Feedback

4k

I'm not sure what went wrong but I just trained a model on very clear, detailed pieces of art and the outputs when the model is selected are simply jumbled blobs of color. No subject nothing just looks like the initial noise from a single step or something. Why is this happening? Is it broken?

I really enjoy using the canvas to stylize and reimagine images, but it leaves something to be desired. Simple quality of life improvements like positive and negative prompts not disappearing every time you close the generator window or switch tabs, or the ability to dial or input exactly the amount of magnification you'd prefer for the canvas itself, as opposed to presets like 10%, 25%, 50% etc., and the ability to move about the canvas without having to constantly switch between the select tool and the pan tool in the tool bar would be greatly appreciated. I mentioned I like the stylize generation features, but I'm confused about why trained models appear to have no influence on the output of the generation. Now, the set of models provided by your company, like your Krea model, Luna, Ghibli, etc. do have a dramatic impact on the output, but it's almost without exception of poor quality. Perhaps that's due to my prompting, but I've never seen a good picture result from using them. The community models, on the other hand, all seem to have decent outputs, but it clearly makes no difference about which model I use. My suspicion is they all have Stable Diffusion XL as their base, because the results are very similar. But if the purpose of those tools is to "stylize" images, wouldn't you expect different models to have a greater effect on the result of the image? One more suited to your preferences? Instead the resulting image is almost entirely due to the prompt and nothing else, making switching between the many models pointless. If it is possible, could you add a dial similar to the text to image generator where we can decide how much influence the model has upon the output? This would reduce the dependency on the prompt itself to generate a specific style, and free up the valuable prompt space. In fact, a wider variety of settings and dials in that domain would be great. Being able to control the number of steps, the clarity, and sharpness of the image would be helpful. What I'd also like to see is a more robust suite of tools to modify and enhance the images in specific ways. The "enhancer" is a fine tool for its specific purpose, but it leans almost entirely toward an uncanny photo-realism, regardless of the settings or prompts you employ, and for anyone preferring a more artistic styled image, its almost worthless. Now, I don't know much about how things work on the back end, but what I'd like to see is the option to enhance your stylized image in the style you prefer without a significant loss in clarity or quality. The opposite, in fact. Upscaling an image in the Canvas just makes a larger version of the flawed original, which is not ideal. To that point, it would be great if the option to set the resolution of the output in Canvas was a built in feature. It's clear that limit of pixels on the image likewise limits the detail and quality of the output, so being able to decide that beforehand might resolve that issue. In addition to this, I'm actually confused as to why the website is compartmentalized the way it is. In-painting, (as clumsy and simplistic as the brushes are at the moment) is only a feature in Live generation, patch enhancement is only available in Enhancement, and stylize is sequestered to the Canvas. Would it not be preferable to unify these tools into a one stop shop program so all features are ready and available at your fingertips? Apart they seem more like kitschy novelties rather than a solid program whereby one can decidedly craft an image, with the added labor of saving and uploading images time and again and heading to different tabs to see how they might fare in producing the result you desire. I understand that would be a big undertaking, and pardon my ignorance if any of these suggestions are silly or unfeasible, but being able to stylize an image, in-paint (either live or in generations), patch enhance, and then upscale and enhance all in one place seems like the obvious direction for your software. Anyway, thank you for your consideration. Apologies for my being so long winded, it never helps in my prompts.

been looking for the delete project feature but I couldn't find where it is, not sure if it's missing or my eyes are just too exhausted :O

Draw a mask, add a prompt and generate the content from the prompt in the masked area.

im trying to generate images but it stays loading and nothing happens

To be able to see how many of our 50 (which is weird as in your pop up it says 60) free generations we have sued and when we will be able to access more

after hitting stylize button, I could see the second counting down for image generation, but after that nothing appears. I tried restarting my browser and used a different one, this problem was still no solved. please fix this bug, thank you!

Hey there, KREA devs! Hope you're having a wonderful evening! There was a feature back in the old canvas where besides text prompting, you could also supplement (multiple base) images to influence the results. I'd honestly love to see the return of that feature in the near future. I loved chucking in different noises and ISO grains to dither and make the results more cinematic and retro-realistic so to speak. I'd love to hear your thoughts and if it's on the roadmap! Keep up the great work guys, you're honestly the most hardworking and reachable devs out there! Much love. Ke

i'd like to be able to delete ai models I've created through training. give me a button for that.

Up to 2 images at once with 2,3,4 to a maximum of 5 faces per image.

Feedback from a user interview: "Inpaint" didn't actually let me inpaint. It just tried to remove an object. I would change the functionality or change the name.

It would be nice if you can add more images and traing again the model instead of making a new one.

For example if there was ability to save/restore 3 negative prompts. neg prompt 1 for realistic portraits, neg prompt 2 for landscapes, neg prompt 3 for illustration. instead of copy/paste from txt file (or even typing from scratch), it would be nice to be able to click on let's say button 1,2,3 in the negative prompts section to be able to quickly restore one of the 3 different negative prompts depending on what type of picture you were creating

If you could implement aspect ratios like 16:9 9:16 and the rest of the set. This would greatly help in you tube shorts, video productions, Instagram sharing, Pinterest and other platforms. But maybe you already are working on it???