Global language support for Generative Fill and Generative Expand in Photoshop (beta)

[MeredithStotzner]

Creativity speaks your language! Global language support for Generative Fill and Generative Expand in Photoshop (beta)

We are excited to announce that Firefly-powered features in Photoshop (beta) - Generative Fill and Generative Expand – are now testing support of text prompts in over 100 languages.
Simply enter the desired prompt in one of 103 languages, then click Generate.

What languages are supported?

What if my prompt isn’t accurately translated?

We currently support prompts in over 100 languages using machine translation to English provided by Microsoft Translator. Because of the nuances of each language, it’s possible certain generations based on translated prompts may be inaccurate or unexpected. We are working hard to identify and resolve any issues. To report incorrect translation results, click on the Poor Result flag in either the Properties Panel or the Task Bar for the specific variation. 
 
Click the Tell Us More to indicate the result didn't match the prompt.


What if my prompt says it’s an unsupported language?

• Please provide feedback on the language services in this public forum thread.

Current known issues (Adobe Photoshop Version: 25.0.0 20230725.m.2254)

• Short (1-3 word) non-English prompts may result in an error with Generative Fill/Expand.
  Workaround: Make your prompt a little longer
• Prompts requiring Enter key for conversion in the middle (I­­nput-method languages such as Japanese, Korean, Chinese etc.) unintentionally create a generated image.
  Workaround: Copy/paste your prompt into the field. 

[AxelMatt]

@MeredithStotzner Can this be posted as an Announcement for better visibility? Or can this be pinned?

[CShubert]

Hi @AxelMatt it is pinned and set as a feature post.

[vdthanh]

Been using the new AI generative feature for a while, I have some feedbacks that I wish you guys could have a look into that but don't know where to post so I comment here:

1.  Generative fill doesn't work if I choose a small area to apply fill. That will result in standard/guideline violation even though I leave the command bar blank (I don't tell the app any specific thing I'd like it to generate)
2. Option to let the app generate more than default 3 as there are such time like hard script input - the app is less likely to generate what I want
3. Option to let user keep a specific detail between the options (I may not describe it clearly but it's something like blending the results - eg you generate a vase of flowers and you want the flowers of option 1 and the vase of option 2). Of course it can be manually done but I still wish this for easier use
4.  Option of the input like neural filters: whether the generation is in a new layer (w or w/o layer mask) or current layer.
5. Option to generate high resolution fill, though it would take a lot more time but still OK

[Elisabeth31241255blk1]

I miss creating text, for example: Write text "Hello world" with blue color merged styles.

[GABRIEL31368664lswz]

What happened? This looks like the worst version to me.
Now 90% of the time I use Generative Ai, it doesn't work. I will have to downgrade if possible.
- Every time now it's Something Wrong Error 408 or 406.
- Service is not responding, check internet
- Any details to remove(black prompt) - images violate user guidelines.

[ApplianceServiceTech1]

I experienced the same thing. 

[luisam85494213]

The generative fill refuses to follow the prompts imputed. When trying to remove people from a street photo, it just kept replacing the selection with other people or elements (like trashcans, dogs, concrete blocks, children). Prompts tested:
- remove woman

- remove person
- remove element
- fill with backgound
- fill with backgound street
- sidewalk
- empty sidewalk
- fill with sidewalk
- fill with empty sidewalk
and so on.... On all attempts the element selected was replaced with another thing, not removed

[classbrain]

Just a hint. Don't say "fill" or "remove". Just type what you want to see.
It doesn't appear to have been trained on those sorts of prompts.
Weirdly, I've found that it does best when no prompt is given many times.
It does do different types of fills based one the shape and overlap of the
selection. Also if it's being stubborn about replacing one person for
another. I've found I can do a rough paint out of part of the person and
then make a selection, hopefully with a slight overlap of what I want to
see, and then it does a surprisingly good job. The shape and overlap of the
selection does appear to be key.

[rickk62939765 · Accepted solution]

Doesn't work any more - it seems you have to explicitly use the remove tool as blank prompts will often lead to a scene-appropriate object being filled instead.

[classbrain]

I use blank prompts all of the time. A lot of the success depends on how you select the space. Also, if you are trying to remove an object and the blank prompt doesn't work, just. give it a nudge in the right direction, like "wall" or "gradient" or whatever is a simple direction toward what you need. Do play around with the shape of the selection. It makes a huge difference in how it generates.

[rickk62939765]

The word 'Tapir' is in English yet Photoshop always says its in a language it doesn't understand in generative fill.

Result: No tapirs 😞

[what now toons]

I hate this new generative fill!!!  I left the other version to get away from it.  If I lay down a color blend, and decide I want to darken the top a little, or blend another color, it just wants to erase the previous and start from scratch.  NOT what I want!  I'm a professional, I've used Photoshop for decades, I don't have time to relearn how to use it.  How can I go back to two days ago when it worked the way it always had?!!!