Prepare photos for a three-person AI scene
Choose clear adult portraits, assign roles and avoid common input problems.
Make permission the first step
Use your own photos or photos you have permission to use. Each image in the current Gas recipe should show a different adult. Do not use a stranger's portrait simply because it is available online. Synthetic adult portraits can be useful for technical tests, but those tests do not establish how well real customer photos will work.
Choose a readable face
A sharp face under even light gives the workflow more useful visual information than a distant or heavily edited face. Avoid sunglasses, masks, strong beauty filters and extreme shadows. Crop away unrelated people without cutting off the main subject's face. Keep an original copy so you can choose a different crop later.
Keep the roles unambiguous
Prepare three separate files and label them for yourself: dancer, selfie walker and person by the car. Check each thumbnail before submission. The current mode expects three photos; a collage or group photo is not an equivalent substitute. One-photo and two-photo modes remain unavailable.
Check the live requirements
Upload limits and accepted formats are shown by the live interface. Follow those limits rather than a screenshot of an older version. Public generation is currently on the waitlist, so wait for access before uploading. Clear photos reduce avoidable input problems but cannot guarantee likeness or a successful result.
See the Gas workflow guide for the current recipe and its limitations.
Visit the Gas effect page for current availability and the waitlist.