Project by AiArtLab
Colorfulxl is out! But who cares then we have so great 1.5 checkpoint?
https://civarchive.com/models/185258/colorfulxl
Thank you so much for the feedback and examples of your work! It's very motivating.
About v 8.0:
This is a unique synthesis of cinematic, photos, anime and art.
I think this is the final version. Most likely I'll do ColorfulXL now. At the moment I don't see anything else that can be improved. Thanks to everyone who has been with me all this time (more then year)! Try different resolutions and check 1-step generation with DPM adaptive (but other samplers, with 10 steps may be faster)
My settings
Portrait: 768*1088
Landscape: 576*1280
Samplers: DPM++2M SDE Karras - 20 steps, best (work with 10)
Also work: Euler a, DPM2, DPM++2s a, DPM++ SDE, DPM++ 2s a Karras
DPM adaptive - 1 step
VAE - included! (civitai bug requires indication of a VAE that is not on this site: vae-ft-mse-840000-ema-pruned.safetensors)
About v 7.3:
A very controversial and complex version.
Very sensitive to negative prompt, sampler, resolution and number of steps. Probably bad at art.
In fact, an attempt was made to remake SD1.5 globally. Which often leads to incorrect anatomy. Still, I think it's an interesting version and I'm curious what you can create with it.
Before
After
Most huge train details
https://wandb.ai/recoilme/finetuning/runs/nkfsrchd
About v 6.2:
I don't have luck with hires training (1280 - just throw in trash millions steps and $70)
So, i returned to v3.7 as a base and trained on 960*576, 576*960 and 768*768. Some cinematic added. But its not photorealistic model (i like art), mostly semi-realistic
Please respect my time and resources. If you use my model to make money - contact me for fair use
About v 5.0:
It was hard. But cheap) I spent less then $30 and ~1000 civitai credits on train.
At 1st of all i preserve ability to generate in 9-12 steps on 768/832 resolution. By my opinion quality is better then Turbo/LCM. But it works only on some samplers
So, at 1st of all i continue playing with negative training. At this time i train deformed, worst_quality, oversaturated and so on loras
And merge it with different weights, like -0.3
You will be surprised how many trash in SD 1.5!
At same time i train realistic checkpoint. Realistic is tricky
For example "woman, covered by snow", looks like
But thanks to Supermerger team and my brilliant cosine similarity merge algo, we may fix it to something like this
best ratio, a photo of attractive stunning Ukrainian woman, messy bun, covered by snow, skin pores, behind glacial mountains, snow, high detailed skin, film grain
And last
Anatomy not ideal at hi res, images was cherrypicked from batch 4 on 832*1024. If you want more stable anatomy try 768 on 960 and low
Have fun!
About v 4.6:
I create 'ugly' dataset
And train on it, and merge with 'add difference'
After that anatomy little better on 768*968, 968*768 and so on, before
after
But it was oversaturated:
After a longer study, I discovered that this method gives a very large shift and, unfortunately, while correcting in one place, it breaks in another. This is no better than other methods available to us or direct training.
So, i train again, but on "good images", and merge with cosine similarity
Strange, but i found what it may produce hi quality images at low steps in some cases
About v 3.7:
What's on the train. Compared to SDXL it is 10 times faster, I overfitted out of habit, saved the low lr 4e-7 and the adafactor (although I didn’t cut much). After 4 epochs the result is absolutely normal. Over the course of a night and an evening, she learned the realistic model, so much so that she forgot everything she had been taught all these years. I trimmed it with merging. Charged it for another 4 epochs - again it floated into realism. Trimmed it again)
Something like that little by little. Taught in a multi aspect ratio with an emphasis on portrait and landscape generation. Few people like to generate squares. Dolls/cartoon - etched to a state by a certain semi-realist. Hands/faces in the distance/composition - better, but not perfect
The optimal resolution is 768 by 960. A negative prompt has a very strong effect on the quality, see the examples - I did well with this negative, but with the basic one it was kind of crap. Idnk why is this so
Of course, it wasn’t possible to make SDXL from 1.5 overnight, but I’m happy as an elephant with the result)
Contacts
Models
Description
Inpaint model
FAQ
Comments (17)
Using the 3.1 version for a lot of images, but CivitAI fails to detect it. I've installed their extension and it detects all other models and cross-posts there. But for Colorful v3.1 it doesn't detect it....
One example here
Colorful is my favourite model, one of the most the capable I've come across when it comes to accurately depicting the prompt. Not just good but empowering work. Bless you
Somehow I can't generate any images with the 3.1 version; just gives me nothing, like a scewed image! version 3 works okay, though! any particular thing I need to do in automatic 1111?
This model seems to have trained on an unusual amount of bare breasts. Using no prompt and this model gives roughly 50% containing them
Sounds like a plus to me! LOL, You likely NEG them out of need be of course.
it contains animatrix, animatrix merged with rev animated in both directions, rev animated - hentai/anime model
@recoilme soo basically yes, you do get boobs.
Any info/description on what changed in 3.1? I know 2.5 is supposed to be better at photo-realism, but what about this vs. 3.0?
i trained lora on aestetic images (modern art) with pyramid noise + noise offset and merge it with 0.3
What dimensions is this trained with?
512x512
Link to colorfulxl seems to be 404 ?
Does the Clip Skip matter much?
Hi
I wonders the same thing and found this answer: (if you already not have figured it out by now)
https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5674
from what i remember someone did mentioned that turning clip-skip:-2 will make generation to prioritize your prompts then follow by +lora +base_model, and clip-skip:-1 are to strictly find a best leverage that stabilize between lora+prompts+negatives+Ti+model_base. I could be wrong if it deeper into sampler-processing, however most makers might mention which clip-skip are best for solid usage-contents. cheers
Okay, dumb noob question, why is the pickletensor bigger than the safetensor? Is the pickle uncut? I'm asking because some other models have pruned vs unpruned versions with the non-pruned ones being twice the size or more.
safetensors packed in fp16, pickle - fp32 (2x size). No difference in outputs, pickle - just for compatibility with some old soft