"DAMN!" (c) - my first thought when the finetuning ended and I saw the final results.
BEFORE USING THE MODEL, READ THE DESCRIPTION, SPECIFICALLY THE PROMPTING AND PARAMETERS SECTIONS.
If you want to support me or help me cover the costs:
https://boosty.to/coyotte/donate or Bitcoin 1CoMZvrXtRXbz7BCzdvxifSjHqhaf1yUZs.
Every donation will go towards developing chekpoints, workflows and loras (this stuff is costly). I have big plans, but little means to fulfill them 💀
YOU ARE NOT ALLOWED TO USE THIS MODEL AS A PART OF YOUR MERGE, UNLESS YOU CREDIT ME IN THE DESCRIPTION.
Important prompt structure for Illustrious-based version.
Rule #1: More tags ≠ Better output. It's actually other way around. Do you really need 89 tags?
Rule #2: Use danbooru tags.
Rule #3: If your prompt causes physical nausea to read - fix it.
Rule #4: DON'T use Pony loras on Illustrious.
Positive prompt:
✅ Try: masterpiece, best quality, realistic, [your subject, actions, clothing, etc].
❌ Avoid: 3 paragraphs of random junk you saw in Discord
Negative prompt:
✅ Try: bad quality, worst quality, low quality, [something very specific to the prompt you're working on right now]
❌ Avoid: dozens of redundant anatomy tags like "missing limbs, bad eyes, floating limbs, etc", negative embeddings made for specific checkpoints
Pony's quality tags like "score_9, score_8_up, score_7_up" AREN'T doing anything on it.
If you choose to ignore the guide and are getting trash results - please also ignore the "rate" button.
My simple yet organized workflow I've used for creating the preview samples: https://files.catbox.moe/147f5i.json
RECOMENDED PARAMETERS FOR BEST RESULTS:
◦ Sampler: DPM++ 2M SDE / DPM++ 3M SDE / DPM++ SDE
◦ 30-40 Steps.
◦ CFG: 3-6 (low for more realism, high for more stylization)
◦ The model was on trained mostly on 1024x1024, 832x1216, 1216x832, 896x1152, 1152x896 resolutions, so it could help the quality if you used these (although that's not necessary - it behaves nicely on other aspect ratios as well).
◦ Highres fix: Upscale by 1.35-1.5, denoising strength ~0.4, upscaler - 4x NMKD Superscale or Siax.
For the distant shots of people you probably wanna use highres fix/upscaling and/or adetailer.
PARAMETERS FOR WORST POSSIBLE RESULTS:
◦ Sampler: Euler A
◦ 15 Steps.
◦ CFG: 11
◦ 512x768 resolution or whatever.
◦ No highres.fix/upscaling, no FaceDetailer/ADetailer
Description
FAQ
Comments (25)
the legend returns
Lustify was taking all my free time and money, but now that it's finished, I just HAD to return to DAMN
Im from Pony, so its irrelevant, what i say.
Best adherence, what i saw, but surprisingly, it needs to prompt heavily for realism(realistic, photo\(medium\) and similar tags).
That's interesting, because as you can see in my preview pictures, I don't explicitly prompt for realism, yet it's realistic. My bet is that it has to do something with the rest of your prompt or even negatives.
Can you share your whole prompt?
My Pony prompt is:
photo \(medium\), source_photo, real life, photorealistic, highly detailed, 64K
Negative:(uncanny valley face, uncanny valley eyes:1.2), (ugly), airbrushed, (low quality, worst quality:1.2), smooth skin, porcelain skin, unnatural, muscular, 3D, 2D, painting, CGI, render, illustration
You can say, these are bullshit, but works.
Now, when i removed the realistic tags, it became semireal(2msde, 5cfg), with character lora.
So yes with character lora(which is common usage).
@robertreeve359 ahhh, was the character lora trained on anime or real pictures?
Also, we are talking about the REBORN version, the Illustrious one, right?
I've tried countless pony models and they all have that 'pony face' with them. This one os the only one that has seemed to fix that issue, so well done on that. Unfortunately it seems to struggle with cunnilingus. Though this is definitely the best model when it comes to realistic, like not stiff 'cover photo' or 'perfect images' but feels like I'm actually scrolling through instagram photos with all the different angles and pov, lol. It's great
Keep refining, dont drop, like the pony, what remained unfinished, half-torso.
I will try to. Pony wasn't adapting to what I i was going for back then, that's why I've abandoned it. But now that I have it out the way with Lustify, I'll give it another try.
Although I still prefer Illustrious/NoobAI to Pony
@coyotte Yess. A new pony would be awesome. I dont find big difference between realistic! Illustrious models and Pony. Pony can be still more realistic. Avid realistic users dont touch anime specific concepts anyway. I acclimated to the limits of realistic models.
edit: commenting v3 (outdated) damn! is my first word when using this model, albeit not producing the realism i had hoped for, that's exactly what i did say. the result is beyond any pony model i've tried.
I definitely like v3 much more than v4, i find v4 hard to use (i'm using ipadapter and dmd2 lora). the results i've been getting from v4 are quite bad (using same parameters as v3)
Thank you for the feedback.
Are you using v4 (reborn) as if you were using a pony model? Because it is Illustrious/NoobAI-based (I've made a hybrid base model for training) and requires different quality tags, not score_9 etc as these WON'T do anything. Also it's important to use DPM 2M SDE sampler, not Euler A/regular DPM 2M.
I've tested it on a lot of prompts + highres fix + adetailer (where needed) and it gave me good results consistently. Otherwise I wouldn't publish it, I'm really picky about my stuff. You can try the embedded workflow from preview pictures, it's a simple one without anything fancy, but it has everything from above set up.
Anyway, a new improved Illustrious-based version will be out soon, it should be more stable. Can't tell if it actually is because I didn't have problems with the model in the first place.
@coyotte There is no problem with the model. But its up to you, what path do you choose.
Drop some concepts(furry, anthro, etc) and go for realism, or remain semireal with all possible knowledge.
@coyotte hmm, i'll redownload and give it more shot, i've developed new workflow which may suit hard to tame models (tamed bigasp2 even if the result seems out of place). i'll definitely return the result soon
sorry man, but honestly v4 is just not for me. i need ipadapter for consistency, and dmd2 lora for fewer steps. without those i had to run the sampling up to 157 seconds a generation, and i just don't think it will worth continue trying when i can generate better result in 17-48 secs. if i may advise: illus and pony are far from realism, you're looking for lots of hurt if you want to pioneer in this new breed. Unless you want to finetune the best of both world (pony and illus) into stylized art rather than realism, i suggest you improve the already good and proven concept v3. btw if you want to continue researching this, i suggest using gemini, google's ai chatbot. i developed new better workflow consulting gemini as my assistant. helped me finetune and troubleshoot what i lack. got much more in 2 days compared to months stumbling around reading texts and forums.
aaah, you're the brand name behind lustify... thank you, man! i never thought you will respond to small fry like me with no achievement yet. Great work with lustify and damn! v3. i'm still keeping lustify for a long time as my go to for nsfw.
Lustify developing lustify would be best, train more position, angle, props, etc
Damn... are you sure this is named correctly? Can't believe I have missed this one, I thought lustify was amazing. I think this should be named as WAIT WHAT!?!?!?
The face is easy to collapse and cannot be repaired by detailer. How to solve it?
Hires-fix.
What do you mean collapse? Prompt the face, look, eyes, emotions extensively. Add quality tags, add absurdres, detalied xyz(face, eyes etc).
it is a good model, but it ignores every angle command. You can prompt 1000 times "frontal view, frontal angle, from front" or "whataver", you can even add loras to force angles, they don't work as well... that's why i won't use it anymore... it's a pitty, because the results are good, but it ignores angle prompts, the model just desides the angle view for itself, you have no way to change the angle.
"frontal view, frontal angle, from front" aren't danbooru tags tho. You are trying to speak to this model in a language it does not understand.
Research this:
https://danbooru.donmai.us/wiki_pages/tag_group:image_composition
Danbooru tag for front view is "straight on"
Details
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.