Follow me on X/Twitter for exclusive content: https://x.com/Citron_Legacy 😁
I've made over 200 Art styles. Checkout my collection here: https://civarchive.com/collections/10766871
👨🦱🐂 Please support the artist by checking out their page 🔞: https://twitter.com/AfrobullArt
Description
,<lora:AfrobullNAI:1.0>,
FAQ
Comments (11)
What exactly is the difference between NoobAI and Illustrious?
I'm experimenting with it to learn more about it.
TLDR - NoobAI is supposed to be better at understanding Natural Language prompts instead of only Booru-Tags.
Full comments:
Its my understanding that NoobAI was "built on top of Illustrious's data" but supposedly has better Natural Language prompting.
According to the Checkpoint descriptions:
NoobAI boasts that its been good at Natural Language for at least 3 months.
Illustrious v2.0 says its good at Natural Language but it only got released 10 days ago. (Per the model descripts on CivitAI Illustrious v1 and v1.1 don't claim to be good at Natural Language)
Illustrious v2 is a private model that no one can download and thus no one can train Loras for. This is silly because checkpoints have very little value without Loras so I'm checking to see how the NoobAI creators are supporting their community.
NoobAI has "better technology" (maybe not everyone things Natural Language is better tech so I say it in quotes) sooner and is still open to the community where as Illustrious is very slow to get new functionality and is closed to the community.
All that said, I've made over 450 Loras on Illustrious v0.1 during the last 4 months and its v0.1 is great! Its just the later versions: v1.0, v1.1, and v2.0 that I have problems with. I like Illustrious and I adopted it earlier than most creators so don't think my critique of Illustrious is me hating on it, I just need to keep experimenting with anything/everything.
I'm still learning NoobAI I'm having a hard time training Loras for NoobAI because it just works differently. I don't know if that's because NoobAI is better or worse than Illustrious. Its just an observation that training on Illustrious is extremely easy (compared to SD1.5, SDXL, Pony, and Flux) but I don't understand NoobAI training.
(Disclaimer dates and timeframes mentioned here are only valid as of 2025/03/28 -- things may have changed if this comment is read at a later date.)
@CitronLegacy Hey, as Illustrious 2 is now public, may you show some light which one is best?
@krigeta Hey Krigeta!
Personally I think NoobAI is still best.
I made an Illustrious 2 checkpoint but I had a hard time getting it to work because it didn't follow the prompts very well. Maybe I just need to learn the trick to prompting it?
Now that someone has asked I'll release the checkpoint so you can see what I mean https://civitai.com/models/131986?modelVersionId=2034861
Pros: It makes really creative art.
Cons: Its hard to control.
(Disclaimer this comment was made on 2025/09/28 - Roughly 6 months after my previous comment so things did actually change. Its fun how the AI world is always evolving.)
@CitronLegacy great! thanks for the finetune and may you please share your experience training concept lora for like fight poses lora? like how to curate the dataset for like dragon ball z fight scenes type where two or more characters are present?
@krigeta For a fight scene hmmm... I don't have an answer but I can tell you what I've been thinking.
I made this dataset a long time ago https://imgur.com/a/FQH89BQ but I don't think it would work. I think I need to have different characters in these poses and more importantly I need a way to describe each image differently. "Hitting elbows" or "left punch from person on right side" vs "right punch from person on left side" are too wordy and I think it would confuse the AI. Its something I've been thinking about for a long time and also thinking about again recently but I don't have a good idea yet.
Maybe each separate pose/attack is its own lora? But that doesn't seem a nice as one lora with all the poses.
@CitronLegacy indeed a lora per pose would be too much for nothing, but the Idea is like there are a lot of NSFW loras, the concept is same but with different angle and the pose is not exact always but kinda close, in short:
A lora that has the knowledge of poses and complex fight poses so if we prompt in that way even using regional prompting and contronet it wont ruin the characters
@CitronLegacy check this lora https://civitai.com/models/1989787/groping
@krigeta Maybe its just me but making porn of two people seems easier than interesting combat. Groping is basically "grab the thing" the rest of the pose can be whatever. Also there is a ton of porn trained into base models (100k+ images of groping on Danbooru) where as there is barely any good art of two people fighting anywhere.
That said, all I have this this one dataset that I showed you. Either its good enough as is or I need to find more images but where can I find good images? Groping is easy to find but good choreography is not?
I love the Super Broly movie but even that didn't have many good keyframes for fighting poses. (it had lots of special effects which allowed me to make this https://civitai.com/models/1100059). Sadly good choreography keyframes are pretty rare in Shounen Anime lol
Good
Thanks! You have nice Loras too!
I reviewed and direct posted on all of them. 😁
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.








