Option to Refresh Images with Same Seed or Generate with New Seeds

[fractaltr33]

It seems that each new generation is based on the same seed, this makes the picture very similar.

You print promt and press generate
It turns out a picture with seed 324 for example
Enter a new prompt
And the countdown of seeds all the same starts from number 324

I would recommend that you use a new seed every time you generate.

Well, as an option, it would be nice to add the ability to select a seed number / random.

Thanks.

[Tracy.H]

Thank  you for the feedback. This is actually a feature of Firefly. When you choose a different Style or other parameter, the images will keep the image similar to the original. Without the same seed, you will generate a brand new image if you intend to only try a different Style. You can use the back arrow to go back and start a new session. Your suggestion to have an option to select whether you want to keep it or not is an interesting idea. Do you think this would be confusing to users who are new to Gen AI?

[The Scott Valentine]

It's not clear how seeds are treated when refreshing. Having the ability to retain the seed values enables refinement and development of ideas across sessions, while generating a new seed that is far away from the current one allows more exploration.

Please include an option to manage the seeds for generating images, including the ability to: save the seed value and reuse it later; generate a new seed that is closer to or further away from the current seed; enter the seed value (or upload in some way).

[The Scott Valentine]

I don't think it would be confusing if the UX included some descriptions along the way - perhaps a toggle like in some of the apps that enables/disables context popups or help overlays?

[Kamesenlin]

A veces cuando se genera imágenes genera la misma incluso si el prompt es diferente pero con algunas cosas en común. Ignora el resto del contenido.

[杨29363534p0s1]

When I'm trying to use the txt2img function, it seems that after pressing the 'refresh' button, the new 4 images are sometimes related to the 4 old images, as I saw image with similar structures at the same position of the 4.

I believe it would be useful if there are options with this. What I expect is:

•  To generate totally unrelated image, like using txt2img with sd-webui locally.
•  To generate 4 new image based on one of the previous image, like midjourney.
•  Setting how related the new image is with the old ones.

Other suggestions are:

• An option to generate text effect more like a mask on a normal image, so that the effect is not so twisted.
• A dictionary or wiki to look for the tags.
• A chatbot which can better turn natual language into prompts.

[Tracy.H]

@杨29363534p0s1 Merging your thread here so we can track the demand for certain features. I have also noted your additional suggestions separately.

[PhilipThonbo]

Remember to always keep a seed specially with the font styles. Because there is a major industry trying to translate the style of titles to other languages in books, tv, movie, cinema, and other places. By then it would be nice to have a seed that was the exact recipe to recreate. And way way too often we dont get much original footage. so if there was a way to encode it into the file. Or else i dont think we would ever get the seed.

[Matthew241430003xuj]

Hi, Philip!
I'm sorry, but I don't understand your idea.

What do you mean by "a seed"? Is it metaphorical or a technical term?

[Tracy.H]

We are working on determining what data needs to be saved with your image. The seed is definitely part of the "recipe". Right now, you can find the seed in the URL.

[OzanE]

Is there any way that we can control the seed of the generative fill? I want to complete parts of a video and I need more or less the same feeling/structure in the generative-filled sections. Any Ideas?

[Petri31713965gpuf]

The seed (integer) which was used to create the image should be visible in the UI.

User should be able to either use random seed (-1) or use an existing (manually set) or old seed.

This would allow consistent set of images e.g. same character in different environments, wearing different clothes and so on.

[Professor M Cokkinos]

Actually, I am trying to generate a series of images for an animation sequence so this works

[Julie_Lee]

Does the generative AI in Adobe Firefly not have a seed number?

I want to create multiple images in a continuous and consistent style, but it's very inconvenient since it only produces images in different styles each time. Is it not possible to do this with Firefly?

[droopydog500]

It does have a seed number, but it is not exposed in the user interface like it is in Night Cafe, for instance. The best way to accomplish what you are trying to do is copy the link for the photo you want to start with and go back to that image using that link as your starting point again.  Since this is not part of the user interface or published features, there are no guarantees.

[Highness]

I would recommend downloading the image and then use the 'upload reference image" feature to copy the style.  As for using the same character, Firefly doesn't contain that feature at the moment.  You can read and support this idea here: 

P: Maintaining a scene or character across different prompts 

[JoshSommers]

Suggest adding the ability to optionally randomize the seed with each generation.

[Jill_C]

What do you mean by "randomize the seed"?

[Prettysleepy]

Firefly is great, but I'd love to have a new seed button to regenerate the scene.

[Sergio33427231aqx9]

It would be nice to be able to choose the seed in each Generation with AI. That is, to be able to see and edit the seed used in each generation to be able to create content more accurately.

[Vivek12]

Hi, Thank you for your valuable feedback. Our product team is aware of this feature request, and they will take appropriate action based on its feasibility. Meanwhile, please upvote this idea to get it prioritized correctly. 

^VS

[Kartika Rawat]

Thank you for sharing your valuable feedback! We truly appreciate it and will relay it to our product team. If you have any further questions or need assistance, please feel free to reach out. We're here to help!

[stacy_4568]

Comment demander le seed d'une image sur Firefly ?

[droopydog500]

Hello @stacy_4568,

The seed is no longer visible and can no longer be changed in Firefly Text to Image. The seed is visible and can be changed in Text to Video. In text to video, it is at the bottom of the left hand panel:

My best,

    droopy

[caio_4244]

Hello, I hope this message finds you well.

I'm working with Firefly Web and Photoshop on my Mac. I would like to test Firefly's text-to-image feature using the same seed to achieve similar results. How can I capture the seed value from a generation? How can I add this seed to a new generation to get a similar result? Is this process possible only via API, or can it be done in Firefly Web and Photoshop on Mac?

[Oh.N8]

Hey @caio_4244 
I m not exactly sure what you are asking. 
However, I am going to make a suggestion. 
1) After you generate, copy the link and post it into a Google Doc or Word Doc. Having the links allows you to save a future generation.
- you keep the seed.
2) copy that link into another tab change the prompt accordingly.

Keeping a copy of the link is important. You have the seed. when you start later, you start fresh. I suggest copying 2 or 3 images and links. 

 

https://firefly.adobe.com/public/t2i?id=urn%3Aaaid%3Asc%3AEU%3A395c95d7-8492-4132-9ee1-3a91a0e5027d&...

Cheers,
O.Nate