can recommend venice.ai's generation capabilities, and it at least talks the talk on being privacy focused (again, not an expert, so don't take my or their word for it, to be safe), though it does still rely on mainstream models (Grok, WAN, etc.) that may balk at certain prompts for legal reasons. It's also not remotely cost-effective for generating videos. The 3k I spent on a PC to run things myself would probably pay for itself in about 6 months, at the rate I burned through ~$100 worth of venice tokens. But for sheer convenience, it's pretty solid.
Given that this guide was put together by a mainstream, commercial AI with all the legal restrictions that entails, this guide *should* be safe for any section of the forums - though the videos it can generate with the right prompts, would not be.
2 Years of Service
I managed to find a MUCH better workflow for video generation. In addition to being in line with what commercial AI video generators can produce (and in the case of a lot of the options I've come across, far superior), it has the added benefit of working equally well as either a pure text-to-video generator or a text-plus-image-to-video generator, and adds audio to it either way (I haven't spent as long trying to get the audio right, so that aspect is still hit or miss for me so far). That "few years behind the times" I mentioned in my previous post has been fast forwarded all the way to the present. Attached is a revised guide that should still walk you through from a fresh Windows 11 install to having a functioning video generation workflow, with the caveat that I have only skimmed it and not actually gone through step-by-step myself.
As before:
-this guide was put together by the commercially-available version of Claude, so the instructions should be safe to post in any normal NSFW environment, though it is capable of producing videos that wouldn't be
-worst case, if you run into a wall in trying to follow the steps, a free Claude account (or your AI of choice) should be able to bridge any gaps between steps in the guide
-this was tailored to my brand-new PC's specs, so ymmv.
My relevant specs:
RTX 5080 GPU with 16GB VRAM
64GB DDR5-6000 RAM
2TB NVMe - drive type is far less relevant to getting it running, but I'm sitting at about 125GB combined for my ComfyUI folder, which includes all the models/tensors/whatever as well as a few dozen generated videos. The videos themselves are a tiny fraction of it (far less than 1GB total so far), since they're only 16 seconds apiece. I also still have everything I downloaded for prior attempts, so it *should* be significantly smaller if you skip straight to this second guide. I'll clean out the old stuff when I start running out of room, since I figure there'll be 3 or 4 more new workflows to try by then... which I've obviously never said about anything in the past that I never got around to cleaning up...
Fair warning: several pieces of the workflow/canvas you'll end up with will be in Chinese (at least I think it's Chinese). You do not need to have any idea what they say for this to work - I certainly didn't.
Apologies for the double-post, but I wanted to preserve the previous for reference rather than replace it. Anyone curious about just how far they've come in the models/training itself, leaving aside the hardware advancements, should take a stab at running each of them on the same machine and comparing the results. I've been absolutely amazed at the difference. It's like comparing claymation to modern, big-budget CGI. I'd offer to post some samples of each myself, but I'm still not familiar enough with everything to be sure I can remove all the metadata and such that I wouldn't want floating around out there (as it stands now, I can open a blank canvas, drag and drop a prior video into it, and it will not only discern the fps and such, but the exact prompt used to generate it, so uploading THAT anywhere is right out). If anyone has any tips on that front - and can point me to an appropriate thread in which to post them - I can look into doing so if anyone's interested. Eventually, might also be able to offer turning other people's prompts into videos (I'm not particularly creative myself, so could be a win-win if someone has the prompt-writing skills but not the hardware), but no promises.
As before:
-this guide was put together by the commercially-available version of Claude, so the instructions should be safe to post in any normal NSFW environment, though it is capable of producing videos that wouldn't be
-worst case, if you run into a wall in trying to follow the steps, a free Claude account (or your AI of choice) should be able to bridge any gaps between steps in the guide
-this was tailored to my brand-new PC's specs, so ymmv.
My relevant specs:
RTX 5080 GPU with 16GB VRAM
64GB DDR5-6000 RAM
2TB NVMe - drive type is far less relevant to getting it running, but I'm sitting at about 125GB combined for my ComfyUI folder, which includes all the models/tensors/whatever as well as a few dozen generated videos. The videos themselves are a tiny fraction of it (far less than 1GB total so far), since they're only 16 seconds apiece. I also still have everything I downloaded for prior attempts, so it *should* be significantly smaller if you skip straight to this second guide. I'll clean out the old stuff when I start running out of room, since I figure there'll be 3 or 4 more new workflows to try by then... which I've obviously never said about anything in the past that I never got around to cleaning up...
Fair warning: several pieces of the workflow/canvas you'll end up with will be in Chinese (at least I think it's Chinese). You do not need to have any idea what they say for this to work - I certainly didn't.
Apologies for the double-post, but I wanted to preserve the previous for reference rather than replace it. Anyone curious about just how far they've come in the models/training itself, leaving aside the hardware advancements, should take a stab at running each of them on the same machine and comparing the results. I've been absolutely amazed at the difference. It's like comparing claymation to modern, big-budget CGI. I'd offer to post some samples of each myself, but I'm still not familiar enough with everything to be sure I can remove all the metadata and such that I wouldn't want floating around out there (as it stands now, I can open a blank canvas, drag and drop a prior video into it, and it will not only discern the fps and such, but the exact prompt used to generate it, so uploading THAT anywhere is right out). If anyone has any tips on that front - and can point me to an appropriate thread in which to post them - I can look into doing so if anyone's interested. Eventually, might also be able to offer turning other people's prompts into videos (I'm not particularly creative myself, so could be a win-win if someone has the prompt-writing skills but not the hardware), but no promises.
Similar threads
Legal Notice
This site provides links to other sites/services, and does not store any files