AI Workflow Any ComfyUI IMG2IMG workflow/tutorial?


This is not a real person, created synthetically on ZIT:
Prompt: "a blonde woman,black shirt, black underwear, neutral background"
Seed:417222973302885
Random seed variance: 538821508460696, 25%, Strength: 6.0
 
if by workflows you mean the jpg's on black forest's github, they're restricted can't be dled
 
The workflows are published on Huggingface like I just said. If you seriously can't google "Flux Klein 9b image edit workflow huggingface" you're not gonna make it.
 
i have the klein 9b workflow but one issue i have is that the resulting character does not look like the original character. i tried increasing the reference image strength (the main character) to 1.3 but it didnt help. what im trying to do is based on two images to mimic the pose of the character in image 2. any tips on how to get it more accurate? ia lso did add the SNOFS lora like you mentioned here.
edit: i mean i added a LORA using the method here. I didnt mean to say you suggested to add SNOPS sorry
 
I can't get it to work properly. It feels like the developer built an NSFW blocker into i2i. Or am I doing something wrong?
 
krea turbo I2I easy mode just vae encode an image and denoise at like .5 . works pretty well
 
Are you using regular 9b or the distilled model? Distilled requires 4 steps, cfg 1. 1.3 is way too fucking high. Do not touch cfg.
If you're trying to use it as a character or faceswap, not for nudifying then just give up now. You need to create a workflow to swap faces with a character lora instead. There's a few good ones on Reddit that use Z image.
 
for asian woman, klein sucks when it come to generate Douyin aesthetics female subject, it skew extremely toward asian baddie appearance and too much setup needed to get correct anatomy when generating 2subject photo. Qwen does perfect job for these 2 goals.
Now the basic workflow for qwen rapid aio can be made easily with claude, if u passionate enough u make it work.
 
ok noted. i think the default model the klien template uses is the distilled one ill check. if you can drop a template or link one that would be goated
i am not doing a face swap. i use face fusion for that. im trying to do make different poses with the reference images.
 
If you let Klein move body parts around, the model takes over influencing the image instead of the reference, you're not using it correctly. A workflow isn't going to help you if you can't understand how the model works.
 
Crazy how complex it is to get a working model/workflow for i2i with cloning face. I was thinking i2i would be much more simple and easy to just spam a tonne of reference images out for better i2v but christ, i've been all around the world on various models, controlnet, ipadapter face id etc i'm burnt tf out
 
I'll repeat myself. Train a lora and inpaint with facedetailer. Keep doing dumb shit and your outputs will be shit.
 
I understand your pain, Comfyui is a goddamn maze. I've barely managed to get the hang of 2 workflows.
I recommend this workflow by poppylover2.
If you're at the point where you open the workflow and things appear broken or you don't have a specific file, search it on google and it should come up on the first result, after that it's a matter of figuring out where each thing goes.
This post also helped me get an idea of where things go.
As for the face issue, I have not figured out any perfect solution.
What I personally do is have more in the "negative" input of the workflow than the positive one. Check spoiler for what I use most of the time.
The second thing I do, since I'm changing a specific base image, is put the original on photoshop then overlay it with the generated one.
I edit it all to match, then erase the face on the generated image.
It's by no means perfect, but so far it's the best I've come up with.
blurry, soft focus, out of focus, smeared details, motion blur, low detail, artifacts, oversharpening halos, noise,
pixelation, distorted features, body modification, altered anatomy, changed physique, different body shape,
larger breasts, enhanced bust, exaggerated curves, tiny waist, hourglass figure,
muscular physique, increased muscle mass, broader shoulders, bodybuilder,
athletic transformation, changed proportions, altered silhouette,
different facial features, beautification, glamourization, idealized body,
unrealistic anatomy, stylized anatomy, exaggerated sexual characteristics,
age change, gender change, weight change, height change,
plastic surgery look, cosmetic enhancement,
oversharpening, artifacts, extra details, invented details,
hallucinated features, synthetic texture, unrealistic skin texture
 
My next attempt today is going to be using sdxl and loras for a random NSFW image I want and then using sarcastictofu's (civ red) klein gguf faceswap and maybe his character dataset creator then potentially using the output good images to maybe train a small lora of a character... I'm determined not to have wasted all this time 😂
 
Can you recommend an LLM that works well with the RTX 3070 GPU in my laptop? I'm new to this.
 
Please help with WAN/VACE inpainting workflows - where to find support now?
 
Hi guys, i need your help. :peepoBless:
I have the portable version of ComfyUI and want a workflow that allows for img2img—specifically, taking an image and editing it to achieve realistic results.
Does anyone know of a workflow I could use that takes advantage of my 5060Ti 16GB card?
I’m not very experienced with ComfyUI, but if I have a workflow and some basic instructions, I can manage. :pepoLove:
 
I use Anima as my main model since its incredibly coherent for its size and also decently fast too. It's mostly for anime but there are decent enough realistic models I use. Are you trying to specifically make images better / copy the composition or trying to inpaint? As in adding/changing/removing objects in the image?
I can give you very basic workflow and tutorial. You have a decent GPU for this. I do all this on my 4060 with 8gb
 
I'm looking for a realistic "erase and replace" style NSFW image editor.
I'm not sure if it fits what you have.
 
You must log in or register to reply here.
Back
Top