Are we allowed to like AI art?

[Reiwa]

Originally Posted by Ekco [You must be logged in to view images. Log in or Register.]

one benefit of the coming Butlerian Jihad is everyone gets a complimentary grill

[You must be logged in to view images. Log in or Register.]

Must have sausage. Little tightly coiled piles like a dog shit, but made of edible meat.

Do want.

[Ekco]

we haven't done art in this thread entitled "Are we allowed to like AI art?" in like 20 pages so..

"can you do a Patrick Nagel inspired High Elf Female"
[You must be logged in to view images. Log in or Register.]

[OriginalContentGuy]

Originally Posted by Ekco [You must be logged in to view images. Log in or Register.]

we haven't done art in this thread entitled "Are we allowed to like AI art?" in like 20 pages so..

"can you do a Patrick Nagel inspired High Elf Female"
[You must be logged in to view images. Log in or Register.]

Ekco, I watched your video about your embedded machine learning personal assistant. And while it's definitely attention grabbing, I really want to know what your actual workflow looks like with this tool. What are your early insights and/or first impressions about your experience? Also did you code it yourself?

[Ekco]

Originally Posted by OriginalContentGuy [You must be logged in to view images. Log in or Register.]

Ekco, I watched your video about your embedded machine learning personal assistant. And while it's definitely attention grabbing, I really want to know what your actual workflow looks like with this tool. What are your early insights and/or first impressions about your experience? Also did you code it yourself?

https://github.com/Ekco-S64QTN6/ollama_rag_agent

it's 100% vibe coded and i have no idea what i'm doing, this is my first time even looking at python and i'm using Gemini mainly to teach me concepts and explain things as we go, and for the code i let Gemini, DeepSeek and Chatgpt argue/code review each other pipe that back into Gemini and have it explain to me like "structurally where should this go in the main loop" i ask what's best practices a lot.

it's a lot of "let's add tts to it?" "llms don't have persistent memory how do we go about adding that?" "walk me through installing PostgreSQL" "hey this single file is getting really long can we break this out into multiple files" and me going Ooooh so that's what refactoring code means..etc..etc

it's inefficient anyone who actually knows this stuff could have done it in 1/100th the time and it's probably the worst way ever to actually try and learn something


my early insights/first impressions are it's just a glorified autocorrect (LLMs) that you can attach some other tools to like in my case letting it run commands and scripts, talk, display system info like neofetch does and in theory actually index and parse the documents i have in the data folder locally so i can ask it questions

[You must be logged in to view images. Log in or Register.]

[Ekco]

Originally Posted by Ekco [You must be logged in to view images. Log in or Register.]

and in theory actually index and parse the documents i have in the data folder locally so i can ask it questions

this is the part i'm currently stuck on and have taken a break, they just released a new Nvida driver and python version and yada yada so i had to go through and update all that and make sure it still works, then the actual QA/bug fixing of checking if the model is getting the information it's spitting back from it's pre-trained model, the actual pdfs in it's folder or just straight up hallucinating

which has been a major headache so far

i have an entire document of testing i run it through

### Kaia Functionality Testing Scenarios

This document outlines a series of tests to verify Kaia's core capabilities, including Retrieval-Augmented Generation (RAG), command execution, memory management, and persona adherence.
1. RAG (Retrieval-Augmented Generation) Tests

These tests verify Kaia's ability to retrieve and synthesize information from its knowledge base (./data and ./personal_context directories). Ensure the relevant files are present in these directories and the index has been rebuilt.
From 'Artificial Intelligence: A Modern Approach by Stuart Russell and Peter Norvig.pdf' (or its .txt equivalent):

Query: What are the four approaches to AI as described by Russell and Norvig?

Expected: Kaia lists the four approaches (acting humanly, thinking humanly, acting rationally, thinking rationally) and provides brief explanations for each, without self-referential statements about being an AI.

the persona adherence section is fun,
[You must be logged in to view images. Log in or Register.]

[Ekco]

[You must be logged in to view images. Log in or Register.]
[You must be logged in to view images. Log in or Register.]
[You must be logged in to view images. Log in or Register.]

[Ekco]

https://github.com/comfyanonymous/Co...readme-ov-file
[You must be logged in to view images. Log in or Register.]

gonna try to get this shit working now, Google and Chatgpt aren't cutting it

Detailed Analysis of Each Model for Your Setup:
1. Stable Diffusion (Your Primary Local Option)

Can it be installed on Arch Linux with a 3060? Yes, absolutely. This is the perfect use case for your hardware. The 12GB of VRAM is excellent for running modern Stable Diffusion models.

Can it achieve the style? Yes, but it requires the right model and LoRA (Low-Rank Adaptation). The base Stable Diffusion model won't naturally mimic Patrick Nagel perfectly. You will need to:

Use a specialized checkpoint model fine-tuned on artistic styles. Look for models like "Anything v5", "DreamShaper", or specifically "Patrick Nagel style" models on platforms like CivitAI.

Use a Nagel-specific LoRA. A LoRA is a small file that teaches the model a specific style or character without needing a full multi-gigabyte model. Searching CivitAI for "Patrick Nagel LoRA" would be the fastest path to getting the exact style.

Your prompt engineering skills will be critical. The improved prompt I provided earlier is exactly what you would use in a Stable Diffusion UI.

How to Run it: The most user-friendly way is to install Automatic1111's Web UI or its more modern fork, ComfyUI. Both have detailed installation guides for Arch Linux, usually via Python venv and pip.

[Ekco]

[You must be logged in to view images. Log in or Register.]
[You must be logged in to view images. Log in or Register.]

gpu go brrrr

[Ekco]

[You must be logged in to view images. Log in or Register.]

getting closer

[Ekco]

same problems i have with all the models so far, if you just want a random image and aren't picky it's fine. if you're trying to get something very specific it turns into a nightmare trying to get the prompt just right

[You must be logged in to view images. Log in or Register.]
[You must be logged in to view images. Log in or Register.]
[You must be logged in to view images. Log in or Register.]