- Jan 22, 2018
- 1,089
- 1,481
Tried it, following the instructions to even include the file type of the image (such as PNG), got no feedback that it even worked and there was still no image. Decided it wasn't worth the frustration.
- Thread starter TCMS
- Start date
-
- Tags
- anal sex bdsm big ass big tits character creation dystopian setting exhibitionism female protagonist footjob futa/trans gay group sex handjob interracial lactation lesbian male protagonist management masturbation multiple penetration oral sex pregnancy sandbox sci-fi sex toys simulator strategy text based titfuck trainer transformation vaginal sex virgin voyeurism
So I finally got to the point where I see the Stable Diffusion UI under the local address and port. (Im running windows versions of pytorch, pip, git and the CPU torch version). But:
A problem arises when I add the COMMAND_ARGS the FreeCities instructions tell me to add to the webui_user.bat. After I add these the UI doesnt appear anymore and the command window gets stuck. Are the COMMAND_ARGS in the FreeCities maybe tailored towards a build with CUDA/GPU usage? I want to run it via CPU.
A problem arises when I add the COMMAND_ARGS the FreeCities instructions tell me to add to the webui_user.bat. After I add these the UI doesnt appear anymore and the command window gets stuck. Are the COMMAND_ARGS in the FreeCities maybe tailored towards a build with CUDA/GPU usage? I want to run it via CPU.
I know that with the Automatic1111 WebUI the UI does not automatically open after the required arguments have been added to the .bat file and you're left with just the command line window, which I'm hoping is the case for you though I am unsure since we're not using the same exact software (Nvidia GPU). If you see a line like "Model loaded in ___" then it should have loaded everything it needs and is waiting for inputs.
Another thing to do would be to open Stable Diffusion via the .bat with the added arguments, wait for it to load then open Free Cities and check the Game Options, specifically the AI image generation settings. Look at the sampling method option and see whether or not it can communicate with Stable Diffusion.
If you see the above, then something went wrong somewhere and more troubleshooting required.
But if you see a list of samplers, then it's working as intended and you'll likely be good to go after you adjust any settings that are showing red ERROR messages in Free Cities.
Thank you! I wouldnt have thought of simply checking FreeC without seeing the Stable diffusion UI.
So I did and I had to change the default sample method to something else. The text turned from red to white and all good. Or so one should think. Because now the cmd window gives me an error message:
Does that ring a bell? (the test image does show nothing probably because there are still errors)
So I did and I had to change the default sample method to something else. The text turned from red to white and all good. Or so one should think. Because now the cmd window gives me an error message:
Does that ring a bell? (the test image does show nothing probably because there are still errors)
Just returned to the game today after a few months away. Prior experience, had no problem with it. Today, after updates, getting what you're reporting right here. I was planning on stepping away from it for a while to see if future updates will fix it, but then I saw you with the same problem and here's hoping somebody can point you/us in the right direction.
The reports I found with a brief search having the same error
You must be registered to see the links
and occured under different circumstances so I'm not sure what the fix is as I'm using a June build of A1111 and don't feel like updating at the moment to see if I also run into this problem, but
You must be registered to see the links
suggests adding "--no-half" or "--no-half-vae" to COMMANDLINE_ARGS. Just from that little console snippet I have no idea if the problem lies with the model you're using (unknown) or the VAE (if you're even using one), and that's on top of you going with CPU generation which may require other settings that I just wouldn't know about. From the error message, there's a number format mismatch going on so hopefully one of those two arguments can force a match, alternatively try "--precision full".Another option is to try another model (not SDXL based) and see if something similar happens.
- Jan 22, 2018
- 1,089
- 1,481
So looking over everything I've done over the course of like 6 or so years, it's clear FCDev tied most of the benefits of the game to being an alpha male. However, with everything we have now, the most content in the game is actually as a futa. Being a male is just the typical vanilla game, while being a woman actually limits you even more because cooch. A futa unlocks both the male and female content, including a futa only event where you can knock yourself up and make what the game considers a clone of your own person.
Now I assume this is only because futas/women got slightly more content because of the heavy breeding content, but it still means males and females still misses content.
Now I assume this is only because futas/women got slightly more content because of the heavy breeding content, but it still means males and females still misses content.
- Jan 18, 2021
- 329
- 564
While this is partially true, I think that your point of view is too narrow. In general, it is not possible to see everything in a single playthrough so you always miss some content. It is true that some content depends on what sexual organs are available, so having both gives you more opportunities. However, there is a lot of other content in Free Cities that depends on your choice of Future Societies, on what you have built or not in your arcology, on how you have reacted to some events early in the game, etc.
Saying that you should have both sexual organs to see more content is a bit like saying that you should pick Ethnic Supremacy or Subjugationism for your Future Societies because this is not mutually exclusive with other Future Societies so then you miss less content. While this is partially true, this only applies to a small part of the content. You will still have to replay the game several times with different options if you want to see more content.
Adding "--precision full" didnt help, "--no-half-vae" is already part of the suggested COMMAND_ARGS in the FreeC instructions... so I tried changing that into just "--no-half".
Aaand that did something! My command window now looks like this:
Whenever there is a new picture it fills up these bars (which takes minutes! can I speed this up?). The problem: I still dont see any pictures ingame:
Whoa, nearly 20 seconds per step? That ain't good. What's your GPU? What size pictures is FreeCities trying to make?
I've seen 20 seconds per step sometimes using the default size (512x768 iirc) and a GTX 1060 3GB, although normally I get less than 10. I don't know what causes it to sometimes take much longer. If SlightlyTitillated is in a similar position, restarting might help? Sometimes I think my GPU is just having a bad day.
Also, using xformers could help if you aren't already, and if you're using a graphics card like mine the low vram option would probbaly do you some good too.
If their setup is working as they first mentioned it, they're using CPU generation so it's going to take longer to complete a step of an 512x768 image than an 8 year old GPU rendering at 1080p will.
Try adjusting this setting:
My first guess is that it's just taking way too longer to render so Free Cities just gives up on receiving an image. Try adjusting that value to see if you'll get an image to show up. As for speeding up the generation, the best option is not to use CPU generation. Unfortunately this is an area where the parallel processing capabilities of GPUs really show their strengths; even if you ran stable diffusion on a server with a $3,000 CPU, any GPU with 4 GB VRAM made within the last decade is almost certainly going to outperform your very expensive paperweight.
If that's not an option and you're going to stick with CPU generation, then see if using xformers helps. As you're not using a GPU, I don't think any of the arguments like --medvram will help as you're not using VRAM. You could change the generation settings as well, by reducing either the image size or using a different sampler than takes less math to run. On a similar note, you're also going want to decrease the regeneration frequency by a lot by making that number in the settings a lot larger.
But at the end of the day, there's a really good reason why the people who are very into rendering images with stable diffusion either have or open their wallets for a GPU.
Thing is I have an AMD card (6700 XT) and Pytorch doesnt support Rocm under Windows so Im stuck with CPU based generation.
I suspect the Stable Diffusion model that Im using "realisticVisionV60B1_v51HyperVAE" . Maybe I need to set it up differently? The Lora model is the one mentioned in the FreeC instructions.
I suspect the Stable Diffusion model that Im using "realisticVisionV60B1_v51HyperVAE" . Maybe I need to set it up differently? The Lora model is the one mentioned in the FreeC instructions.
Ah okay I will look into that. But is there a way to utilize my AMD card for generation? I would love to.
And what would be a sampler that is light on the math?
Try
You must be registered to see the links
and
You must be registered to see the links
? I'm sure there's other versions/forks that run on AMD but I can't recommend any specific ones as I'm on an NVidia GPU. If you can get that working then tweaking the generation settings won't matter as much.Yes I actually do run this already. The folder is called stable-diffusion-webui-directml .
The problem is Pytorch. Due to SSL errors webui-user.bat fails to download it so I have to install it manually. Once on their website you quickly realize that there are only CUDA based torch versions under Windows - or CPU. I mean I could try combining a CUDA 11.8 based version of torch with this but this is destined to fail is it not?
The problem is Pytorch. Due to SSL errors webui-user.bat fails to download it so I have to install it manually. Once on their website you quickly realize that there are only CUDA based torch versions under Windows - or CPU. I mean I could try combining a CUDA 11.8 based version of torch with this but this is destined to fail is it not?
SlightlyTittilated - not sure it will help but on my old PC I was getting over a minute per image for the AI and it was timing out - and worse with FCHost than running in a browser.
I switched to SD Forge (search 'forge' in this thread, there are a good set of instructions, it needs some different settings to the guide in the game) -
But still pretty difficult to enjoyably use the AI generation on an older computer and for a few play throughs I just copied out the prompt info into the web browser version of Stable Diffusion and imported the images back in manually so that it wasn't constantly steaming away.
I switched to SD Forge (search 'forge' in this thread, there are a good set of instructions, it needs some different settings to the guide in the game) -
You must be registered to see the links
instead of the A111 version that the game recommends for Stable Diffusion and was about 3x faster, and mainly dealt with the timing out problem (it also seemed a bit more robust in terms of the installer except for needing an old version of python). As I understand it (and I don't really understand it) Forge has some better optimisation for low RAM.But still pretty difficult to enjoyably use the AI generation on an older computer and for a few play throughs I just copied out the prompt info into the web browser version of Stable Diffusion and imported the images back in manually so that it wasn't constantly steaming away.
- Feb 20, 2020
- 25
- 18
I don't think there's been image pack updates in a while. Most of the development work in the past few years have been WebGL (3D rendered in browser) or AI.
I cant bring Freec to show the generated pictures. I brought the time per iteration down to 4 seconds but it appears DreeC doesnt want to wait even that long.
Where does Stable Diffusion put the generated pictures? Can I have a look at them myself?