I would like to know what keywords make a difference in the advanced text box for Redefine.
Yes effect.
I’ve noticed that if I write “woman”, then it will make all people in the photo women.
I saw “Realistic photo” and “realistic skin details” so I use those a lot. Making cartoons into flesh.
–
No effect
I tried “impressionism” and “artistic” and “colorize”, and “slim” but those do nothing I think.
What keywords effect the image?
How can I change so not all three women in one photo get the same face?
Is there a way to colorize black and white photos?
–
I have a i9 12900 K with RTX 3060 and normal time to redefine one 1000pixel image 4x size is 30-60 minutes. What is normal?
I came here for this! There don’t seem to be many of us trying to use this creatively. So far I haven’t had any luck with any prompt I’ve tried. I’m trying to get it to take photos to a more artistic or cartoon effect.
Utilize the text prompt if you want to be extremely specific about the details you are looking for. The model responds more to a descriptive statement versus a directive one. For example, use the phrase "girl with red hair and blue eyes " instead of "change the girl’s hair to red and make her eyes blue ".
So clearly describe the subject, environment, lighting, colors, and mood:
“A serene mountain lake at sunset, surrounded by pine trees, with soft golden light reflecting on the water.”
Under the Generative AI models, select “Redefine”.
Move the mouse cursor to the Preview to select an area to click on to render a small preview.
To have more thorough control of the edits, select “Advanced settings”. You must turn on Advanced settings and set the Creativity level to 2-6 in order to access the prompt input box.
Yes it’s one of my worst niggles - especially when I’ve been investigating differences between short and long prompts. I am getting towards an MSc Project if not a PhD Thesis in my Notepad++ files because I save -
my own long and short text prompts
at least one long and one short prompt from a free “text description of an image” site
a ChatGBT prompt
a Google.Gemini prompt
I like rene.zuch ideas on text prompts
I am hopeful we can get something better than the current state of things - we need to increase our requests for improved usability as opposed to glitzy funnies.
I missed this post although I thought last month I had gone back thru everything posted on Gigapixel since v7.0.0
You have probably moved on and discovered how to get better responses from Prompting, but in case you are still looking here are some things I’ve discovered in the last 6 months.
I’ve done quite a lot of testing using long and short text prompts and different permutations of Creativity and Texture and image size.
I offer some general hints at the end but taking your points in turn: -
if you try “woman in white dress at lower left” and also include “retain detail of men in group at centre” it may help
in artistic prompts I find detail helps ie “pre-Raphaelite vivid colors and Paisley patterns” or “in the style of a Monet impressionist painting” or “18th century maritime seascape and costumes”
there don’t seem to be specific keywords that always effect things the way you might expect. I think most people find that direct commands like “make a person slim” will not work but maybe “emphasize a fashion model catwalk pose for the female on the left” might work
sometimes a negative prompt will work " blah blah blah. -no changes to foreground vegetation original detail"
at the moment I don’t think there is an effective way to maintain individual faces other than to limit Creativity to either realistic None or Subtle OR artistic Low or Med which locks out more overall creativity (fantasy) in the image.
if you have Photoshop skills you can paste the faces from a Low creativity render into a 2nd higher creativity render and merge/blend them for the best of both worlds.
there are many free and paid for apps to colorize a b&w photo. Photoshop has a Neural filter which does quite a good job. Try Google searching for what might work best for you.
I’ve notice that some people seem to manage to say change the color of a dress or jacket but that skill still eludes me ! There are lots of apps that will do that - color change or in-painting
regarding your hardware the process time does seem rather slow - lots of possible reasons
what RAM do you have and how much VRAM in your RTX3060? I think you must have at least 8GB for redefine to work at all but in real world use 12GB is regarded as necessary - 16GB would be great!
Look at Task Manager and Resource Monitor when you are doing a 4x render and see how much memory your CPU and GPU (and Disk) are using.
Where are your image files compared to your Program Files?
My system is Windows 10 with an older but good i7 7700K CPU & 64GB ram with Gigabyte RTX4070 Super GPU with 12GB. On a 1000px x4 redefine artistic Creativity 1-4 Texture 1-5 it takes from 30secs to maybe 3min depending on the detail in the original and the settings used.
A general tip I would highly recommend is to use online AI to generate a text description of your original image. You can tweak the answer as you feel necessary - they just get you started …
Free Image to Prompt Generator is very easy to use. ChatGBT and Google Gemini are also good.
Just drop or upload your starting image and then ask for - very detailed text description of image
Longer descriptions don’t ALWAYS work better than short ! Try the default offered and then you can add - limit to 100 words - to see if a shorter description works just as well. There isn’t much difference in processing time for smaller images but when you get to x6 upscaling an original 1000px image with lots of detail and Creativity a longer description may have a noticeably longer processing time.
You can experiment with different model settings to find what combination of settings works best for whatever style you are wanting to achieve.
Once you do find optimum settings it’s still sometimes very surprising what a completely different choice of settings may produce. Say you find redefine artistic Low Texture 3 works well for you. Just once in a while check out using Recover and or redefine realistic None and Subtle as well as High and Max at Texture 5. Sometimes the result will really amaze you and produce an unexpected golden egg surprise.
A final thing to remember is that a Cloud render will produce a noticeably different final result to Local rendering on your computer. Cloud rendering uses completely different models and interprets Text Prompts using a vastly superior AI knowledge database.
Of course you need Cloud Credits and they can be expensive if you get carried away.
Happy to respond if you have any questions and I’d like to know how you have progressed since Dec 2024.
Best of luck