Real Dream Website (Faster download)
Available at TensorArt: https://tensor.art/models/899250342552672189/
Comfyui Workflows:
Anima text2image with Loras Workflow
Anima img2img (denoise) with Loras Workflow
Klein 9B Edit with Loras Workflow
Klein 9B img2img (denoise) with Loras Workflow
Z-image-turbo img2img workflow
Z-image text2img with loras workflow
Z-image De-Turbo with loras workflow
This model is capable of generating very realistic images that can confuse people's critical sense, please do not use it irresponsibly, harming ordinary people or public figures. Also avoid its use in electoral campaigns, promoting or damaging the image of candidates. Although the field of artificial intelligence is poorly regulated, its improper use can cause legal penalties even in countries with little regulation in the area. Be very careful to not commit crimes in your country and not to promote injustices against innocent people. I do not authorize the use of this model for Lora training using the face of real people without authorization from the biological owner. Do not create images that can embarrass and damage the public image of people. This can cause serious psychological harm, including deep depression. I do not authorize the use of images generated by this model for scam applications and the dissemination of false information.
During the fine-tuning of AI models, I only use synthetic images that do not infringe any copyright, personal image rights, or laws. The models are merges of open models from different sources and may contain diluted data remnants in the neural network. I am not responsible for possible data remnants since they are diluted, and it is not humanly possible to track the existence and origin of this diluted data.
I do not allow the upload of any of my models on Tensor Art. I only do not allow it there because the policy of duplicate models on Tensor Art prevents me from uploading again if someone has already done it.
I have enabled to automatically hide all NSFW posts for several reasons that are not worth listing. I hope you have patience and understand.
Description
This version is the final version after 20 beta versions. This version features more realistic face textures, more consistent scenes and more realistic contrast. Unfortunately it has lost some of the Flux style that it had in version 11, to return to the Flux style it would be necessary to train the model again and this would not necessarily improve the model.
FAQ
Comments (64)
What's new in v12?
This version is the final version after 20 beta versions. This version features more realistic face textures, more consistent scenes and more realistic contrast. Unfortunately it has lost some of the Flux style that it had in version 11, to return to the Flux style it would be necessary to train the model again and this would not necessarily improve the model.
Thanks to Patreon supporters and Buzz donors for helping with the development of version 12.
Any chance of realising v12 to your Patreon's?
@Voltaire77 It's version 11.20 but I'll re-release it on Patreon.
I've no life, so I download ALL pony-based models and create an internal competition.
for "realism" tests the number of points was multiplied by 3. The main parameters were the "creativity" of the model, (ability to create complex unusual scenes), and the quality of the picture.
As you may have guessed, the 11th model was in first place in both tests. So I highly recommend supporting the author.
Could you give more details on how you benchmark models? This is the most relevant topic when I'm developing a model.
This is a long process, and there can be a lot of text. I'll try to be brief.
1) I installed ComfiUI (well, this is in addition to 4090 :)) and made several presets. 768x1152 / 896x1280 / and 5 others.
2) I selected optimal parameters (GFXscle etc) for each model - they are different, they cannot be compared directly
3) I prepared prompts (there could be a whole article here) 6 prompts with people, and 4 others. A very important point, which I checked is the ability to do 2 or more people, how much the quality of the scene drops. The second very important - how EXACTLY the conditions of the prompt are met.
All models cope well with portraits, we all know this, so the portrait is a bad benchmark, but still.
I know the models' weak points - for example, different people in one frame "old man" and "guy", or non-standard poses - not just a photo "for a passport", so I call it "flexibility" of the model.
I separately assessed creativity, but these are more like bonus points, since this can usually be implemented with a prompt.
Penalty points for b/w (your model has 0, although sometimes I came across them, but not in the test), or for very bad drawing of faces.
8 models that have updates for 3 months, 10 presets with 7 resolutions with 7 images each, and voila )
In second place, if you're interested, is "ponyRealism v2.1"
There were models who made very successful individual pictures, but overall they were much inferior (goddessOfRealism, tameThePony), plus I had to make my own flow for them.
@sinatra Sorry, I misspoke when I said "in both tests". I meant in the two most important criteria - adherence to the promt (flexibility of the model) and average picture quality.
Since I did this research for myself, I did not bother to rate each picture, but just glanced at them, so despite the correct approach, I may be subjective.
In fact, most pony models are quite similar in quality, and although yours stands out noticeably, I do not think I could tell which picture was made in which model without notes.
A little later I will add some pictures from the tests
@kostyg31337 Could you explain to me why ponyRealism v2.1 would be so good? From my point of view it was quite relevant when it was released but it seems outdated now.
@sinatra and who would you put in second place without testing?
I gave a lot of points for realism and accuracy, for me these are the most important criteria for a pony model.
Perhaps, I just thought about it now, this is due to the fact that I worked a lot with pony 2.1 and subconsciously made the prompts suitable for her.
yeah no, PonyRealism pales in comparison to Real Dream, especially if we count versions. even Real Dream v3 is better than PonyRealism, imho.
I don't normally give so much buzz, but I have spent hours on your other models. So you earned it buddy!
The download link for version 12 is also available in the free section of my Patreon. If you like the models I develop, I recommend subscribing to my Patreon for at least the free subscription, from time to time I make beta versions available for free.
Just take my 1500 buzz! It's not like it's my life savings of buzz or anything! 😭😭😱
目前最好的真实模型
V12 is fire you absolutely knocked it out of the park probably my favorite pony tweak yet.
Genuinely incredible. Breathtaking realism.
Great Job, I sent 500 Buzz to support your development.
V12: When two or more characters are drawn, the model has trouble rendering eyes in all shot types except the closeup.
To fix use refiner function and upscaling. Refiner: first 30% Pony Realism Model, Last 70% RealDream. hi-res fix: latent space with noise add back of 0.65. Final resolution: 1024x1024
The download button is now working again and is also available on the Real Dream website's download links
It is not, as of 13:26 GMT.
@sendmefreestuffa8246 Use Real Dream website link.
Yeah, I can confirm that it is still not working as of 18:26 (UTC+2), possibly someone coded time-zones backwards on the time-lock or something... 🤷♂️
For a second, a link pops up for paying for membership on civitai.green, and then it just redirects to civitai.com instead. Seems like a bug on this ever-more-complicated-and-broken site... 😩
Either way, thank you so much for providing external links ❤️
Still not working. Prob should ask for Civitai again. Thanks anyway.
@threeRthreeDe The download button on Civitai has now finally started working for me 👍
12 is definitively a step up in realism from 11. Its a Pony and as such has inherently a problem with drawing distant faces. One can use a third party face restoration model but it has to be said that the aging face restoration models aren't up to newer standards also they actually mess up better non-pony models so these have become a bit unwieldy. Where Pony shines is poses and its far superior to standard SDXL in that regard. Its very comparable to standard SDXL quality but SDXL has still a few exceptional models such as Art Universe 4.0 and they still have their justification of existence. This is one of the best realistic Pony variants as far as i can tell, maybe even the best and as such highly recommended. There are other good Pony models of course such as 2dn which are not as realistic but more a choice for sharpness and style. Very good work indeed.
rdxl11 has become one of a handful of "reliable" tools, if other checkpoints don't deliver what I am aiming at :)
I'm curious how rdxl12 does :)
Many thanks generally though! <3
It really is one of the best versions I'm testing... it was worth every BUZZ invested to have it.
Real Dream V12 is my favorite realistic Pony model to date. It somehow manages to retain all the benefits of Pony while also converting things into a realistic style that doesn't make characters look off-putting. Plus it's very good at hands - much better than most other models I've tried - they usually just need a slight touchup with an upscaler/hiresfix. Amazing job on this one!
the eyes are weird unless you upscale/add detail, the eyes are considerably worse than V11 imo, one is always bigger than the other and its almost like its lazy eye,
If you use a1111 or a fork of it, use the adetailer plugin.
the eyes are completely normal if you use pony prompts and not "HD, ultra HD, Nikon image" I don't know why but they are
This is the first comment I have ever made on this site and it had to be to commend you on your amazing work. This is the best model I have ever used. The best are RDXL > The Araminta Project and then Juggernaut for me in my opinion. I am still very new to all this but your model blows my mind. Thank you.
This is incredible!
amazing high quality, love it
ok, im a little suspicious of the people who are going, "its incredible" like this is amazing. That being said its great, i can finally put in "large breasts" and get attractive large breasts, not over sized ugly breasts. So thank you!
It's best if you use pony prompts rather than the ones on his images. I had to learn about score_9 and score_8 up and stuff but I've not used anything else since this came out
its good!
I absolutely love it.
Great model! It works much better than other realistic models for using homemade trained loras of characters. Thanks a lot!
I was wondering if you had tips for improving the rendering ever further, such as specific tags that works well with this model?
For example, should I include the tag realistic? or add 2D or anime content as a negative prompt?
In the sample images you uploaded, you did not include the usual score_9, score_8_up, are they necessary with this model?
this is better than pony realism v22 which was my main checkpoint, bravo
Same here :)
Real Dream SDXL Pony is the best realism checkpoint in XL. I's very good for nsfw and sfw. These days this is my #1 pick to generate an image.
good!
Useful
waiting for lightning version 4 step for rdxl 12
Spooder gorl
purdy nice
So far have had the best results, missing loras and embeddings etc...
Like it, works well with UltimateUpscale
used it quite a lot over the last few days really like it a lot
IMHO - Hands down #1 Realistic Pony model. I've tried more than I care to admit. No notes.
Can you explain, how is this realistic at all? All sample and user pics are semireal at best. Try another models and you will be suprised.
@vijizu164 any suggestions?
@SleezeBagDiffusion https://civitai.com/models/820629/ms-ponyreal I've tried many 'realistic' models too and this one has surprised me very pleasantly. Give it a try, don't judge it only by the test images on the model page.
@SleezeBagDiffusion most balanced model is the newest cyberrealistic pony. It means its has much better realistic quality, than this checkpoint, while keeps compability with character loras, celebs. That is very important to me. There are checkpoint, which surpass everything in realism(Humana, YAPRM, TAMe, Uncanny valley(korean bias), Real viwes, pornmaster, but i recommend these, if you only want generate 1girl, because they are not compatible with lora girls. These models trained heavily on realism, and lost therefore the pony roots.
So my only good advice at this time is cyber 6.5. The creator has a history to slowly and steadily manages to create better and better models. I would stick to that, and wait for next releases, which will be improved again and again.
Edit: if you will use cyber, use 2m Karras. And add random img number to prompt, like IMG.6521.JPG, use walking on street or dancing in nightclub, it helps a lot in realism.
@zipazuqu141 Hmm interesting, I've tried the Cyber Realistic Pony ( I primarily use LORA's for specific characters) and haven't like it as much as the popularity would suggest. I'll give it a try again soon!
amazing work
Any guide for it? Like recommend tags for the model etc?
https://danbooru.donmai.us/wiki_pages/tag_groups
But this is a censored checkpoint. If you want hardcore, pick another one.
New models in beta version are available at my Patreon
One of the best models out, Excellent hands, feet and body positions. Never needed to resort to a LoRA unlike many other models. Prompting is a fantastic experience - like using HAL vs an Atari 2600. As Sinatra notes the models is slowly evolving away from Pony (specifically the rating tags). Earlier versions were also much better with booru tags in general with newer version being much less consistent. Would be very helpful if it could count like other Pony/XL models, e.g. one person and 1girl do not work most of the time. Some diversity issues, e.g. other models are more aware of the differences between a Hawaiian, Amazon and African jungle. Also hard Asian bias (could be all Pony have this issue). Use the ethnicity you want in the positive prompt and add Asian to the negative prompt to get more consistent results.
Can confirm: the model works very well!
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.