AI does not produce an identical result every time: even with the same photo and instruction, results can vary. This is different from a failure that prevents the image from being generated.
The following resources contain examples and comparisons from earlier models; they do not describe every option or limitation of current models:
Prompt examples for earlier models
Framing study using earlier models
- If no image was generated or an error appeared, note the displayed message and describe which stage stopped when asking the AI Assistant for help. A generation failure should not be explained simply as normal AI variation.
- If an image was generated but looks different from what you expected, review the instruction: describe what you want to transform and what should be preserved. Use the prompt examples as a reference, rather than rules for every model.
- Check the framing of the input photo. Full-body, half-body and close-up photos give the model different information and can change the result. The framing study illustrates these differences in earlier models; check the result with the model you plan to use.
- Do not expect identical copies when generating again. AI is not deterministic; details can vary between runs, even without changing the photo or instruction.
- A prompt that works for one person may not work for two or more, and the reverse can also happen. Test each participant count you plan to offer separately.
- Test in conditions close to the event: the same camera, lighting and framing. An approved sample made in different conditions does not guarantee the same result at the event.
- After adjusting the instruction, generate new samples and check the results for each participant count. Only offer options at the event that you have tested and approved.