A model that focuses on PoV perspective (cowgirl, reverse cowgirl, blowjobs). Skin texturing from analog-diffusion-v1 (many cross merges back to analog-diffusion to keep some of the texturing, lighting and poses). Dreamlike Diffusion 1.0 to add in more color range and versatility with different aesthetics. r34 to make it have better compositions. f222 to have some simpler anatomy.
Downsides of the model are
Bigger boobs generally
Face styling can be too stylized sometimes
Doesn't work well at 512x512 (needs bigger sizes, or more abstract sizes). I am running 384x640 for many of these images, and then using hires fix scaling (upscaling in A1111/SD-Webui) to 768x1280.
Needs highres or SD upscaling to look great in the eyes and face.
Only PoV poses, and has a hard time with other sexual poses.
Upsides:
Responds very well to camera settings, camera models, post-processing.
Responds well to lighting and environmental conditions
Tuned to limit body horror but it still happens occasionally.
Due to using Dreamlike Diffusion 1.0, this model has the following license:
License
This model is licesed under a modified CreativeML OpenRAIL-M license.
- You can't host or use the model or its derivatives on websites/apps/etc., from which you earn, will earn, or plan to earn revenue or donations. If you want to, please email us at [email protected]
- You are free to host the model card and files (Without any actual inference or finetuning) on both commercial and non-commercial websites/apps/etc. Please state the full model name (Dreamlike Diffusion 1.0) and include a link to the model card (https://huggingface.co/dreamlike-art/dreamlike-diffusion-1.0)
- You are free to host the model or its derivatives on completely non-commercial websites/apps/etc (Meaning you are not getting ANY revenue or donations). Please state the full model name (Dreamlike Diffusion 1.0) and include a link to the model card (https://huggingface.co/dreamlike-art/dreamlike-diffusion-1.0)
- You are free to use the outputs of the model or the outputs of the model's derivatives for commercial purposes in teams of 10 or less
- You can't use the model to deliberately produce nor share illegal or harmful outputs or content
- The authors claims no rights on the outputs you generate, you are free to use them and are accountable for their use which must not go against the provisions set in the license
- You may re-distribute the weights. If you do, please be aware you have to include the same use restrictions as the ones in the license and share a copy of the modified CreativeML OpenRAIL-M to all your users (please read the license entirely and carefully) Please read the full license here: https://huggingface.co/dreamlike-art/dreamlike-diffusion-1.0/blob/main/LICENSE.md
Description
Merge of pyro_pov, analog-diffusion-v1, f222, dreamlike diffusion, and r34 models.
Fix random noise issue.
FAQ
Comments (24)
I tried it out and also checked the other comments, but for me, it still creates some weird not lookalike creations. I use NMKD Stable Diffusion and all the prompts given from the pics
What dimensions are you trying? Are you doing a highres fix or img2img upscaling? This model has problems at smaller resolutions but works much better with a second pass to upscale it. Also, probably not a perfect model but let me know if these help.
Thanks for the fast reply! I use 640x640, just gonna experiment a little and keep you updated if some weird stuff appears again. Like a dog out of nowhere :D
I think the r34 model might have some oddities that might creep in. Weirdly I didn't have any dogs show up but was running some tests yesterday and got a dog (and woman) and a dog-headed woman. Not sure I can explain it though, seems very rare.
Thanks for the feedback :).
I think it kinda depends on how close you want the render to be to the original ref pic, when i put a higher number in I get better results, with lower numbers I get this wild relics
What number are you referring to? Thanks
@twilightBOO the initialization image strength ( influence number ) (1<=x=>9)
this model has a ton of positional just needs some more POV angles unfortunately it feels very limited as alot of the angles are the same
How do I find this model on huggingface? Colab's Dreambooth doesn't recognize the Civitai link
I will try to get this up on huggingface. Thanks!
Tossed it up there, the full 5GB unpruned version https://huggingface.co/twilightBOO/pov-skin-textures-dreamlike-r34-v2
I've checked, there's no yours ckpt in huggingface, could you reupload it?
@fgj There is a ckpt on here you can download (the pickle tensor) which I would suggest. Unless you have a need to get it through HF specifically?
Unlike other models on this site, the model doesn't get imported directly to Draw Things. Why is that?
Hmm not familiar, maybe because I am mostly only using the safetensors. I can try to get up a ckpt.
Although I really have no idea what you are talking about, it looks Draw Anything does not support safetensors extention.
What can I do about it?
@Steve12343 The pruned pickle tensor, the 3.xGB version should work if you need a ckpt. Not sure how Draw Things imports it.
@Steve12343 Draw Things imports safetensors just fine.
Download the model. In DrawThings pick the folder you downloaded to. Normally DT should show you ALL the models in this folder you've ever downloaded for the usage. I advise you to use the normal Downloads folder on Mac. DT asks you to delete the safetensors-file after it has implemented it.
Have fun! ;)
Loved using this on my setup when I was restricted to just CPU.. Just finished setting things up for my AMD 7900 XT, had to use ONNX route... Is there anyone who knows how to convert these models to onnx?
Maybe someone can give you better info/tutorial but possibly use https://github.com/huggingface/diffusers/blob/main/scripts/convert_stable_diffusion_checkpoint_to_onnx.py to convert it over. https://huggingface.co/twilightBOO/pov-skin-textures-dreamlike-r34-v2 for the model to use with this script (diffusers version is up on hugging face).
python convert_stable_diffusion_checkpoint_to_onnx.py --model_path "twilightBOO/pov-skin-textures-dreamlike-r34-v2"
I'm only really guessing but maybe that will help.
@twilightBOO oh wow, thanks a ton!
Can you upload your pov models in 1.99gb pruned version?
Should be labeled now, but I suggest the safetensors version. Likely won't prune the ckpt for this version.