Doggystyle POV

Well, this (and many other, at least from this author) t2v loras works for i2v, sometimes you need just put the prompt as "POV doggystyle sex, Scene starts with a woman. After a black screen scene cuts to she is fucked by a man from behind, view from above." Or it may needs some additional loras, but in that case - works fine without any other loras, checked rn for an answer (whole day testing author's loras on i2v, pretty good results).

@forfreelsd368 Thanks for this advice! The POV doggy works now but as with his other loras the genital details are missing and the sex looks poor (not the movement only realism of the sex) even with HN and LN at 1. Thats my first problem, second problem is that for example that the side cowgirl riding lora wont work, i took your sentence changed it a little but i get an doggy fuck as an result. Kinda weird.. I guess its not easy to find the balance between these loras+the input pic and the prompt. But i will figure that out within time. BTW i use the boundbite10 V2 checkpoint it has an baked in 4step speed lora and mostly i use 4-8 total steps, resoloution always depends on the input picture. CFG always on 1 because of the speed lora. Anyways the missing sex details issue is my biggest problem atm, any advice?

Thanks for your answers! I'm using an DaSiwa workflow and DaSiWa Checkpoint Boundbite V10 Lightspeed. I also forgot to mention that i use I2V thats why your recommended vaginus lora doesnt work for me. I already added DR34ML4Y and some other loras to support the genitals but the results are far away from good enough. Should i try your workflow (not sure because its t2v) and if yes which checkpoint should i use.. You said smoothmix.. I'm about to try it out. About the noise: My checkpoint+workflow doesnt even allow me to change noise not even in the subgraphs so that might be an issue too. Again thanks for your time! Im very new to this so i have to figure things out alot at this point thats why im very thankful for any help!

@az420 Yeah i have the low noise around that strenght. I guess the issue is that if you want to use it properly on i2v you must train it again? The checkpoint and workflow i use is very good so i dont really think thats the reason for the issue. Today i tried the sideview BJ and the dick had almost no details even with LN on 1. Question is am i the only one with this issue?

@teeay25225 How are your Sigmas Split? What's your LowNoise LoRA Stack? Are you using one of the Lightning LoRAs on Low? How many steps do you have dedicated to the LowNoise pass, and what's your CFG/Sampler/Scheduler for Low?

Think of your LowNoise as a Logo-Placer. Your HighNoise sets up the 'box' and how it moves or behaves, whether someone opens the box, puts their hands in the box, hands the box off etc. Then your LowNoise pass comes through and slaps a bunch of Logos/Stickers on the visible edges of your boxes. Like, "Ok, this bouncing box is actually boobies. So you get a nice boobie sticker to cover this side of the box. And what do I see down here? Oh well that must be a Willy Holder. Gonna have to cover up any panties I already see and slap a big ole Willy Holder sticker right over that entire area. Ope.. gotta poke a hole in it real quick.."

So now you've got a scene full of moving boxes with stickers that make it look nice. But- If you have too few Sampling steps left over to the LowNoise pass, and you pick a Sampler that's half-blind, you're gonna get some weird stickers or missing stickers. Which sounds like what you're describing.

There is one other possibility. If you're using the wan-2.2_Text-2-Video High+Low main models with the wan-2.2_>Image<-2-Video 4-step Lightning LoRA's, you'll also get the missing stickers effect. Since the function of the 4-Step Lightning LoRAs is to make the possible movement of the boxes very limited, and to reduce the selection of available sticker options, try running a really short (21 frames) test with the LowNoise Lightning LoRA disabled/off/0.0

@NeonScreams Thanks alot for your answer. I tried it without speed lora with the Boundbite Truevision checkpoint but there still wasnt enough details! I tried different samplers, steps.. cfg and so on. Then.. One render had more details with sigmashift low from 5 to 6. But this is not really the soloution. I then returned to the speed lora and found out that at a higher resolution (in this case my fav final render res. 1072X608) the details were coming. That's not the case with other loras where you have already details at lower resolutions. Maybe an checkpoint which has female genitals already included would be my soloution? But "smoothmix" is more like anime style and my current checkpoint says it got nsfw included but not really the case.. (Boundbite V10). I did a render at high res and the Three Quarter Rear Doggy and no anus appeared anyways thats why a high res. render seems not like the final soloution, and btw the lora at this render was both high and low on 1. MAYBE if you have the time and energy you could test my setup, this would give the soloution i guess, but maybe just by reading my config will be enough: Workflow: DaSiWa WAN 2.2 i2v FastFidelity C-AiO-12. Checkpoint Boundbite V10 Highspeed (it has an baked in 4step speed lora) and kinda everything on default.. for example Sigmashift low+high at 5. Step to swap: 2. CFG most time at 1. At the I2V subgraph i cant modify noise, only start+end steps and sampler+scheduler. Clip: umt5_xxl_fp8e4m3fn. VAE: Wan 2.1... At the lora stack i only put one az420 lora high+low (supporting loras didnt help like i said). I also tested different res+samplers+more steps indeed. Maybe i need a whole other checkpoint+workflow? If you tell me exactly what to do i could test again. I'm willing to learn because i love video editing and also do regular video editing with Davinci Resolve. Thanks for your time.

@NeonScreams Thanks alot for your answer. I tried it without speed lora with the Boundbite Truevision checkpoint but there still wasnt enough details! I tried different samplers, steps.. cfg and so on. Then.. One render had more details with sigmashift low from 5 to 6. But this is not really the soloution. I then returned to the speed lora and found out that at a higher resolution (in this case my fav final render res. 1072X608) the details were coming. That's not the case with other loras where you have already details at lower resolutions. Maybe an checkpoint which has female genitals already included would be my soloution? But "smoothmix" is more like anime style and my current checkpoint says it got nsfw included but not really the case.. (Boundbite V10). I did a render at high res and the Three Quarter Rear Doggy and no anus appeared anyways thats why a high res. render seems not like the final soloution, and btw the lora at this render was both high and low on 1. MAYBE if you have the time and energy you could test my setup, this would give the soloution i guess, but maybe just by reading my config will be enough: Workflow: DaSiWa WAN 2.2 i2v FastFidelity C-AiO-12. Checkpoint Boundbite V10 Highspeed (it has an baked in 4step speed lora) and kinda everything on default.. for example Sigmashift low+high at 5. Step to swap: 2. CFG most time at 1. At the I2V subgraph i cant modify noise, only start+end steps and sampler+scheduler. Clip: umt5_xxl_fp8e4m3fn. VAE: Wan 2.1... At the lora stack i only put one az420 lora high+low (supporting loras didnt help like i said). I also tested different res+samplers+more steps indeed. Maybe i need a whole other checkpoint+workflow? If you tell me exactly what to do i could test again. I'm willing to learn because i love video editing and also do regular video editing with Davinci Resolve. Thanks for your time.