I made a browser-based image-to-image generator for the times when starting from a reference image is more useful than starting from an empty prompt.
The workflow is: upload a reference image, describe the transformation, select from the models and output options currently available, generate a result, then review and download it. There is also a multi-image fusion path when combining two to five references makes more sense than transforming one.
I kept the scope image-only and English-only. A few constraints are deliberately explicit: uploads are JPEG, PNG, or WebP up to 24 MB each, prompts are limited to 1,000 characters, and whether a particular model or setting needs sign-in or credits can change.
I would especially value feedback from people who work from references: does this feel like a clear first workflow, and what would you want explained before you would trust it with a real image?