in term of realism, this model and epiCRealism are the best to produce realistic images. But in term of training LoRA, I found difficulties with these models. I prefer Beautiful Realistic Asian V7 as the checkpoint to train my LoRA to generate realistic person images. Despite being so powerful in generating such realistic images, these two models tend to generate "European" faces. I trained my LoRA once in this checkpoint but it turns out, that almost all faces resemble the "European features" although my model reference is an Asian person. The second difficulty is the prompt. Yes, it has been simplified to produce quality images. But I can't produce the same result with these models as I generated the images from the BRAV V7 checkpoint. Maybe there has to be a different approach of writing prompts in these models. I need some time to tweak it.
Hi, thank you , I will try this model. Can you do a comparison video between Automatic1111 and Forge webui in terms of speed and quality? My simple comparison with an RTX 3060 and 12GB RAM yielded great results: with a 20-step generation, Automatic1111 took 23 seconds for one generation, whereas Forge webui dropped to 15 seconds, which is great!"
I have the same card and that's in the same area for what I get with SDXL and ADetailer. SD 1.5 models are closer to 10secs with ADetailer. You didn't say which model type you were using but yeah, the performance increases are great, it just needs to sort some other bugs/incompatibilites out and I'll switch from A1111 to Forge full time.
EpicRealism and Epic Photogasm have several different versions such as X, Y, Z and Last Unicorn. Is there much difference between them? And which one did you use in the video?
Hey, if you go on the models page and click about this version, it could contain some information. I'm not sure myself as I haven't used those versions.
I believe they're just tweaked or upgraded versions depending on what you want and it's age. I agree that it's not exactly clear from the names but from the ones I've tried, the difference is negligible in terms of quality, although the generations are different. I ran exactly the same prompt and seed in Last Unicorn and PureEvolutionV5. Last Unicorn gave a close up portrait, while PureEvolutionV5 gave a waist up portrait.
Love it when people analyse prompting techniques. Great vid!
That was the best checkpoint overview I've ever seen, absolutely perfect.
Realy like this one, would love to see more!
in term of realism, this model and epiCRealism are the best to produce realistic images. But in term of training LoRA, I found difficulties with these models. I prefer Beautiful Realistic Asian V7 as the checkpoint to train my LoRA to generate realistic person images.
Despite being so powerful in generating such realistic images, these two models tend to generate "European" faces. I trained my LoRA once in this checkpoint but it turns out, that almost all faces resemble the "European features" although my model reference is an Asian person.
The second difficulty is the prompt. Yes, it has been simplified to produce quality images. But I can't produce the same result with these models as I generated the images from the BRAV V7 checkpoint. Maybe there has to be a different approach of writing prompts in these models. I need some time to tweak it.
i love these
Thanks!!
What is the UI being used for these tests?
Automatic1111
@@BitesizedGenius Thanks!
👋
it is really silly some company wanted to censor ethnicity in photorealism generation, ethnicity is fact and unique!
Hi, thank you , I will try this model. Can you do a comparison video between Automatic1111 and Forge webui in terms of speed and quality? My simple comparison with an RTX 3060 and 12GB RAM yielded great results: with a 20-step generation, Automatic1111 took 23 seconds for one generation, whereas Forge webui dropped to 15 seconds, which is great!"
I was thinking about reviewing forge, but i'm running a 4080 which is overkill for anything so i'll put it on my list :3
Thank you, i read from some forums people who had 6 and 8 GB VRAM had huge speed bumb@@BitesizedGenius
I have the same card and that's in the same area for what I get with SDXL and ADetailer. SD 1.5 models are closer to 10secs with ADetailer. You didn't say which model type you were using but yeah, the performance increases are great, it just needs to sort some other bugs/incompatibilites out and I'll switch from A1111 to Forge full time.
This SD 1.5 checkpoint is one of my top choice for generating realistic pictures.
as a noob i feel like i got the best results with this checkpoint, been using it for a month, so glad to see you cover it
EpicRealism and Epic Photogasm have several different versions such as X, Y, Z and Last Unicorn. Is there much difference between them? And which one did you use in the video?
I see Last Unicorn but my question still is about the difference between the versions
Hey, if you go on the models page and click about this version, it could contain some information. I'm not sure myself as I haven't used those versions.
I believe they're just tweaked or upgraded versions depending on what you want and it's age. I agree that it's not exactly clear from the names but from the ones I've tried, the difference is negligible in terms of quality, although the generations are different. I ran exactly the same prompt and seed in Last Unicorn and PureEvolutionV5. Last Unicorn gave a close up portrait, while PureEvolutionV5 gave a waist up portrait.
Thanks for the review. EpicPhotoGasm is a good Realistic photos checkpoint.
awesome. I really loved your review format. this was useful for me even though I have already used this model for months.
That was a solid checkpoint tutorial man.
can you make a tutorial on XY PLOT ?
th-cam.com/video/EEpyGVCLWnM/w-d-xo.html
@@BitesizedGenius thanks so much
I cant tell difference between realism and ai anymore after t his 😮
can you do latest leosam helloworld sdxl?
This is a nice Checkpoint
Agreed, it's a good model. I still prefer RealVis for SDXL (which just released a version 4) but good alternatives are useful.
@@Elwaves2925 what do u think about epicrealism sdxl?
@@rangorts I don't actually have that one, thanks for the reminder that I need to get it.
Your tutorials are so good
very nice and informativ review !
Thanks for this, what about hands or fingers? Are the results better?
Good idea to include that and yes the hands are pretty good imo!
@@BitesizedGenius thank you for replying too fast and for the info!
EpicRealism is better
Nah
Actually, those are from the same person who produced the models.