Been a while. Post your ai furry videos, ask setup questions, get better at genning. Fap.LOCAL SETUPhttps://rentry.org/sdg-link>A1111 and Forge/reForge Installation Guideshttps://github.com/AUTOMATIC1111/stable-diffusion-webuihttps://rentry.org/fluffscaler-install>ComfyUIhttps://github.com/comfyanonymous/ComfyUIhttps://rentry.org/comfyui4a1111>AMDVladmandic Fork/SD.Next recommended:https://github.com/vladmandic/automatichttps://rentry.org/sd-nativeisekaitoo/TRASH/ RESOURCES>Models & Lorashttps://rentry.org/trashfaqhttps://rentry.org/trashcollectshttps://rentry.org/trashcollects_lorashttps://civitai.com/Wan-Video: https://github.com/Wan-Video/Wan2.1CogVideoX: https://github.com/THUDM/CogVideoMochi: https://github.com/genmoai/mochiHunyuanVideo: https://github.com/Tencent/HunyuanVideoLTX-Video: https://github.com/Lightricks/LTX-VideoVarious others: https://github.com/topics/image-to-videoAi Video Generation sites:https://www.vidu.studio/https://runwayml.com/https://www.klingai.com/https://hailuoai.video/https://lumalabs.ai/https://openai.com/sora/https://app.pixverse.ai/onboardhttps://viggle.ai/https://pollo.ai/>ControlNethttps://github.com/Mikubill/sd-webui-controlnet>Openpose editor extensionhttps://github.com/huchenlei/sd-webui-openpose-editorEXTRAS>AI boorushttps://e6ai.net/https://aibooru.online/>DiscordSD: discord.gg/StableDiffusionFurry: discord.gg/cNyuGvbT8U
close?
>>78169867damn...90s cartoons went hard
another attemptjust used the wan 2.2 with experimental NSFW lora. you know, the one on civitai that says "nightly updates" but hasn't been updated in over a month?it's kinda cool. even a little movement helps it look like these are bodies and not just renders of bodies.
Hmmm..idk if anyone here will know but if I have a 5090 (32GB VRAM) what setup should I be using for wan 2.2 for the best results? I figure this thing can handle some hefty workload considering I jumped from 8GB of VRAM (I didn't do it for AI, but fuck it I'll provide some content)Just need to know what workflow to use or what my setup should be. Like is there a specific model I should be using for 5090 that handles porn well and does high quality results?
>>78170385>you know, the one on civitai that says "nightly updates" but hasn't been updated in over a month?The furry one?
there's a nigga who used to make requests of Ai generated pictures into videos for Slop, and i have saved some of them (also because one or two are requests of mine)
lmao the captcha is >OWO PP
>>78174005n i c e
>>78170728Anyone??feels like a shame not to put a 5090 to use genning delicious furry porn. Need some guidance
>>78174005can you catbox this? would like to see the workflow!
I shall contribute a couple of videos generated with Wan 2.5 to this thread. The second has an audio track produced by Wan's audio feature, so here's a catbox.https://files.catbox.moe/vd9huj.mp4
>>78171265One of my favorites!
>>78174630Ironically i think you should ask that in /slop/
>>78174005>>78174981I wish i could but the original is work from somebody in /msdg/ https://desuarchive.org/trash/thread/76803049/#q76816589 and here its the second version as he call the first a failure
>>>/d/11381873I forgot how to crosspost, anywaysThere this fella on d's /sdg who currently has working loras or something with ai generation that has traction idk i'm a bypasser
https://limewire.com/d/odrrh#pmvjneQW4l
This shit still sucks.
>>78180865nah
>>78179875/slop/ is for images though, really not gonna be applicable to video genning
>>78175962>with Wan 2.5Uh what. Did it advance THAT much in 1 month? thought we were on wan 2.2?wtf happened to 2.3 and 2.4
>>78180785catbox for this? Nice
>>78182618sexo..
>>78181890It's behind saas garbage for the foreseeable future because money.
>>78168559My 3080 does well enough for imagegen but I'm guessing a XX90 is a must-have for reasonable vid-gen times?
>>781841133080 can do 480p 81 frames wan 2.1 with a lot of offloading, and it takes about 70 mins for one gen without the light2x lora.
>>78184113From my understanding (and someone can correct me if I'm wrong) the main key thing that affects video genning quality is the resolution (height x width) that you set your gen at.So a 480x480 gen will give pretty scuffed fucked up results (weird movement, fucked up anatomy, etc) whereas 1024x1024 will produce actual quality with less chance of errors.The thing is the higher you set the height x width nodes the more VRAM you will need to complete the gen. Lower cards will just get out of memory errors. I use 1024x1024 as an example, Idk what the "ideal" resolution is for things like 4090 and 5090 cards. If someone wants to chime in on this they can, I'm no expert but I believe this info is all accurate
What's the meta for local video setup?
>>78184113Experiment with low quant GGUFs. Q4 is probably the lowest you'd want to go. Reduce resolution, frames [length], CFG, and steps. You can upscale later and stitch videos together if you really want. Look for FAST workflows and LORAs designed for low step / 1 CFG workflows.
>>78185503What would be the opposite of this setup? What about if you have a 5090 and high specs?
>>78185562Use a full-precision model or Q8_0 GGUF. Higher resolution at the start to keep crisper details, like text. I still wouldn't go too high since upscaling is always an option later and you don't waste time and resources.You can also opt for longer videos, but WAN can start to loop if you go too high. Look into the RifleXRope node to try and address that. Honestly even with a high end card you can still benefit greatly from some of the time saving measures intended for lower end gpus, so I would still look into fast workflows and just crank some settings up to take advantage of higher tier cards.
>>78185811>Use a full-precision modelAny recommended one to try? currently I use "wan2.2_i2v_high_noise_14B_fp8_scaled.safetensors" and the low noise version as well.I still don't fully get what all the 14B, FP8, etc all even mean in these models, this was just the one I use. Which one would do better?
>>78171265needs a part 2
>>78189118>>78189240>>78189297>>78189318>>78189379Look at this gen made by someone recently on msdg
>>78189434neat
>>78189434>>78191455gen more like these pls
>>78189434What model is being used?
>>781707285090fag here still, looking for advice still if anyone knows
>>78193612no idea honestly
>>78194535I'm on a 3090 and I've been relying on ComfyUI's workflows for video gens. You can check out the Wan 2.2 ones at: https://docs.comfy.org/tutorials/video/wan/wan2_2The full sized Wan2.2 models might be too much for my machine or the workflow I'm using just isn't working right, but if you've got a 5090 then you've got the best consumer setup available. I'd go for the full 14B parameter low and high noise workflow and see if it works for you.Good luck!
>>78194854Thanks for the advice bud
>>78194924>>78194854Should of probably posted one of my video gens as an example.I can get Wan 2.1 video gens working using the ComfyUI workflows, and the smaller Wan2.2 models, but the big Wan2.2 models keep exploding for me. You'd likely get the best results with those if you can get them working though.
