I’ve been testing Sora and wanted to share some feedback after trying to create a bedtime routine diagram in Spanish for my toddler.
What works well:
- I really appreciate that Sora offers multiple image options (usually four) with each prompt. That gave me a better chance of finding something usable, even when the results weren’t perfect. This feature is helpful and sets it apart from other tools like ChatGPT.
What needs improvement:
- Context memory: Sora doesn’t seem to retain context from previous prompts, making it hard to refine or revise an image based on earlier results.
- Structure consistency: Steps were often cut off, skipped, combined, or repeated, which made it unreliable for visual sequences.
- Mismatched visuals: The illustrations didn’t always align with the text. For example, brushing teeth might be accompanied by an image of a child clasping their hands together with no toothbrush in sight, or brushing teeth while also reading a book.
- Grammar issues: Spanish phrases were sometimes nonsensical — for example, “elegir los dientes” (choose the teeth) instead of “cepillar los dientes” (brush your teeth), or “decir dos libros” (say two books) instead of “leer dos libros” (read two books).
It would be great if future updates included:
- Better context retention for iterative image editing.
- More control to fix or regenerate individual parts of an image.
- Improved language accuracy for common educational phrases.
- Tighter alignment between text instructions and visuals.
Here is an example of the results for what should have been a ten-step bedtime routine; notice only eight steps are included.
Thanks for your work — there’s a lot of potential here, and I hope future versions will make it more reliable for creating helpful, multilingual visual aids for young kids.
