>>1916919 >>1916922 Did you prompt for furry,anthro?

>>1917087 Yeah I realised. I don't get how the showcase art is so different from what I get? Also what model is that?

>>1910320 >Fantina content >>1918761 It's Noobai.

>>1918876 thanks. gonna try it out

>>1918876 I could use a bit of help. I am getting really amateur results out of noobai

>>1918876 That black mage is very cuddlable.

Any idea what model or models this guy uses? the art is phenomenal https://x.com/verbotenvertex

>>1919491 It's definitely something local, possibly even a bespoke setup. The actual style reminds me of older models like Yiffy, that really soft kind of pastel look was common in that era. I can tell you right now these are also edited and quite heavily so, local rawgens don't tend to look this good or be this complex while maintaining coherence. I'm guessing he's generating locations, backgrounds and objects to create a scene, then characters, then edits them all together while touching them up and making tweaks. These for example, were obviously edited from multiple different gens. That last one looks like he crafted both girls out of mirrors of same image, so his process is quite involved.

Finally installed SillyTavern. Now, I have the painfull task of setting up everything for my ERP adventures based on Mythras.

Shrek and his boy wife

>>1920223 Damn, whole point of shrek is that he got a wife as ugly as himself.

>>1911633 This is great, just protip: always count the knuckles. The AI loves to fuck them up and most people miss them.

>>1918876 The thigh highs looked like part of her skin until around her ankles. It's a little confusing.

>>1911633 >>1920601 Hotfix.

i2v image source: >>1723140 >>>/ais/5897 More /loli/ junk. Vocal generation used Japanese voices for source audio. Video gens for Kaolla kept dropping details between segments annoyingly. Imperfect, but can't keep dumping time / energy into re-rolling. Tried Q8 & fp16 to see if I'd get less smudgy motion with 2D stuff. Didn't seem to help. Anime style LoRA for WAN 2.2 may have cleaned up motion slightly for 2D, otherwise placebo. Back to fp8 instead of fp16 as I saw no visible difference right now. SmoothMixWAN Q8 works & didn't run OOM. I prob need more system memory regardless, but RAM prices are up. PROTIP: If you've used uv to install anything run the command "uv cache clean" or it'll keep wasting space on your OS drive with unnecessary package archives.

>>1920759 Neat. Saved.

Trying SmoothMixWAN 2.2 on an older PonyRealism image. i2v source image: >>>/ais/5323 Past attempts >>>/ais/14259 >>>/ais/14261 Sectional workflow can prompt different actions per section. Had 3 sections for establishing shot, show banner, walk away. Found limit of WAN 2.2 text generation when I tried to make banner with text "Where are the video games?" it'd fumble trying to make a full sentence so seems I need to keep it simple or maybe try different text encoder.

Three days after installing Sillytavern I must say... I'm pretty disappointed. I'm not sure if it's the prompt, the Mistral 7b, the configuration or what, but it feels stupid and boring. I'm not sure what I was expecting, maybe something better than Perchance's RPG AI or similar but with bigger control.

>>1922210 >face of someone who goes on a rant about how Donald Trump sends too much money to israel

>>1924394 not brown enough

>>1919208 Start by prompting with artists,tags, and art mediums Noobai recognizes. For me it's watercolor_\(medium\),limited_palette.

Pls help. How does one combine a model with a lora? I want to use this lora: https://civitai.com/models/340435/cervina7-ponyxl-style-lora With the following model on comfyui https://civitai.com/models/340435/cervina7-ponyxl-style-lora >>1924570 Can you give me a verbatim prompt to copy to see if I am using this correctly? I have noobaiXLNAAIXL_vPred10Version

>>1924570 >tags been awhile since i've done anything with gens, when i stopped every new model was using terrible natural text training data. Has this improved at all, are models finally reverting back to the vastly superior tag system yet?

>>1924583 watercolor_\(medium\),limited_palette,(wamudraws:0.3), (krekkov:0.6),2girls,hug_from_behind, sabrina_\(pokemon\), misty_\(pokemon\), I used r3mix for the first two and this version of noob for the last two: https://civitai.com/models/1201815?modelVersionId=1491533 >>1924594 >Has this improved at all, are models finally reverting back to the vastly superior tag system yet? For me Noob is the best model right now for anime and furry since it's trained on both danbooru and e621 tags. The one thing that hasn't changed is that you will train a LoRA, inpaint to hell, or both if a concept isn't working out of box. And there is a lot of straight up useless tags.

>>1924655 Is it possible to have two different python versions installed at the same time? I want to run Automatic1111 (which i have installed now)or maybe forge and comfyUI to run noobai. or maybe I can run noobai on automatic1111?

Tried WAN 2.2 assertive cowgirl LoRA: https://civitai.com/models/1566648 & anime creampie LoRA: https://civitai.com/models/1941545 Used Chel & Tulio voices from The Road to El Dorado for voice cloning. >>1924848 See my 3rd attached image. If a WebUI installation has a "python_embeded" folder then you use the python executable in there, which can be any version that is needed. This forgoes using the python installed in AppData. There are however cases where you need the python in AppData to get started though. I've had to install multiple python version in AppData & change the PATH in windows Environmental Variables back & forth between them in those cases (which is annoying). The directory containing the python executable is not always named python_embedded either. In my example: python_embeded\python.exe --version Command checks the version of the python executable inside the python_embeded directory (assuming you ran cmd from the base directory as shown). You can also pip install packages and requirements this way. Example: python_embeded\python.exe -m pip install triton-windows==3.5.0.post21 This command would install triton-windows to "python_embeded\Lib\site-packages" directory. (assuming you needed triton) >>1924344 I went from TheBloke_LLaMA2-13B-Tiefighter-GPTQ to Beyonder-4x7B-v2-GPTQ model & I was not very impressed either. >>1924583 Base NAI model kinda needs artist tags to make something decent. NAI checkpoint merges can help with that since they're generally intended to get a decent result with less effort. I'm using JANKU (EPS prediction, I think) https://civitai.com/models/1277670?modelVersionId=1896532 which is Illustrious XL + NAI XL + training. I'd still highly recommend playing around with artist tags to get variety of styles. Here's a heap of artist tags you could try to see which ones you like (there are tons more than this): (artist:wamudraws), (artist:incase), (artist:alex_ahad), (artist:rariatto_/(ganguri/), (artist:barbarian_tk), (artist:hews), (artist:optionaltypo, retro_artstyle), (artist:akableak), toon \(style\), flat color, (artist:gerph), (artist:ixy), (artist:koyorin), (artist:iesupa:0.9), (artist:alkemanubis), (artist:cutesexyrobutts), (artist:kuvshinov_ilya), (artist:naga_u), (artist:kanikama), (artist:ratatatat74), (artist:futa_/(nabezoko/)), (artist:andava), (artist:wlop), (artist:mika_pikazo),

Not my style, but I found it neat. https://civitai.com/models/2117217/illustrious-lucky-star-anime-style

Messing around with clothing tags is pretty fun. The autocomplete extension helps a lot in finding specific clothing tags to combine and make custom sexy outfits. 1st pic uses highleg leotard, thigh cutout, cleavage cutout 2nd pic uses cross-laced top, plunging neckline, garter straps, pelvic curtain, ornate clothes

Anyone know any good models or loras that could make something like this character?

Some weird stuff. image-2-video source images: >>1785485 >>1846460 >>1928859 I clearly hadn't experimented enough with clothing since I only ever saw like two or three of those tags via auto-complete.

>>1929517 Unless that's the entire corpus of that's character's appearance, you can probably make a model of that exact character. Monochrome comics are entirely doable training data, though some cleaning may be needed. https://civitai.com/articles/8370/nanashianons-cool-lora-training-guide-for-attractive-people

>>1929556 Very good result with the first one, I like how she moves around like an animate doll.

So, is NSFW content in sillytavern and KoboldCPP actived by default or do I need to jailbreak it? Also, I see Wam can be installed in Kobold, does it work well there? Also ignore the deleted message.

>>1929939 I don't think I've ever encounter a censored local model, you can fuck your home grown bots as much as you want.

>>1930500 Ok, then is just me not making a proper configuration be it in prompt or in the IA. I already have more that 30 entraces in the world's lore book and I expect to add at least 10 or more.