Join the Woollyfern Patreon (free & paid options): www.patreon.com/Woollyfern Topaz Labs Free Demo Download (affiliate link): geni.us/kzlG Midjourney Guides: woollyfern.gumroad.com ➡many are free and paid ones are included with some Patreon tiers
I think in a lot of cases the overall lighting is better for photographic images of people in 6.1, but yea, there's over-beautification much of the time.
I found that the “beauty bias” extended to my cref model that I use constantly. Her freckles disappeared and her eyes have more glamorous makeup. I retreat to v6 when I use her for a more natural look. By the way, that’s both in photos and my preferred illustrations.
The prompt to text is slightly improved. It can't handle long phrases. Actually Leonardo ai Phoenix is superior. So far, I have not seen anything in V6.1 to brag about, but I'm confident in the long-term development.
Can someone help me with ref image. I try to put my picture if a piece of furniture into generated by MJ image but it changes the furniture. I want to be the exact same furniture but it changes it. Any suguestions how the prompt should look like?
Midjourney uses reference images as inspiration. It will create something new from them. You can get a piece of furniture similar to your image, but Midjourney is not set up to replicate it exactly.
Thanks for these tips! I'll be applying them in my work tonight. @woollyferncreative Question at 04:58 , I didn't know a new Midjourney version changes the personalization 'id' P Code number. If so, can this be a way to reset errors? Like we just wait for the new version and run the personalization Task/Rank Images page all over to fine tune our style preference (mine is cinematic + photorealistic for my projects) from the previous Midjourney version? Thanks
For V6.1 they implemented an entire new model/algorithm for how they use our image pair ranking data to generate personalization codes, which is different from their personalization model for V6.0. Each new --p code since the V6.1 release is a reflection of all image pairs you ranked up to and including that date. Think of it as a snapshot in time in your image ranking journey. Your codes are not reset. Midjourney just has a new approach for trying to generate more accurate personalization codes. Hopefully that made sense!
@@woollyferncreative After I wrote the above question for you I went ahead and ran the the personalization Task/Rank Images page all over again but I only selected the images that looked like a real life-like photo, from food to people, from animals to cityscapes and one sci fi. I also selected realistic colors and moved away from neons images and those with sparkles. Then I used a prompt I ran in 6.1 last week to do some cover drafts for a writer. Same exact prompt in 6.1 and just now I got totally different results that resembled a still from a movie. So it works! 😀 BTW the prompt I used, if u want to test it, was for a pirate theme cover tests last week's came with bright intense tones, after rerunning today I got muted, very realistic colors and yes, the ID changed: Here's the prompt I'll erase it later- after I get the image I 2X zoom out and take it to Photoshop for fixes: a handsome athletic English cyber soldier pirate with dark hair in a science fiction azure skies, dark blue and dark black, in the style of Michael Garmash, realistic movie still, dark black and beige, soft-focused hyperrealism cinematic colors golden ratio, ambient occlusion, high key photography, arabesque sci-fi, towers, sci-fi futuristic minarets, walled sci-fi city, fortress gate, cinematic light, insanely detailed and intricate melancholy, Photorealistic, ultra quality, fashion style, Studio Lighting, rtx, shot on 70mm, in the style of john collier --ar 2:3
@@woollyferncreative Last message before I get back to 'work' doing AI art for three art-packed short story anthologies. The 6.1 revised new ID is doing more detail after a new personalization run than the 6.0. In that prompt one of the models has a sleeveless leather vest and only when I opened in browser and zoomed in notice the improvements. The hems are not just extremely well made but show irregular wear-and-tear in sections you don't see unless you zoom in: the leather vest has sections facing the front to detail wear and tear more, like weapons pockets, than sections where a person would barely touch or mangle if wearing a vest like the upper shoulder fabric. And the light on the vest matches the same source of light reflected on face and arms in same tones. And the best tiny detail, the exposed top of the arm under the sleeveless vest has a slight color variation, less tan, than the rest of the arm exposed to sunlight, just like any person going outside under the sun for hours and arms are tanner than skin under the shirt. I've never seen such miniscule attention to these color details before, and I only selected less than 100 images with photorealism under the personalization Task/Rank Images to override the 1,100 plus from the 6.0 personalization. So, it doesn't take much to get a new clear slate id. Bye and looking forward your next tutorial!
@@deanm9652 they haven't set a timeline for it, but it's currently in training. V6.2 will be the next release. If I had to guess, V7 is probably at least a couple months away.
Thank you for your video! They are all so clear and easy to follow! But 1 question please, where I can find the Seed number? I cant find it on midjourney or discord either ):
Thanks! To find the seed number for images you have already generated, open your image on the Midjourney website, click the 3 lines in the upper right, then Copy, then Seed. This will copy the seed number to your clipboard. Note: this will not work if you open an upscaled image -- you need to open the original image from the image grid. You can use that seed number in future prompts by typing --seed a space and then your seed number. If you are wanting to compare results between v6.0 and v6.1, you can use any seed number you want. It's just a random number. I typically use --seed 1111 when I'm doing prompt comparison.
Yes, SREF numbers can be discovered by adding --sref random to your prompts. Check out this video for more info: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-ndou30IMmQA.html. I also have several videos where I share some of my favorite SREF codes, for example: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-RhhgukQGWu8.html. Thanks for the comment!
MJ still overly exaggerates describing people. Anyone with a little extra baby fat is described as obese. It even uses phrases like fat ugly and obese which is extreme for many use cases for describing especially when you try to use the prompt and it says it goes against their standards