What makes a good input photo
AI effects can only work with what's in the photo. A few tips that consistently improve results:
- Use a clear, in-focus image with even lighting. Avoid heavy shadows on faces.
- For portraits, use a photo where the subject's face is unobstructed and roughly centered.
- Higher-resolution inputs give better outputs. Anything from 1024px on the long edge usually works well.
- Avoid heavily filtered or already-edited photos — the AI works best on clean source images.
Supported files
- JPEG, PNG, and WebP images.
- Most effects accept files up to ~20 MB.
- Animated GIFs are not supported as input — use a still frame instead.
Why a generation might fail
Generations can fail for a few reasons: the upload was corrupted, the AI model couldn't find the expected subject (a face, an object) in the image, or the input was flagged by our content safety filter.
If it fails for a system reason, your tokens are refunded automatically and you can retry. If it's flagged for content, the tokens are still refunded but you'll need to choose a different photo.
What the progress bar means
While your image is generating, you'll usually see a thin progress bar under the "Generating..." text. It's an estimate based on how long this same effect has taken on previous runs — not a guarantee. Some runs are faster, some slower.
If the bar reaches the end before your image is ready, it just stays there with an "Almost done..." note until the result actually arrives. And for a brand-new effect (or an unusual setting we haven't seen many times before), you'll see the spinner with no bar at all until enough data has accumulated.