- This topic has 30 replies, 10 voices, and was last updated 1 week, 3 days ago by
.
-
Posts
-
In a response to a comment i posted on dreamlights youtube he acknowledged this would be a good upgrade, particularly loras. So fingers crossed that this product continually develops
I have Aire and it's fun but honestly as others have said if your computer can run it without stress than it's better to learn Comfy UI and have the whole universe of tools and flexibility available. There are so many more models out there than Qwen and being able to inpaint/mask/controlnet and use Loras and different checkpoints is a lot wider.
If you want to start from scratch with a prompt, totally in agreement with you here. Although I still suck at it. What got me interested was the idea of AIRE was using some of my scenes and "beautifying" them.
In essence it's true that using Comfy, checkpoints and lora's is way richer, there's no denying it. It's a diffenret tool for a different goal I guess?Yes! Agreed, and I still do use it, it's much more convenient and easy to run on a render just to see what you get - and Qwen using natural language is nice as opposed to some of the random prompts needed for lots of the fun SD models ðŸ˜
I use AI to help me prompt, and when it comes out weird it can be great to diagnose!
There's an interesting thread over at Sick-n-Twisted'Otica ( https://www.renderotica.com/community/forums.aspx?g=posts&t=285815#post565753 ) run by someone (ADP001) who seems to know a lot about AI used for images. He's linking to and promoting an open source tool that can be run locally, and will make naughty pictures.
as others responded you can indeed run Qwen models locally! Depending on your Vram power, there's also the Flux/Krea line which is incredibly capable in various ways. But Stable Diffusion models can be best for keeping original render detail integrity and inpainting/controlnet for depth and more complicated scenes. Unfortunately they've got the annoying prompting input most of the time.
Check out ComfyUI's list of existing workflows: https://comfy.org/workflows/
I run free models locally for most of the time but there are frontier models that are API only, and it's really not too expensive to do some fun work on existing renders via their cloud tool and some of those models. Like all API systems it's all about keeping from getting too iterative and suddenly waking up after a 3 hour generation session and finding you spent $15+ on one image (experience!)In terms of the ability to create mature like @Ethiopia mentioned, the Stable Diffusion and beyond model community ecosystem has an impressive amount of model and Lora options to create 'mature' (best word to use, 'erotic' gets you scolded) on CivitAI: https://civitai.com/ that you can tweak and find by style.
Wait did some research and looks like the best place, like with all things models, would be HuggingFace/uncensored weights and Loras live there http://www.huggingface.com as other repositories have censored more (Edit to my edit: look for "Civit Red", they apparently put it in a new safe sexy model space)
There's also got to be some sort of (or many) thriving community/repository/or discords out there for adult-tuned model discussion and reviews, haven't looked into it as I'm doing more illustrative stuffs but the internet is for porn/rule 34!
There's so many tiers of effort and generative tools for convenience and cloud versus local and tuned models for various focii and fetishes etc it's kinda dizzying; I think it comes down to how much do you want to invest in/what's your GPU power for running comfy, how much do you want to do uncensored (which will need to be local), and where your needs and commitment and time and equipment falls on the flexibility/control versus 'ease of use and image generation' scale!
civitai.red is their naughty version.
A point that's been thrown around is to compare the newcomer (AI) to what we did with Poser then Studio. There's a huge learning curve and we're still going uphill. Instead of buying software and models, we're investing in hardware to run the AI locally. As much as I love 'playing with dolls' in Studio, even a staunch traditionalist like me can see what's coming. Among the flood (tidal wave) of AI slop that's all the same, there's a slowly growing trickle of truly excellent and original imagery being produced.
I know Huggingface of old, for llm downloads. But that's been mostly for SillyTavern RP. It's been a nasty surprise to find out that Nvidia has bought them a few days ago.
Not out of the box. I dont know enough about python as to whether you can hack one of the files. Dreamlight said he was hopong to include use of other models in later updates
It's just a daz bridge to throw your image into qwen image edit. It's running locally so that means you have daz sucking up vram and the model.
Yeah, both Daz and Comfy open together is a VRAM nommer.
it was actually Aire - which is really cool, not a knock, it's much more accessible than Comfy - making me download Comfy that led me to move to actually just straight using it to create AI, learning the workflows and what different models can do (found a really nifty double pass template that uses Krea 2 raw and distillations + interesting checkpoints to turn out some lovely work, add controlnet to lock in Daz staging and poses and the sky's the limit.
Hypothetically, I'm no currently learning inpainting and masking, it's powerful but finicky.
Basically it's a great gateway to expand from for a Daz-AI polish workflow
- You must be logged in to reply to this topic.
