Animated images to realistic styleSlight inpainting availableNothing that'll get me bannedSuggestions for a better board than /b/ or better chan welcome
>>949769766
>>949770223Thanks
Pls
>>949770303
>>949770539Ty!
a challenge
>>949770239
>>949770870
>>949770920
>>949770985
>>949771571impressive bro, what are you using?
thanks
>>949771645>>949559703 >>949771767
please
>>949771554Ty
>>949772636
>>949772211amazing ty
I didn't prompt for pubes or rain, but i'm not entirely upset
>>949772923ty>>949773073i like the colorful pubes
>>949772882
>>949772947
>>949775175
>>949778789ty.. i like it a lot o7
>>949774045
>>949773387
>>949779016thanksss
what about furries?
>>949773481
>>949779845
>>949780026
>>949783199
>>949783709
>>949789344
>>949786953
>>949791709
>>949792007
i am curious
>>949791605
>>949792941
>>949793162
What happened with the people that were good at these?
>>949790717>>949792341>>949780176are these being outputted directly like this (tiled) or are you stitching them together afterwards?If the former, can you show me how?
>>949795938Not him, but I use ImageConcat after an Image Resize node, but I think the former can only tile two images
>>949795938There are nodes for joining images, but since these are multiple outputs run individually, not a batch gen, I use after the fact photo joining picking the successful ones.>>949793470
>>949796011that's some pretty impressive stuff right there, I'd like to see those workflow settings/prompt
>>949797537https://litter.catbox.moe/5tx176b9lat326pv.png
>>949798620thanks
>>949798620aw, its broken
>>949802898
>>949801692>>949801367https://litter.catbox.moe/mxg6yekx4w7csjql.pngforgot to select anything for 1h
>>949805413thank you!
>>949805498>>949804444>>949805359thanks kindly, I'll check it out later
I tried it with my favorite picture. Can you do better?
>>949808452>>949808033What model was used here? Despite the significant change, all proportions are maintained.
going for a walk and then I'll so some requests, so request away, preferably with no penises because I can't get those to work >>949816066same as in >>949805359different prompt but I think I didn't use any extra LORAs in those two besides the first two
can we get a /slop/ board for ai shit?
My latinx loli please.
>>949816637It is link to an image, i don't know how to read it's workflow XD. What is Basic Model name ?
klein 9b is incredible
>>949817626because it made that drawing look like a 2016 youtube thumbnail?
>>949818413I mean yeah
>>949818607it didn't retain any of the character of the original, or even the actual action being depicted
>>949816155
>>949819163terrible
>>949816296
>>949816189
>>949816272
>>949816388
>>949816840
>>949817270Go to https://comfyui-metadata.com/ and upload that image, it's the fastest way becaise the workflow is a bit complicated the models are flux-2-klein-base-9b-fp8 with a qwen_3_8b_fp8mixed clip and a flux2-vae VAE>>949819932tell me how to do penises
>>949815916>have no idea how to prompt a toothless smile>Color matching hair is hard
>>949816446>>949817270>drop image into comfyui>workflow appears
>>949819932left is cool
>>949821331very nice
>>949820475>"Detailed Prompt">make image1 into a realistic photographic nsfw amateur render, Make this photo high-quality, do not change pose or composition, woman with black hair surrounded by erect penis cum on face happyhttps://files.catbox.moe/yi3k2v.png
>>949820295Thanks
>>949817626>Can't figure out how to convince the AI not her arms
>>949817626>>949818413>>949818749>>949822347I spent too long trying to do accurate photorealistic conversions lately. Seeing this thread, I guess I'll share the random shit I put together.>1. use image recognition to generate a prompt. currently using qwen3.6-27bWonder Woman, woman kneeling between man's spread legs, chest pressed against erect penis, penis compressed between breasts, breast job, tit fuck, red rope binding arms behind back, red bicep straps, man's hands gripping black glossy hair pulling head back, golden tiara with red star, teeth clenched, strained facial expression, cheek blush, thick semen dripping from chin, semen splattered across chest and gold bustier emblem, indoor window lighting, specular highlights on glossy skin, static medium close-up, 50mm lens, breast compression>2. do a first pass on an image edit model, prompting to just change to photorealistic style while not yet changing anything else. Currently using Qwen Image Edit 2511I just do this step for rough edits. The image edit outputs are usually not useful by themselves, but I clean them up with SD img2img.>3. swap to stable diffusion with controlnet depth, img2img once with strong controlnet, then a second pass with low strength. I use PonyRealism but it's probably shit and outdated, so use whatever. The point is getting an accurate composition with photorealistic colors and minor anatomy tweaks. More precise anatomy corrections can be made with further iteration or swapping to a better model. Pay attention so you don't accidentally lose details like I did with her teeth in this step.>4. my best local realism model is currently Wan2.2. To simulate img2img, apply lens blur to the StableDiffusion output, ask Wan to focus on the image.It takes a long time to generate these and it often requires a lot of trial and error so I don't have an example. Results look better when they don't turn into body horror.
>>949822216>>949822473No a counter argument, just my take:>1.Image recognition? Like, WD14?>2., 4.Not a strong enough GPU to fiddle with Qwen Image Edit, specifically in reference to gen speed>3.SDXL Img2Img does give me more realistic results, but less accurate. I suppose you or method would help, but when I pull out SDXL it is for specific processesLong story short, I'm lazy and these edits are quick and dirty and not representative of a caring genner.
>>949822216
>>949822263
>>949822916>Like WD14No, an LLM with image support. Like a local ChatGPT that you can give an image to. I'm using qwen3.6-27b-heretic-uncensored-finetune-neo-code-di-imatrix-max. You need to give it a prompt or instructions on how to format the output. >GPU to fiddle with Qwen Image EditThis is actually the lightest model I used. It's "only" 20GB, plus the NSFW Lora. >but less accurateThat's what the strong ControlNet Depth on the first pass is for. It conforms the shape of the image to the original anime source, while using the color, lighting and image context you supply it.
>>949822473>>949824169I appreciate the tips, I made that first wonder woman and would love to be able to generate images that can maintain the original's expression and composition even if it's more work, I like the idea of feeding it to a sdxl controlnet>>949822347I had that same issue and had to make the prompt more detailed and run a few seeds before it finally understood it's not her arms
>>949824644
>>949824758The quick summary of my thoughts: Convert the color and lighting first while maintaining as much of the physical shape as possible. The result will be cursed, but it gives img2img a lot to work with and it's easy to fix the cursed anatomy and faces in two or three steps. Final step is feeding it into video, and not just for animation. Pic related. The dick and cum on the tits is fucked up, and I could fix it with more outputs or Loras, but generally speaking they're good at buffing out the subtle CGI look to things.
>>949825104Nice
>>949825583
>>949826396
>>949824758>>949825583Trying to adjust my process without needing so much VRAM. Manually write an accurate SDXL prompt for your model, use strong ControlNet Depth on text2img, manual corrections in Photoshop, img2img, inpaint/ADetailer.
>>949826396Absolutely LOVE the body hair by the way. Love a thick natural woman
>>949826673
>>949826396What about this one but she's holding the garment to the side, showing her pussy. The second one on the left is my favorite one honestly
the original drawings are still better, but i have a feeling there's going to be a post someday that's going to be so good it'll change my mind completely
>>949768420
Bump
