Crazy Frog AI Photo Guide: Choose Better Reference Images
By CrazyFrogAI · Updated · Based on the current tool and published examples
Quick answer
Choose one clear, well-lit photo of each person. For a Crazy Frog AI chase, upload the lead first and the chaser second; for an invisible motorcycle, upload one rider. Keep faces unobstructed, avoid crowded group shots and use only photos you have permission to process.
What makes a useful reference?
Start with a visible face, ordinary perspective and enough detail to distinguish the person. A waist-up or full-body image can provide clothing cues. Strong beauty filters, tiny faces, dramatic blur and hands covering the face reduce useful information. A higher-resolution photo cannot guarantee a better video, but a severely compressed crop gives the model less to work with.
One person per slot
In the two-person tool, photo one is the foreground lead and photo two is the chaser. Choose separate references even if the people are together in your original photograph. Check the preview and slot label after upload. If you remove and replace a photo, confirm you have changed the intended role before generating.
Faces and clothes can compete
A face-only crop supplies little outfit information. A distant full-body picture may show the outfit clearly but leave the face too small. Choose a balanced reference, then describe one important clothing detail in the prompt. Avoid asking for both exact clothing preservation and a completely different outfit in the same instruction.
File preparation and limits
The browser accepts JPG, PNG and WebP and prepares images before submission. The server validates processed JPG/PNG references, up to 3 MB each. Chase mode requires exactly two photos; solo ride requires one. If an upload cannot be prepared, export a standard JPG or PNG and try adding it again. Choosing a file only shows a local preview; confirming generation sends the references to the video provider.
Examples you can inspect
Each of our four example pages shows the two AI-created reference portraits and the resulting video. These are illustrative synthetic people, not customer testimonials. Compare the face, hair and clothing with the output, and notice where motion or perspective changes their appearance.
Before pressing Generate
- Confirm both roles are in the right slots.
- Make sure you have permission to process each photo.
- Choose a simple motion and one camera direction.
- Review the displayed cost; replacing a photo does not make a new paid generation free.
If the output has the right scene but an unstable face, change one variable at a time. Start with clearer references before adding a longer prompt. See the troubleshooting guide for technical versus creative failures.