Check my exclusive models on Mage: ParagonXL / NovaXL / NovaXL Lightning / NovaXL V2 / NovaXL Pony / NovaXL Pony Lightning / RealDreamXL / RealDreamXL Lightning
Recommendations for using the Hyper model:
Sampler = DPM SDE++ Karras or another / 4-6+ steps
CFG Scale = 1.5-2.0 (the lower the value, the more mutations, but the less contrast)
I also recommend using ADetailer for generation (some examples were generated with ADetailer, this will be noted in the image comments).
This model is available on Mage.Space (main sponsor).
You can also support me directly on Boosty.
Realistic Vision V6.0 (B2 - Full Re-train) Status (Updated: Apr. 4, 2024):
- Training Images: +3400 (B1: 3000)
- Training Steps: +724k (B1: 664k)
- Approximate percentage of completion: ~30%All models, including Realistic Vision (VAE / noVAE) are also on Hugging Face
ᅠ
Please read this! How to remove strong contrast.
To make the image less contrasty you can use LoRA [Detail Tweaker LoRA] in a negative value.
ᅠ
Orange Color = Optional
ᅠ
I use this template to get good generation results:
ᅠ
Prompt:
RAW photo, subject, 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3
ᅠ
Negative Prompt:
(deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime), text, cropped, out of frame, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck, UnrealisticDream
(deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime, mutated hands and fingers:1.4), (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, disconnected limbs, mutation, mutated, ugly, disgusting, amputation, UnrealisticDream
ᅠ
Euler A or DPM++ SDE Karras
CFG Scale 3,5 - 7
Hires. fix with 4x-UltraSharp upscaler
Denoising strength 0.25-0.45
Upscale by 1.1-2.0
Clip Skip 1-2
ENSD 31337
ᅠ
Thanks to the creators of these models for their work. Without them it would not have been possible to create this model.
HassanBlend 1.5.1.2 by sdhassan
Uber Realistic Porn Merge (URPM) by saftle
Protogen x3.4 (Photorealism) + Protogen x5.3 (Photorealism) by darkstorm2150
Art & Eros (aEros) + RealEldenApocalypse by aine_captain
Dreamlike Photoreal 2.0 by sviasem
HASDX by bestjammer
Analog Diffusion by wavymulder
Life Like Diffusion by lutherjonna409
Analog Madness by CornmeisterNL
ICBINP - "I Can't Believe It's Not Photography" by residentchiefnz
Description
A custom VAE has been baked into the model.
This is the final version of the model. Wait for news about Realistic Vision based on SDXL.
Thanks to all who have used and are using my model, I appreciate you very much.
FAQ
Comments (115)
Hey, everybody! I mixed up the model and converted with the new name version 3.0 instead of version 4.0. Please read the comment from Civet_AI (https://civitai.com/models/4201?modelVersionId=105674&commentId=189126&modal=commentThread). I will fix my mistake now.
A big request to those who downloaded the full version of the model in .safetensors format! Re-download the model in this format as soon as it will be available for download, you downloaded version 3.0 by my mistake. Sorry for my mistake!
How about the VAE? is it correct?
@kanto The noVAE version will be replaced with a VAE version (in process). The noVAE version will be on the Hugging Face.
Thanks for correcting!
Just downloaded it, are the files updated to 4.0 now or shall I delete it and wait for the proper version? The file I downloaded has the filename realisticVisionV40_v40VAE
@jonasmillard741 Can you share a hash of this model? :)
@SG_161222 This? afcc6a9cac
@jonasmillard741 Yeah, that's the correct model.
@SG_161222 Thanks =)
i'll wait for vae vers but examples r awesome
Wait for it, coming soon :)
Is it worth separately uploading the noVAE version on Hugging Face instead of CivitAI?
Yeah I think it's fine, if someone wants the noVAE version you can link to Hugging Face so it is less cluttered here.
should i use the ft-mse-840000-ema-pruned VAE next to the model or it's already baked in the model (i'm using the full model by the way).
Hi. There is a custom VAE baked into the model, but you can use ft-mse-840000-ema-pruned VAE.
Read this, please.
The model has been completely reuploaded on CivitAI. The version with baked VAE has replaced the noVAE version. The noVAE version has been moved on Hugging Face, to download this version of the model follow this link [Hugging Face RV4.0 Repo]. Also there is a custom VAE file.
Sorry that this happened, but I fixed it :)
No need to apologise. Thank you for your efforts.
why Model file in civitai ~2GB, but in hugginface ~4GB , what is the best ?
@happynguyen91 not sure what you're talking about. This model is 4GB on both civitai and huggingface.
@happynguyen91 you might be looking at a 2GB pruned version vs a 4GB full version. personally i m not sure if the extra 2GB of full are worth it
So great! What file/model of Realistic Vision V4 would you recommend for traing lora?
Hi. I've never done LoRA training, but you may need to use the full version of the model located on Hugging Face, as it doesn't have VAE baked into it.
If I’m using this as a base to train a LoRA should I use the VAE or non-VAE version?
I have the same question (commented to bump it up the algorithm)
Hi. I've never done LoRA training, but maybe you need to use the full version of the model located on Hugging Face, this version without the VAE baked into it.
Also will the next update use SDXL or will you stick with 1.5
can I use it for commercial use?
Hi! Of course you can use the model for commercial purposes.
What does it mean "...fp16 version" from the menu on the right?
Its 2GB larger so I'm wondering what the difference is?
What benefits does it have over the regular 2GB one?
Hi! In the pruned version (2GB model), unnecessary weights have been removed. The model should be a bit faster, but the generation result should not change. But it is not suitable for training, as far as I know.
@SG_161222 I see four files. All have "fp16" in it.. the not pruned, bigger files also. I want to use the model to train a person with Dreambooth. What would be the best RealisticVision model version for that?
@N3rdlicht an "Full"-Version, not an "Pruned".
ok thanks
Can anyone please specify why different versions like inpaint etc. I understand VAE/noVAE. But why would i want a version (no inpaint) if they're the same size. Having inpaint is always handy. So why without it?
Hi! The inpainting version of the model is only needed for inpaint and outpaint actions, but it is better not to use it for image generation.
@SG_161222 is there going to be a noVAE release for 4.0?
@bl4ckfyr3 This version of the model has already been released, you can go to my Hugging Face profile in the model description :)
@SG_161222 do we install the VAE version in the VAE folder or SD folder with the other models?
@gxngxn Hi. The model needs to be placed in a folder with other models.
@SG_161222 Oh.. thanks for specifying
Can somebody explain what VAE is please?
VAE is like a color theme. :)
According to the description, version 3.0 has merged a lot of models, and how do you choose the Multiplier parameters of these models? Are there any metrics to evaluate how good a parameter is? And another question is, are so many models merged one by one, and is the order of merging important? Many thanks for your kindly explanation!
Why add CLIP SKIP: 1 of Usage Tips? I think the previous V3.0's Usage Tips is also CLIP SKIP: 1, using the last layer output of text encoder .Are there any difference between V3.0 and V4.0 in CLIP SKIP?
I'm using SD local and it is giving me an error saying :
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument index in method wrapper__index_select)
Why is this happening?
Restart a1111
Appreciate your model! I noticed on v3 (havnt tried v4, looking forward to it) that it frequently generated imagines with people who had more wrinkles than preferred. Any recomendations to limiting this?
Hi, there is one way in the model description, it is related to LoRA Add Details, read this, it may help.
Not gonna lie, I feel like this is the best model out there for realistic photographs. Well done 👍
Thank you very much :)
Can you upload it to HuggingFace please?
Hi! The model is already available on Hugging Face (https://huggingface.co/SG161222/Realistic_Vision_V4.0)
@SG_161222 Can’t pull. Says model library is unknown. Is it still in private? Would be fabulous if you could make it public.
When using Nickleback FS upscaler the image turns gray, any idea on how to fix that?
what are the downsides of using the inpainting version?
what is difference between V4 in Civitai and V4 in Hugging face ? i see they are not the same , 2GB and 4GB ?
Would also be interested to know the difference
Probably the encoder, safety checker and vae, or the precision.
v4 has a vae included? I always forget--in A1111 does that mean I need to turn the vae off that usually runs or is that not a problem?
If you set it to Automatic or None then it will not try to add the other VAE on top. I found when left on vae-ft-mse-840000-ema-pruned.ckpt it will darken images.
I usually get the non vae of models so I can choose but giving this one a test see if bake VAE is better.
I have also read but cannot confirm that if you select a vae in A1111 it overrides the models inbuilt one and does not add it on top. If that is true then you have the choice. But I don't know if that is true, anybody confirm?
@creeduk As far as I remember, In the AUTOMATIC1111 settings there is a checkbox that is responsible for how this function works.
By default, it usually uses the built-in VAE and does not overlay a separate one, but this can be disabled and then the external VAE will be used.
@Zmey_G That checkbox (I think) only makes it swap and use VAE that has the same name as checkpoint. If you have a checkpoint for a specific model it will use it, but baked in is different (as far as I know).
@creeduk It is quite possible that this is so. Unfortunately, I cannot confirm this 100%.
Information from some AUTOMATIC1111 setup guide, which I read at the very beginning of mastering web ui. It said that if you uncheck this box, then the selected VAE will be used, if you set the menu item 'Automatic', then the VAE with the same name as the checkpoint will be used.
I think it really needs to be checked. Unfortunately, the set of equipment that I have does not allow me to reliably state whether this is true or not...
@Zmey_G I found the VAE is on hugging face so I guess it would need both baked and non baked model, then some testing with none vs loading it then automatic to see what matches we would see how the built in VAE is treated vs the independent one. Maybe if I find some time :)
Just tried out v4 and somehow, it's giving me worst results than v1.3. With the same prompt, I was able to get much more quality results in v3. So personally, I'm sticking to v3 and don't plan on using v4, unless I'm doing something wrong.
same
RV3 is better for my prompts
I get very realistic faces with Realistic Vision, especially with a lot of the available lora and lycoris, but I notice that lora which modify bodies don't seem to work quite as well - either the model seems to ignore the lora, or they combine to negatively impact output quality. Do you have any tips for folks who want to exert more specific control over bodies, or is that not in the scope of this model? (NOT a criticism - it's perfect for so many use cases, I wouldn't expect any model to be universally applicable).
I have images that remain blocked, awaiting validation by a moderator, does this happen to you too?
Hi! Yes, it happens, but after a while they're published.
I've yet to find one that can replicate the star trek bridge. All seem to just make a "bridge" as in that piece of road.
Can you upload the 4.0 no vae please
Hi! 4.0 (noVAE) is on Hugging Face.
lol why am I getting so many DUDES in my 1woman prompts??
Hello, I would like to ask, what is the difference between V4.0 (VAE) and V4.0-inpainting (VAE), is there any change with or without inpainting, which version is the best. I don't quite understand what it means to be a model with inpainting
An inpainting model is more suitable if you want to modify a specific portion of an image - say you want to change a table or remove someone in the background of an image, you would "inpaint" the image and replace it to whatever your prompt is.
good job
Maybe is the best photorealistic model I ever used!! Thanks for sharing!!
Thank you! :)
Realistic Vision V5.0?
Hey, everybody! I have something new. What do you think about this <example images>? Yes, I said that 4.0 is the final version, but in my spare time I tried to improve the model anyway. (This post is not related to SDXL)
I like it! Especially how it keeps the eyes brown, I hate when they get changed to blue or green when I have weighted words like (brown eyes) or (black eyes)
5.0 Looks more real and the skin is more detailed.
Massive improvement in 5.0, it actually looks "real" as opposed to "realistic"
That looks fantastic indeed. 👍👏
These examples look fantastic! As others were saying, the skin, faces, light and expression look more life-like and less artificial.
Wow! That looks so good.
Well, this looks definitely better, at least in those samples you've provided.
Interestingly Fanning looks least like her in that example so I wonder what other famous models will look like.
Anyway - definitely waiting for RV5 now :)
please upload it
When you guess it must be the maximum now, the dude came up easily with a v5.0, absolutely fantastic. I would be very pleased about.
Already waiting for that real improved version. Amazing work as always ✔ ❤
V5.0 def has better skin detail for sure... looks great
@malcolmrey Maybe I should have used a different negative prompt. And I'm not that great at using LoRA. Is it possible to contact you on Discord or some other way?
@Stable_Confuzion @anniebunnie @lol1fkkkr @sintica102 @drawaline @luisillo22 @malcolmrey @pieke @alice_dee @SpeedyYT @ByteCann Thank you all so much for your feedback. The model will be ready for release soon.
What do I think?
I think you are some God that came from the Latent Space heavens so that you can provide us users with greatness..
I think they look amazing!
This is insane. The leap from 4 to 5 feels like the transition from 1.3 to 2.0., if not bigger. The skin shading and subsurface level of detail is crazy for something out of SD1.5, looks more like cherry-picked MidJourney pictures.
Please also keep an eye on male subjects. Most models are only optimized for females and do a crappy job of generating males.
just adding to the list, i'd love to be able to try v5. example images are incredible!
@Stable_Confuzion @anniebunnie @lol1fkkkr @sintica102 @drawaline @luisillo22 @malcolmrey @pieke @alice_dee @SpeedyYT @ByteCann @digitaleyesightment @aslansbane @Noise_Connoisseur @Uneternal @aihotgirls https://civitai.com/models/4201/realistic-vision-v40?commentId=203541&modal=commentThread
When I use img2img and the person I try to copy or improve or whatever has a closed mouth my images still have either open mouths or half open mouths or even closed mouths with teeth showing "through".....like 1 out of 10 or 15 has a closed mouth then. What is this with the teeth even if the input image doesnt show any kind of teeth? Must be a prompt issue. What would you guys suggest?
Love it. Nice to see a model remixing mine sitting on the top most popular :)
Oh, hey! Good to see you here. Thank you for the great model :)
Realistic Vision V5.0 on Hugging Face and <new example images>
I'm opening up access to the model on Hugging Face. Note that this is the version without VAE. Use with this model vae-ft-mse-840000-ema-pruned. The version with baked VAE will be uploaded to CivitAI soon.
Realistic Vision V5.0 Hugging Face Link
Awesome! Thanks so much 👏👏👏
Thank you! ❤️ Real Vision SDXL coming anytime soon?
@anniebunnie Hi. Yes, I am planning an SDXL based RV but I can't give a specific timeline.
@SG_161222 Awww Sweet! I'm really excited hearing this. I like your models a lot and I create LoRa of myself to use it with your Model. Lot's of Love. Happy to hear,
@anniebunnie :)
Details
Files
realisticVisionV60B1_v40VAE.safetensors
Mirrors
realisticVisionV40_v40VAE.safetensors
realisticVisionV60B1_v40VAE.safetensors
realisticVisionV40_v40VAE.safetensors
realisticvision.safetensors
realisticVisionV40_v40VAE.safetensors
realisticVisionV51_v40VAE.safetensors
realisticVisionV40_v40VAE.safetensors
realisticVisionV40_v40VAE.safetensors
realisticVisionV40_v40VAE.safetensors
realisticVisionV40_v40VAE.safetensors
realisticVisionV40_v40VAE.safetensors
realisticVisionV40_v40VAE.safetensors
realisticVisionV60B1_v40VAE.safetensors
realisticVisionV40_v40VAE.safetensors
realisticVisionV40_v40VAE.safetensors
realisticVisionV40_v40VAE.safetensors