Bug Reports

HHello, i think you have a big problem with smile persons generation, please fix it

While the photorealism models work great they are sometimes unreliable. In some instances the images are very realistic, at other times they create cartoonish images. This has happened with both the RecraftV3Raw and Photorealism image generation models.

The drag-in is cool, however when image is on a 45 degrees wall like this, the perspective is often slightly wrong, and there's no way to do a free transform mesh (like adjust one of the four corners) to match it. Also sometimes there are some rough edges at the border of faces, and getting more annoying when there is some blur object in front of it (like a plant). Might just need time to optimize the algorithm, the experience is still amazing.

now it seems my mouse sticks to the image and doesn't let me do anything else

the app doesn't work

But the word I'm using is "Fart". Last time I checked, that isn't profanity

I've tried following the video on expanding an image and it was really awkward on a laptop using a touch pad to select the outer boundaries. When I finally managed I ran the prompt and the results were unusable they were so bad. Nothing like what I saw in the video. I have attached screenshots of my original image in the frame. I exported this and brought it back into the canvas following the video instructions. I selected the outer edges where I wanted the AI to fill them in and you can see from the screenshots from my history the selection and the results I got. Also working in the canvas I selected the outer borders, or so I thought, the results were really terrible, including one totally white blank image, and you can see from the screenshots from my history that the entire image had been selected. This was not what I was trying to do. I'm not sure how it happened, but this shows that it is really awkward trying to select these outer borders with a bit of the image included. I think it would be reasonable to assume that if we create a frame, put an image inside it and size the image down leaving empty borders of frame around it that we want to expand the image into those empty surrounds using the same style and continuing and expanding the image in a logical way. So why not eliminate the need to select anything altogether? Of course if we do want to select and area to add something with a new prompt that the user will use the selection tool to do that. It's obvious which the user wants by their actions leading up to clicking the generate button.

I sent a refferal link (invintation for earnning 200 credits) they registered, but i did nit get any credits.

In Linocut V3 99% of the cases the image fits the whole image size and it "Touches" the border. Almost always the images fit all the image size, resulting in some borders being cut, specially on the top border. This results in a flat unnatural line at that part of the image. See the examples attached. This was not happening in previous version 20B.

I can type with the keyboard for the mobile app, but it does not correct typos that are shown.