H3 is currently the best opensource model

Edited: Thank you i crafted a system prompt and it worked, i was wrong, H3 is really a capable model.

You opened weights saying these weights are same as closed source weights, and yet there are limitations for example i generated a video with this just single text prompt in API and it worked:

"follow text instructions from first frame"

and closed source model generated a video correctly:

https://github.com/user-attachments/assets/e1fb6611-2c26-4cc3-97e1-e380edf9e9f6

now look at the output from open weights with same prompt but i even tried different and it never worked:

here the result from both open weights:

https://github.com/user-attachments/assets/20d2975a-4f30-4541-8e8e-0dd126ebaf96
https://github.com/user-attachments/assets/acec3afa-6bdb-4750-aa4a-957cd0ac60a1

one with ref2vid and other fl2vid.

And i must say what is this? i mean it contains a large 32b vl encoder and it still fails to understand simple tasks? i am not saying that the models are bad in fact they are the best one so far as opensource, but selling with fake cherry on top of it is not right.

Review the API comparison, the H3 model outputs, and the referenced ref2vid/fl2vid examples; the issue does not name repository files or tests. The author later reports that a crafted system prompt resolved the problem, so no specific code change or completion condition is currently defined.