Zeiz

joined 1 week ago
[–] [email protected] 2 points 1 week ago

Hello! Unfortunately, it doesn't recognize it. Seed is nothing more than a random number that AI consults to make calculations. It's like a constant. Change the model, change the way of making calculations and the old seeds no longer give the same results. If it is a realistic photo, you can use the website https://imageprompt.org/image-to-prompt to extract a detailed and suitable prompt to use in Flux, the new model. Then save the new seed.

[–] [email protected] 2 points 1 week ago (1 children)

In this case, you have 3 options:

  1. Without using seeds, generate batches of images several times with the same prompt until you find the face you are looking for;
  2. Change the prompt slightly to include "mixture of Mexican and Italian" or something like that; or
  3. put the name of a famous person in the prompt.

I always try to choose option 2. Both the current and previous models always gave my characters a unique touch. I'm still learning a lot, and this new model has also brought me frustration. However, for those who like to find patterns to write prompts, the previous model was extremely bad and unpredictable. The current model has greatly reduced this problem and it was then easier to understand the AI's behavior.

[–] [email protected] 2 points 1 week ago* (last edited 1 week ago) (4 children)

Calm down, you're rushing and probably don't know how to use the tool. Here's how I did it:

  1. I converted your image into a prompt using the website: https://imageprompt.org/image-to-prompt
  2. I took the prompt and put it in perchange;
  3. I defined the casual style;
  4. I chose an image to copy the new prompt and seed;
  5. I generated variations of the same seed, slightly changing the prompt;

Will it be perfect? Of course not. The images would still need to go to the "workbench" as has always been done. But the quality is really much better with the new model. You can generate several times until you find the facial features closest to what you want. You might spend about 20 minutes generating images repeatedly, but once you find the seed, you can manipulate it however you want.

Prompt + seed: A casual photo of Close-up portrait of a woman with long, dark hair. Latina, 25-35 years old. She is wearing a cream-colored, lace-trimmed top. The top has a sweetheart neckline that accentuates her upper body. A silver chain with a dark, rectangular pendant hangs around her neck. Her expression is serious and focused, with direct eye contact. Her gaze is slightly backwards, suggesting contemplation or introspection. Her lips are smiling, and her facial features are softly lit, highlighting her natural beauty. Soft, natural lighting creates subtle shadows and highlights across her face and upper body. The background is a dark, muted brown tone, providing a stark contrast to her light-colored top. The image's perspective is directly in front of the subject, creating a feeling of intimacy and personal connection. The composition is centered, focusing the viewer's attention solely on her. The overall mood is introspective and thoughtful, with a classic realism style.. It's a casual photo. (seed:::172525303)

[–] [email protected] 2 points 1 week ago

I was a little bummed, too. But give the new template a try—it’s actually pretty good. There are still a few things that could be improved. The preset styles look a little weird and out of whack. But seriously, give it a try. Take a look at these drawings I made—they’re starting to come close to a standard. I didn’t even have to throw away the old prompts, I just tweaked them a bit.

One tip: set "No style" under Art Style and define your own style manually.

[–] [email protected] 2 points 1 week ago* (last edited 1 week ago)

What’s been working for me is manually setting the style in the prompt and selecting "No style" under Art Style. I still need to experiment a bit more to get better control, but it feels like I’m on the right track. One thing I noticed right away is that this new model produces way less “garbage” than the previous one. Before, I had to generate a bunch of image batches just to find one I actually liked.

[–] [email protected] 3 points 1 week ago (1 children)

At first, I was pretty upset about the AI model change. I had already set a workflow for my project... But honestly, this new model does seem better. I’ll keep learning so I can get as close as possible to what I had before. I’m sure in a few months things will feel normal again and everyone will get the hang of this new tool.