Virtual try-on takes two inputs: a photo of a person and a photo of a garment. The model identifies your body, its pose, and its proportions, then reconstructs the clothing onto that body so the fabric falls the way it would in reality. Finally it relights the garment to match the light in your original photo.
That two-input design explains most of what makes results good or bad. The model can only work with what it can see. If your arms are pressed against your sides, it has to guess where your torso ends. If you are wearing a thick hoodie, it has to guess your shape underneath. Guesses are where realism breaks down.
It also means your lighting gets reproduced, including the unflattering parts. A hard shadow across your chest or a strong yellow bulb overhead will be baked into the result and applied to the new garment. A photo taken facing a window on an overcast day will almost always outperform one taken under a ceiling light.
The practical upshot: one clean full-body photo, shot straight on at chest height, against a plain background, wearing something fitted, will produce better results than any amount of adjusting settings afterwards. Take that photo once and reuse it.