AI Best Chat AI websites / apps

SillyTavern as some others have mentioned in this thread. Open source front end then pair it up with your LLM of choice. Can go as cheap or as expensive (or free) as you want with quality or response time varying to how much you want to spend. Can pull and import character cards from various sites. Though with it being open source, it takes trouble shooting and understanding processes to get it up and running, but it's cheaper than and of those sites and you can scale and tailor it to however far you want to take it. It's also not the only front end you can use, but one of the most popular.
 
I don't have hardware to run anything locally, I use these services on my tablet so it'd have to be as lightweight as a website like kindroid. I also want things as hassle-free as possible since this is something I do for fun every now and then rather than making it a job or even a hobby
 
The chats are very good, also customizability is a nice feature.
They have pretty good voice lines.
But image generation does not care for the title picture of the character at all. The generated images look completely different, which is sadly a big negative
 
Do SillyTavern support NSFW image generations as well? :peepoSad:

I'm also using Kindroid (after Darlink went to shit) but it lacks something that I can't really put my finger on. Kindroid requires a bit of tweaking to get things right, and even then the experience can be a bit frustrating so I'm also looking for an alternative. I don't mind paying as long as it's worth it.
 
It can, but it's just a front end service run on your computer. It can also be run on android as I've seen people mention running it on their phones for corinthianultra. The image generation has to be tied to a back end service or API. I've never messed around with image generation on it due to it being pretty expensive compared to chat completion and you'll have to adjust settings and prompts to get closer to your ideal image. I'm sure there's guides for it though. Setting up SillyTavern is decently user effort/research intensive, but if you're somewhat technically inclined, can follow guides, and google/ask AI for assistance, you'll be fine.

I've tried website chat services only on the free trials, but they all charge a subscription, sometimes subscriptions on top of the entry one, and pay as you go for additional services and after seeing the limitations of the chats/image/videos (not truly new image generated, just pulled from an image library of already rendered images and close enough to your prompt for example) I said nah, I'm good. I've spent less than $20 USD over two years of usage. SillyTavern itself is free, just depends on how expensive the API service you want to use is per token. Look into the SillyTavern reddit or youtube guides to see if it'd be up your alley.

I can't link it here, but google perchance text to image generation. Free and allows NSFW for something you can mess around with. Can't be used as an API for SillyTavern though.
 
I mean you can use koboldcpp as the model launcher instead of ollama, that has built in image gen, audio gen. You can also run A1111 or comfy ui, link your locally running servers for them to silly tavern. It works but its not like the online ones with access to datacenters, its limited by your GPU vram speeds. BUT ITS FREE and its locally running so its more private/secure. Plus linking comfy ui you can add home made Loras to the workflow or home made checkpoints, way more effort but way more customisable and its free (if you have the juice to run it all).
 
I started out with erogen.ai, and it's still decent, but it feels like they've really restricted some content and have paywalled a lot of their better features over time.
 
Local models are ideal, but they aren't there yet for the quality to price point ratio yet especially with GPU pricing being bonkers and 5090s now going for above triple the MSRP. Not to mention people still haven't found out a workaround for the melting 5090 power connectors yet. I want to run local myself, but waiting until the models improve, GPU prices drop, or 6000 series with more VRAM across the lineup while being found at a reasonable price (lol) which will all be quite some time waiting. For pricing and chat quality from what I can get from openrouter or nanogpt, it's much more affordable. However, if PC gaming is also a hobby, the fancy GPU can crossover to both. But yeah, if privacy is of the upmost concern, local is the way to go.
 
Do you guys recommend A2E or Venice studio for IMG to Video creations?
 
Cant imagine trying to build a machine for it now, I was lucky and did a new build in 2022-2023, 4070 ti super 16gb for about £700, m.2 drives for £30-50. I think in total I spent £1,220 if I tried to do that same build now, looking at my build on pc part picker it says £3350.

I was still running some smaller LLMs and even image gen on my old gtx 1080 that was 8gb. I would rather put that money to the side and save up for better components than using it to pay for AI, that’s just me though. Most of the stuff I wanted to play around with and do ive manged to do so ive never really thought about paying for it. Ive done inpainting to photos, trained a model on a person and made a image model that just generates images of that person, cloned voice, done text to speech, moved onto image to video then text to video, then added cloned voice to the video gen, then made a chat model with a tts using the cloned voice then tied image gen to it and it could do image gen for me too.

Also im currently making characters and world lore sheets in silly tavern, that does so much to help with guiding things, you dont need a super complex model I dont think just some good character building.
 
For anyone who wants to generate erotica, chat bots, joi, etc... spicywriter has Muse Spark 1.3 in its free plan right now.

The only drawback is that it might use your submissions for meta's ai training.

As an advanced model, it's pretty capable, but requires more prompt structure than Claude's Sonnet/Opus. It also tends to take instructions more literally than other models.
 
Yeah, can still definitely get mileage off of 8GB VRAM GPUs. From my understanding, to rival GLM 5.3 or Kimi K3 (what I consider high quality, but still budgetish) that I'd need to be running 128 GB of VRAM total for an equivalent local model from ollama. And even then there would be a noticeable difference in nuance, long term memory, and tracking multiple characters. My current card can run some models, but I don't think I'd be able to run anything I'd enjoy after being spoiled trying Opus, Sonnet, GLM, and Kimi. Was thinking of upgrading to the 5090 to mess around with local models and also for gaming reasons, but not at these prices. I'm hoping local models improve and GPU prices drop, but I know it's going to take a bit. Luckily I have some pretty budget priced API routing options to hold me over.
 
Curiosity got the better of me, so I tried the Ultra subscription on FictionLab ($25 fucking dollars a month. I've been using Plus for $9 before for almost a year now).

For those who can afford it and are primarily interested in text generation for roleplay, I can confidently recommend Arcanum on GLM 5.2 - it absolutely blows my fantasies out of the water. It occasionally refuses to generate incest stuff, though, so you have to regenerate a few times, which is a bit annoying. But overall it is just fucking insane how creative and fun it is.

In that regard, Paragon on DeepSeek v4 pro is much more permissive when it comes to subject matter and allows pretty much anything, including very dark themes. So this one is pretty good if you get tired of Arcanum for some reason or you want some dark stuff. The downside is that it’s not as creative, and sometimes takes 30–40 seconds to generate a response. Arcanum, by comparison, takes around 2–3 seconds.

But i don’t think I’ll keep the Ultra sub for another month though, it’s just too expensive for me to pay regularly just to goon despite how amazing the Arcanum is. And here’s the funny part: I don’t think I’ll want to go back to the Plus sub either now, since that I’ve experienced how good Arcanum model can be in comparison with the lower Plus models.
 
I'm not sure this is the best place to ask this so redirect me if there is a better thread.
Does anyone have any good prompts for minimax h3 max turbo on Venice. Either for facials or just some good prompts in general.
 
Is there a good chatbot where I can upload an image reference, do nsfw chat, and generate photos within the same interface? TBIO is ok, but it requires two different chats open at the same time and you can't switch back to a text model after uploading an image.
 
Locally or online? online hides everything on the back end linking services and servers to each other, locally you have to run it all yourself usually.

So image gen is its whole thing, you can get chatbots that can "see images" run comfy ui or A1111 in the background and link it to ollama, koboldcpp or just through open webui. I have done this before, I ran the A1111 or comfy ui server then just added the address to open webui, I already had ollama hosting the chatbot models so technically I had 4 servers running with my set up, ollama to host models, A1111 as image gen, TTS server for speach and open webui as a UI to link all of them together in 1 place. I set up a BAT file so when I run the BAT file it opened up everything and ran everything including firefox and the open webui address so to run all I did was run a bat file, select a model and start talking took about 1 min.

The next step which is what im looking into now is getting a locally running agent on your PC which should basically do everything I just said but hopefully easier to add things and tweak. (so far look into KoboldCPP it can run its own imagegen and editing without having to run A1111 or comfyui, it has built in agent features where you can chat, drop iamges and ask it to edit / gen for you)
 
You must log in or register to reply here.
Back
Top