You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I've encountered an issue while generating images with FLUX. The entire canvas of the image exhibits a visible 'grid-like' artifact, which looks like a subtle but consistent pattern of intersecting horizontal and vertical lines. This pattern becomes most noticeable in darker areas, such as shadows, fog, or smoke. On lower-resolution images (e.g., 1 MP), the artifact is less prominent, but when I upscale to 4 MP or use FLUX tools like Depth or Canny Control, the grid becomes very pronounced, even at 1000x1000 resolution. It resembles a fine mesh or screen overlaying the image. Is this a known issue, and how can it be resolved?
I get the exact same artifact when using flux checkpoints in forge. Same behavior too. It doesn't really show up until using hires fix and upscaling it, but I don't think it's related to upscaling... I can see the grid pattern in the early denoising steps.
I've tried using every upscaler, sampler, and scheduler. I've tried a range of sampling steps, denoising strengths during upscaling, a range of hires steps, and cfgs.
Once I see it, I can't unsee it, and it ruins every image.
I've encountered an issue while generating images with FLUX. The entire canvas of the image exhibits a visible 'grid-like' artifact, which looks like a subtle but consistent pattern of intersecting horizontal and vertical lines. This pattern becomes most noticeable in darker areas, such as shadows, fog, or smoke. On lower-resolution images (e.g., 1 MP), the artifact is less prominent, but when I upscale to 4 MP or use FLUX tools like Depth or Canny Control, the grid becomes very pronounced, even at 1000x1000 resolution. It resembles a fine mesh or screen overlaying the image. Is this a known issue, and how can it be resolved?
fluxtools-lora-depth-demo.json
The text was updated successfully, but these errors were encountered: