Playground v2.5 & Creative Upscale Level Up Your Creative Game!

Announcing Playground v2.5, our latest text-to-image generative model, available to the public and research community today. Playground v2.5 is the state-of-the-art open-source model in aesthetic quality, with a particular focus on enhanced color and contrast, improved generation for multi-aspect ratios, and improved human-centric fine detail. We have also published a technical report for this model, and you can also find it on HuggingFace. huggingface.co...
With this release, we intentionally chose to push the bounds of our current SDXL architecture. While we are excited to explore new architectures in future models, our goal was to quickly deliver improvements to our users (Playground v2 was published in December 2023).
We've also added a new "creative upscale" feature that will enhance your images even more.
Note: Free 5 Creative upscales/mth.
Pro: 100 Creative upscales/mth
Turbo 250 Creative upscales/mth
Thank you to our Playground users for creating so many amazing images
Find us on
Playground AI - ai image generator: playground.com
X (Formerly Twitter) https: // playground_ai

[@playgroundai]

Hello folks! The creative upscaler is now available on Canvas! Have fun!

[@LiJenTu211]

Thats super useful for some situations,thx for your team's effort

[@playgroundai]

Have you tried Playground v2.5 and the new Creative Upscale? Let me know in the comments!

[@OnyxLee]

I tried it in A1111, it doesn't work. 😅

[@playgroundai]

@@OnyxLee A1111 support to come, currently there is support for Comfyui.

[@sad2804]

Filters polaroid style and analog diffusion deleted :( 👎

[@vVinchi]

Intro is so epic!🔥

[@playgroundai]

Appreciate it bro! Your space cat looks awesome! 😁👊🏼

[@kagtisleri2418]

Features are very good. A seamless(tile) option for patterns would be much nicer. Can this feature be added?

[@aegisgfx]

Cant use the service anymore, it always says I am too far down the queue or something. Whats the point of advertising a service no one can use?

[@rogerc7960]

Try one image at a time, or during quiet times ie. usa nightime.

[@AnutaNSK]

Thanks! Your videos help me a lot in my work!🌼

[@playgroundai]

You're welcome 😊

[@loupottetoe6239]

Hooraay Playground, congratz 👏🎉

[@playgroundai]

Thank you! Appreciate the support!

[@viktorgencur7019]

go go power playground !

[@playgroundai]

Thank you! 🙌

[@147adam]

the free version is kind of unusuable now

[@johnK18068]

I like the v2.5. it cuts the amount of time needed to adjust the image.

[@playgroundai]

Definitely! We're working on making our users workflows easier

[@survivalmedly943]

thank you so much,very useful on me for editing inspirational qoutes everyday.

[@playgroundai]

You're very welcome! That's awesome!

[@nks-nqn-xs4yd]

ya probé Playground v2.5, y el Creative Upscale me dejó escalar 5 imágenes en la cuenta gratuita, los resultados son muy buenos, los felicito por eso, siempre se superan, aunque yo no puedo usarlo mucho, con esta versión mi cpu va a explotar jaja, tengo que cambiar de procesador (tengo 2 núcleos), cuando logre mejorar mi ordenador seguramente lo utilizaré, gracias

[@michiganmovieweekendermmwn1941]

When did Playground AI change to 100 FREE! prompts per day? And now they are only generating two variations at a time? When did this happen? We were only alerted to this this weekend.

[@playgroundai]

It was a recent change about 4 days ago, we announced it on the community tab here, as well as our Discord server. The website is also up to date with the new info. You can still generate 4 images at a time on the free tier. kzbin.info/door/UbLzJ30GLE1F5OSWNoYs9gcommunity

[@ShoMorphias]

2.5 seemed alright when I was using inpainting mask to preserve a face. However, SDXL seemed to be a little better on the clothing designs and body position where I was getting more 'yeah maybe this' reactions. I used creative upscale but in my case it didn't pan out as it turns purple eyes to blue. Most things AI like to do that, I suppose. Perhaps some control options like Krea has would help. I noticed I could finagle it with a negative prompt and such to preserve the eyes I wanted. Playground's normal upscale and face restoration (especially) seems to be hit or miss, and mostly miss. But I have salvaged things I wanted before. Btw, Playground's canvas is way more intuitive that Leonardo's. Wish the inpainting mask didn't drop/reset every time it's used though. Not sure what that's about. The mask layer retains in the normal mask edit interface.

[@inknoidrobot1227]

Where do you think I created this cyberpunk dystopian avatar 😂. I was wondering if they were going to upgrade.

[@playgroundai]

Sick! 👊

[@GryphonDes]

Very excited to use Creative Upscale when it reaches Canvas at least -- Board still makes me twitchy LOL - even after all this time!

[@playgroundai]

Really? We often hear so much love for board 😁 but I hear ya being a major canvas user. 👍🏼

[@GryphonDes]

@@playgroundaiI cannot explain why it is this way but I started with Canvas, and Board just makes me feel so claustrophobic LOL

[@liangliang2268]

Hi,I tried to use this model with layer diffuse in comfy ui, the result image didn't have the gray color background as usual, and of course the RGBA decoding was failed. I've tried many other sdxl models, unfortunately Playground is the only one with this issue.

[@JosTemer]

I've been on PlaygroundAI for over a month now...what the hell happened!!! The free version is terrible now and peak time 24/7? I was this close to go pro! But like I've been reading below, I prefer that realistic vision over that realism engine! Why can't it be availabe in the free version????

[@captainpike3490]

I wish Playgroung 2.5 Can work on Fooocus webui or Forge webui, Only available for ComfyUI

[@playgroundai]

Yeah still not available as there is a particular step that was do for this model that most SD platforms don't have but we hear the will be support soon.

[@HiProfileAI]

Why won't it work on Fooocus? I haven't tried it yet. But got it working on ComfyUI

[@marlonalvez9502]

VERY GOOD!

[@playgroundai]

We appreciate it! Enjoy!

[@ATLJB86]

Nice

[@playgroundai]

👍😀

[@equilibrioemocionaldafe0312]

Gostei 👍

[@RafałMalczewski-e6v]

I have a question if your website in the free version generates random images that are, in some sense, similar to what I describe as a prompt. in the free version I am unable to do anything useful using the mask function. How can I believe that if I pay you money for the full version or as you call it pro version, anything will improve or work differently. If I do not have the opportunity to check the image generation using the mask before I make a purchase, even in a low resolution image

[@danielc121]

I was so excited about it, specially in photorealism,but it was disappointed to see that always makes the eyes ,hands and mouth horrible :(,and it tends to make like 3d and not realistic.

[@playgroundai]

In those cases you may have to try different prompts to get those results. Generally you should be able to prompt the same way however it is a different model so it may take some discovering at first.

[@danielc121]

​@@playgroundai,thanks tho,I can see that can't make famous characters like xl but it would be cool to have more videos about 2.5 to explore others aspects of it

[@playgroundai]

It does lack a bit in that department and wasn't really a focus to be honest. but yes more vids to come on v2.5

[@HiProfileAI]

Yes I've had some issues with the mouth and teeth

[@押久保衣潤]

Image generation on top of the generated completed image, in which case the image is merged with the completed image. A huge photo collage-like image is completed. Is there a way to split the image?

[@ppovoa1]

i tried playground v2.5, but not the creative upscale, i will try that

[@Gabriecielo]

Nice demo! Is there any special setting or workflow changes needed for Playground 2.5 in ComfyUI? I tried couple basic or advanced workflow, always get a noise picture. Using fp16 of playground 2.5.

[@playgroundai]

Make sure you have the latest update to comfyui and use this workflow, save as .json then drag and drop. There is and EDM node that is required. gist.github.com/comfyanonymous/6275d1df67f402dc053e3d6991ebe201 This model works best with low CFG 2-4, 30-50 steps, DPM++ 2M Karras although other samplers can work, it's always personal preference with samplers. Hope that helps.

[@Gabriecielo]

Thanks a lot, I was missing EDM part, working and stunning quality@@playgroundai

[@playgroundai]

You're welcome! Enjoy, more to come soon!

[@emanuelec2704]

@@playgroundai how do i fix this edm error? Prompt outputs failed validation ModelSamplingContinuousEDM: - Value not in list: sampling: 'edm_playground_v2.5' not in ['v_prediction', 'eps'] EDIT: Fixed, I just had to update ComfyUI and restart the server. I love it. It struggles with longer prompts, but for short prompts it's stunning.

[@frontdoor333]

when is play ground going to introduce easier face swapping, better stretch to draw ie Like Leonardo feature

[@playgroundai]

Definitely looking at some way to do face swapping in the near future. 👍

[@kemikemi756]

Cool

[@playgroundai]

Very cool indeed! 😁

[@Arthuur9]

Can it generate text?

[@jnm2148]

Doesn't Playground AI have 9:16 aspect ratio? Or is it only for the Pro version?

[@playgroundai]

Yes on the pro version

[@Parth-g5q]

Playground ai safe or not ? Please know.

[@LouisGedo]

👋

[@playgroundai]

Hey hey my man!

[@coskunuyumazer6505]

Language to interface. Support please! Turkish

[@playgroundai]

You can use expand prompt to translate it to english. kzbin.info/www/bejne/poWQq4mog7uXgKcsi=wCeU39nkIDxeY7V7

[@justcd123]

I unfortunately found the creation of face portraits extremely disappointing. When you enter for example a teen beauty pageant winner, no matter what else you instruct, the nose and lips will always look the same. It seems like it is standardized on the look and even is you add "thin lips and small sharp nose" it will totally disregard the instruction and continue giving you twin looking girls with no changes as requested. Even if you add a country of origin, the lips, and nose stays standard.... Very disappointing.

[@playgroundai]

Give it some time, it's a different model and you will find some of your usual prompts will need different words or structure. I've had some images where I had to spend some time finding the right prompts and also this model does benefit from a bit more negative prompts vs SDXL I find. There is more to come on this....sit tight.

[@justcd123]

@@playgroundaiWill gladly do. Routing for the site. Think it is great, so please no not take as negativity on my side. Just in need of photorealism and variation, and at this stage the 2.5 is a clear good move, but it still has some distance to catch up with Leonardo, which I deem to be your greatest competitor outside of the paid only options. With the bid I played with it so far, it seems like your prompt understanding is however superior and the prompt expand option is magical. Thank you!

[@playgroundai]

Not at all my friend, we always welcome constructive criticism. Believe me I do see it's short comings however we've come a long way from PGv1! We're running our own race here and it will be worth it.

[@malarkey102]

Can you do NSFW images

[@HiProfileAI]

Yes, I've tried it, and and yes it s a different model. All the women and men look the same regardless of race. Also prompting for body shape is problematic. Also it gives error can't produce for certain words all the time. I've tried to add prompt info for differences. It gives real good photorealism and detail. However it has its shortcomings. Shorter prompts are better as well. Also I downloaded the model and it gives different results on local vs. online for the same prompt, and same settings. A work in progress, overall a nice detailed model and looking forward to it. I'm interested in rune tuned versions but it has no commercial use licence.

[@nicksonredfield3925]

well, please release webui version asap, I really need it

[@playgroundai]

Are you using A1111? I'll see if I can push the team on it however I think it's up to the developers of the platform. Comfyui support for it currently.

[@nicksonredfield3925]

yes I am using a1111, I tried comfy UI,but it doesnt have function I need, I am using LLM to control SD, only A1111 support thatthank you if you could give a little help on that. @@playgroundai