Best way to create a realistic, consistent NSFW character?

Vlad2012

Vlad2012

Elite Member
Joined
Dec 28, 2012
Messages
1,825
Reaction score
723
Hey everyone,
I’ve been trying for a while to create a realistic and consistent recurring character for my NSFW photos and videos, but I keep running into issues. I’d really like to find a reliable workflow or free tools setup that lets me keep my character’s look identical across different poses, outfits, and lighting conditions both in images and videos. If anyone here has tips, recommended models, or a solid process (ComfyUI, Stable Diffusion, Lora, face embedding, etc.), I’d love to hear your advice. Thanks in advance for your help!
 
Advertise on BHW
You can use ComfyUI, try Flux Kontext in any of the NSFW versions, and apply a hyper-realistic LoRA to keep the same character. You can also use Flux and ControlNet to maintain character consistency and control poses in a more customized way.

If you need to generate video, you can use a workflow based on WAN 2.2 All-in-One (NSFW version), which is very stable and delivers decent results even with 8 GB of VRAM.
 
joshua3100 said:
You can use ComfyUI, try Flux Kontext in any of the NSFW versions, and apply a hyper-realistic LoRA to keep the same character. You can also use Flux and ControlNet to maintain character consistency and control poses in a more customized way.

If you need to generate video, you can use a workflow based on WAN 2.2 All-in-One (NSFW version), which is very stable and delivers decent results even with 8 GB of VRAM.
Click to expand...
Flux + controlnet it's extremely heavy, best to create a LoRa and go from there
 
Alejandro Jose said:
Flux + controlnet it's extremely heavy, best to create a LoRa and go from there
Click to expand...
Alright, in that case it’s a different story, but how can you bring a fake AI influencer to life without spending a fortune, and still get consistent photo sets, Instagram videos, and adult videos for an adult platform?
 
Vlad2012 said:
Alright, in that case it’s a different story, but how can you bring a fake AI influencer to life without spending a fortune, and still get consistent photo sets, Instagram videos, and adult videos for an adult platform?
Click to expand...
Train a LoRa my friend, pick the right base model (don't use sdxl) and good prompting
 
Alejandro Jose said:
Train a LoRa my friend, pick the right base model (don't use sdxl) and good prompting
Click to expand...
Thanks for the advice! I’ve been experimenting with LoRA training so any reason you suggest avoiding SDXL?
 
Vlad2012 said:
Thanks for the advice! I’ve been experimenting with LoRA training so any reason you suggest avoiding SDXL?
Click to expand...
Because SDXL is shit, a lot of artifacts, horrible quality, and generic faces, for some stuff sdxl is great, but if you want your model to look really good, advoid it
 
joshua3100 said:
I use an RTX 3060 with 12GB, and ControlNet with Flux works well — though it’s limited to 1080p.
Click to expand...
Not saying it won't work well, I'm saying it will be more heavy on your resources and it will take longer, if you can get a Lora, it will be easier and more consistent, besides, face consistency with controlnet it's a bit hard and limited unless you use a face ID node or something like that, which will increase the VRAM used, so you want something better in the long term? make a LoRa
 
build a style-locked workflow. train a character LoRA on 20–30 clean, consistent portraits. use A1111 with fixed seeds, same checkpoint, and controlled lighting embeddings. for motion, use AnimateDiff or Pika with reference frames. keep dataset non-explicit, focus on face + body proportions.
 
DojaCat said:
build a style-locked workflow. train a character LoRA on 20–30 clean, consistent portraits. use A1111 with fixed seeds, same checkpoint, and controlled lighting embeddings. for motion, use AnimateDiff or Pika with reference frames. keep dataset non-explicit, focus on face + body proportions.
Click to expand...
That's one way to go, tho I don't like A1111 too heavy on resources, comfyui is lighter imo, would you care to share the embeddings you are using? If that's okay of course, also, why fixed seeds? And how do you identify good ones?
 
Alejandro Jose said:
That's one way to go, tho I don't like A1111 too heavy on resources, comfyui is lighter imo, would you care to share the embeddings you are using? If that's okay of course, also, why fixed seeds? And how do you identify good ones?
Click to expand...
A1111 heavy a bit but u get used to it I don’t share exact embeds, but use IP-Adapter FaceID/Plus + a light TI style embed; negatives: EasyNegative + bad-hands-5. Fixed seeds = repeatable framing/lighting. Hunt seeds via 6×6 grid, low CFG 3–4, 20–30 steps, Euler a; bank the cleanest, reuse across shoots.
 
DojaCat said:
I don’t share exact embeds
Click to expand...
:D
DojaCat said:
build a style-locked workflow. train a character LoRA on 20–30 clean, consistent portraits. use A1111 with fixed seeds, same checkpoint, and controlled lighting embeddings. for motion, use AnimateDiff or Pika with reference frames. keep dataset non-explicit, focus on face + body proportions.
Click to expand...
As to the other, it's impractical to reuse seeds since they don't work well with different prompts sometimes, also, given A1111 is heavy on resources already, adding IP-Adapter with the respective faceid to capture facial features adds a lot in resource consumption, which is impractical for day-to-day image generation, that's usefully for making a dataset -> LoRa, other than that I would not use it on day to day (sometimes even ip adapter messes up the face a little bit which may or may not be noticeable
 
In my opinion, Comfy UI is the best option for NSFW, it's not so easy at the beginning, but it's worth it. Pykaso is also okay, but in my opinion, not as good anymore.
 
Vlad2012 said:
Hey everyone,
I’ve been trying for a while to create a realistic and consistent recurring character for my NSFW photos and videos, but I keep running into issues. I’d really like to find a reliable workflow or free tools setup that lets me keep my character’s look identical across different poses, outfits, and lighting conditions both in images and videos. If anyone here has tips, recommended models, or a solid process (ComfyUI, Stable Diffusion, Lora, face embedding, etc.), I’d love to hear your advice. Thanks in advance for your help!
Click to expand...
Feel feel to try COMFY UI with spcific adult rendering tool or if you are not into the UX/UI I guess small tool like magichour https://magichour.ai/ can help as well, much easier!
 
andrewninjamtl said:
Feel feel to try COMFY UI with spcific adult rendering tool or if you are not into the UX/UI I guess small tool like magichour https://magichour.ai/ can help as well, much easier!
Click to expand...
Thanks man, but I'm not sure if MagicHour accepts NSFW content !
The Chinese tool from ByteDance does the job and creates a coherent character from a portrait image, and it also makes 12-second videos for $0 :)

For ComfyUI I found this


Models used in the workflow:

Next Scene Design Group
https://huggingface.co/CompVis/stable-diffusion-v1-4/tree/main
https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main
https://huggingface.co/stabilityai/stable-diffusion-2-1/tree/main

Next Image Large Gen
https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/tree/main
https://huggingface.co/SG161222/Realistic_Vision_V6.0/tree/main

Samyang Realism Lens
https://huggingface.co/NoCrypt/Samyang-lens-v1/tree/main

Video Generation Group
https://huggingface.co/cerspense/zeroscope_v2_576w/tree/main
https://huggingface.co/KandinskyCommunity/kandinsky-2-2-decoder/tree/main
https://huggingface.co/stabilityai/stable-video-diffusion-img2vid/tree/main
https://huggingface.co/ali-vilab/text-to-video-ms-1.7b/tree/main

Loras: (goes to Config/Models/Loras)
https://huggingface.co/h94/IP-Adapter/tree/main
https://huggingface.co/h94/IP-Adapter-FaceID/tree/main
https://huggingface.co/h94/IP-Adapter-Plus/tree/main

Text encoders:
https://huggingface.co/openai/clip-vit-large-patch14/tree/main
https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k/tree/main
 
Last edited:
Vlad2012 said:
Hey everyone,
I’ve been trying for a while to create a realistic and consistent recurring character for my NSFW photos and videos, but I keep running into issues. I’d really like to find a reliable workflow or free tools setup that lets me keep my character’s look identical across different poses, outfits, and lighting conditions both in images and videos. If anyone here has tips, recommended models, or a solid process (ComfyUI, Stable Diffusion, Lora, face embedding, etc.), I’d love to hear your advice. Thanks in advance for your help!
Click to expand...
Grab 20 selfies → train a 30 MB face-only LoRA on HuggingFace (free, 15 min).
In ComfyUI: SD 1.5 + your LoRA @ 0.8, ControlNet for pose, IP-Adapter with one driver pic.
Lock those three, random seed everything else—same face, any pose, zero cost.
 
Interesting discussion The key to consistency is keeping the same tone style and personality across all posts or content
 
SproutGigsTeam said:
You can use ComfyUI, try Flux Kontext in any of the NSFW versions, and apply a hyper-realistic LoRA to keep the same character. You can also use Flux and ControlNet to maintain character consistency and control poses in a more customized way.

If you need to generate video, you can use a workflow based on WAN 2.2 All-in-One (NSFW version), which is very stable and delivers decent results even with 8 GB of VRAM.
Click to expand...
can you give me a detail guide or recommend a video?
 
Balinummz said:
Grab 20 selfies → train a 30 MB face-only LoRA on HuggingFace (free, 15 min).
In ComfyUI: SD 1.5 + your LoRA @ 0.8, ControlNet for pose, IP-Adapter with one driver pic.
Lock those three, random seed everything else—same face, any pose, zero cost.
Click to expand...
can you help me to make this work?. I don't have an idea how to begin with?
 
Advertise on BHW
You must log in or register to reply here.
Sign up now!

Main Menu

Home Main Forum List Black Hat SEO White Hat SEO BHW Newbie Guide Blogging Black Hat Tools Social Media Downloads

Marketplace

Account Selling Content / Copywriting EBooks / Methods Hosting Misc Proxies For Sale SEO - Link building SEO - Packages Social Media Social Media - Panels Web Design More

Making Money

Affiliate Programs Hire a Freelancer IM Journeys Making Money Pay Per Click Site Flipping Want to Buy

BlackHat World

Dispute Resolution Forum Suggestions & Feedback Introductions Lounge
Back
Top