Mommy's new friend

https://file.garden/aoZZZMbH5m5bEQ84/gdfgjwefryjhadgsdfaga.mp4

what happened lol? did the thread get removed for CP?

>>26860
I'm thinking because of the realistic videos uploaded directly.

>>26861
yeah makes sense

Please upload pics with ideas for animation or your horror clips (with links on external file hosting).

Wednesday not into love: https://file.garden/aoZZZMbH5m5bEQ84/dsfgdstth4fsdgrthdgsgsd.mp4

>>26464
>>26465
>>26467
What new venv? You only need to update and add --use-ck-attention to its .bat. Even stable has it now, so no need for nightly. Add H3 related nodes from comfyui-Kjnodes to your workflow as I said earlier. larryvrh is good but try ltx2v and alibaba-pai ones. A new native node sparse attention have been pushed into stable and it may make things even faster paired with ck-attention possibly without tuning them into shit. Use always 8 steps with turbos, even 2 more steps and final quality will improve a lot. Euler/simple is what you should use, other ODE solvers will burn and SDEs are obviously wrong, but er-sde is an exception, it will often outperform the others.
For vllm in comfy use ComfyUI_Simple_Qwen3-VL-gguf node, read the optimal parameters for your model (i guess you already know that since you use llmstudio) and you're good to go, or use LLMstudio as you said. Feed it a system prompt which will contain jailbreak and https://github.com/MiniMax-AI/MiniMax-H3/blob/main/.agents/skills/h3-prompt-writing/references/base-en.txt or https://github.com/MiniMax-AI/MiniMax-H3/blob/main/.agents/skills/h3-prompt-writing/references/ref-en.txt. I would suggest you using Gemma4 instead of Qwen. Gemma 12b-it is similar size as yours, but if you can use 26b: it's MOE, it's much bigger but runs x4 faster and performs 4x better.

>>26467
I'm not sure how good would this extend work. You can't simply reuse the prompt, you have to re caption based on the last scene/frame/image or whatever. But if it works, good for you

I forgot which lora I used for this one (cuz i have both Larryvx and Lighttx but still couldn't figure out which one's better as result is extremely vary) for this one and i end up trimming the last shot out in vegas because it bugged out.

[[using LLM to write the prompt]]
integrated_multimodal_description:
[Shot 1] The target video uses a 3D claymation style with gritty realism color tones, shot from a fixed wide-angle dashcam perspective mounted on the police car dashboard looking forward through the windshield. At 0.0 seconds into the target video, <Subject 1> (the robber girl) is riding a black motorcycle at high speed down a deep forest trail at night. The environment features dark pine trees lining both sides of the dirt road, illuminated by the alternating blue and red flashing LED strips from the police car's siren lights. The camera remains static with slight vibration from the engine and road surface. <Subject 1> wears a tight black latex suit that reflects the ambient light. The police car interior is dimly lit, showing part of the dashboard and windshield wipers.

[Shot 2] At 00:03.800, Keep camera on fixed wide-angle dashcam perspective mounted on the police car dashboard looking forward through the windshield. <Subject 1> turning her upper body sharply toward the front windshield while maintaining forward momentum on the bike. She pulls a pistol from her waistband with a quick motion and fires multiple shots directly at the police car's windshield. The claymation texture creates visible glass shards shattering outward upon impact, creating several distinct bullet holes in the transparent surface. <Subject 1> (S1) grunts slightly as she stabilizes the bike after firing: <d>[English] Hah!</d>.

[Shot 3] At 00:06.000, Keep camera on fixed wide-angle dashcam perspective mounted on the police car dashboard looking forward through the windshield. The female officer (S2) inside the car pulls her weapon and fires back through the bullet-riddled windshield. A high-velocity bullet strikes <Subject 1> in the upper back, causing a spray of red liquid (simulating blood) to spew out against the night background. The black motorcycle begins to wobble violently as control is lost due to the impact.

[Shot 4] At 00:08.000, Keep camera on fixed wide-angle dashcam perspective mounted on the police car dashboard looking forward through the windshield. the scene show the immediate aftermath of the crash. <Subject 1> and her mangled bike are thrown off the road and crash into a large pine tree on the side of the trail. The police car slows down rapidly and comes to a complete stop in front of the impact site. The sound of a heavy metal car door opening is heard clearly as the officer exits the vehicle.

[Shot 5] At 00:10.000, Keep camera on fixed wide-angle dashcam perspective mounted on the police car dashboard looking forward through the windshield. <Subject 2> (the female officer) walking out from the driver's side and standing next to <Subject 1>, who is now lying on her back in a pool of red liquid, bleeding heavily from multiple bullet holes. <Subject 1> lifts her head slightly and whimpers while begging for help: <d>[English] Please... help me...</d>. The officer (S2) pulls out a handgun with a steady hand and fires a single shot directly into the center of <Subject 1>'s forehead, causing a chunk of clay-textured material to blow out.

Comment too long, view post No.26502 to see the full comment.

>>26498

well since you said H3 works best at 15 sec and everything after that will be straight up *crap* (not to mention the generation time would make me want to commit artistry warcrime).

My plan is to use LLM and some guidance to write a long section. then snip everything using the context loop guide on how to boot the last gen and so on...

>>26498

So this is the result i got (had to crop out and speed up many parts to make it look like a quickdraw)

https://www.furaffinity.net/view/66229250/

Seems like neither Kyte nor Vixi read the fine print. Who knows when 'done' is, seems like they're going to be here for a looong time uwu

>>26714
Hot!

>>26804
We need more of girls being punched to death. Preferably using one hand while jerking himself with the other

>>26615
>>26617
>>26618
>>26619
Love these last four, the last two especially. English text is always a huge plus for me.

>>26617
first image where body After match body Before more or less. consider the body type and distinctive features of body Before, it all looks like two random pictures put together.

bro do you have any anima`s guro lora?

>>135354

idea was something like this.

conscript girl went to take a dump. only to get blown up by stray bomb drone because she's in its sight

>>135814
might get my hands on it when i’m free

>>135814
i made a short demo

>>135889
hot!!

>>135889

Well hot damn that's a well animated explosion, i like it :D

>>26724
nice but please repost video versions to another thread, this one only for images.

Well I'm back

>>26371
Dreadful. What is this garbage?

sadly the voice gen so bad :<

>>26371
amazing.. what model do you use?

using SLA attention so fast for minimax. from 1h to 15m when used without turbo lora.

trying more longer execution. kinda like the result. using eros model
https://files.catbox.moe/10zc6u.webm

Love the design of the petals. Totally on board for this!

>>26824
>>26825
Damn I can't wait to see him chew her tits. Maybe have him bite into her nipples and slowly tear them off.

>>26695
!!!

sample from a fantastic ai-artist... unfortunately he doesn´t specialize on debreastings... and longer vids are not free

>>26747
DARKARTAI was the first who managed to create realistic guro back in a day, but today with krea2 and all new models basically everyone can do it.

>>26695
would have been perfect without the armpit hair

how do you guys get grok to do that stuff?

How about some back views with the laser grid and no spaces after cutting, the laser is really thin for cutting to where you don't see the cut lines until she she starts to fall

>>26728
i'm using sage attention (still some type of version 2) because i barely know what i'm doing and i've just been too lazy to learn how to set up any newer stuff like that
i was generating at 720p 16:9 for a little bit but the results i was getting were bad, even for simple things
when i went back to 0.69 megapixel to channel meme magic i started getting better generations again

>>26725
yes

>>26730
very nice!

>>26735
thanks! i try to keep things interesting even when i'm just running the same slop multiple times

>>26755
>>thanks! i try to keep things interesting even when i'm just running the same slop multiple times
Yay, we still have various different ways to cut the meat. Like >>26679 or https://www.pixiv.net/artworks/132504460 for example.

>>26633
FleshReaper's images always animate so well

>>26740
>>26741
>>26742
>>26743
>>26744
>>26745
>>26746
>>26747
>>26748
>>26749

We've gone from dog food to sashimi. That is amazing work. Freaken' well done sir.

I've never seen the site moderation pop up other than sharing or generating the first turn or two. Anything after that is the AI guardrails limiting what can be output

This story is somewhat rare for IW as it is not in first person. Or to be more accurate you are a none entity watching the porn.

I have made a few changes compared to the original on IW, for the better I think. Though it was a pain to make one that would actually let me share it, I had to remove all the explicit mentions of guro and re playtest it all over again.

I love how this one is basically straight to the porn, if you want to.

Hermione deepthroating her father by turn two levels of straight to the porn.

I have only been told no by the moderation once and it just generated the turn anyway without me making any changes.

It might be because it's new, but this might be my favorite thing on IW right now. I've been on it all day and my hand dick and arm are all too sore to continue level of favorite.

I've only played it using Massive and the turns are usually around the low three hundreds. I have no idea if that is expensive or not.

Comment too long, view post No.26530 to see the full comment.

Dev added a new AI to the list, "Dolphin." Apparently it's custom trained for IW specifically. Haven't done much testing, but it is willing to do Guro content without complaint. And it's a bit cheaper than Smilo

>>26607
It's okay. Haven't run into any guardrails yet. Three free per day, so that's nice

https://infiniteworlds.app/shared/xVTCR2

Description is pretty spot on. Do whatever you want and everyone just shrugs and carries on