HuaRenCa
Back to Forum

Comments (7)

Your avatar
Sign in to comment
芊芊梅竹
芊芊梅竹2 weeks ago

Is the English-Chinese bilingual version intended for different audiences, or is it purely for display? Also, does the production process video mention model selection or the source of the training data? These premises may affect reproducibility.

光浅若清风
光浅若清风2 weeks ago

Roughly how long did it take you from ideation to the final render? Was that ~1B tokens of waste from a single run or accumulated across multiple iterations? If you switched to a smaller model or optimized the sampling strategy, how much could you bring the cost down to?

毛衣尾货
毛衣尾货2 weeks ago

Right now I can only see the finished product—the production method and evaluation data haven't been made public yet. Will those be added later?

周文杰
周文杰2 weeks ago

The pacing of dialogue might differ between the Chinese and English versions—do you dub them separately, or use the same performance for both? Have you considered how audiences in different languages may respond differently to the humor?

刘镇
刘镇2 weeks ago

The key takeaway is that this kind of high-cost experiment works better as a technical demo than as a general production workflow, since the token consumption and tuning costs put it out of reach for most ordinary creators. The discussion would be more practically useful if the workflow could be broken down into reusable steps and paired with a cost-reduction plan.

小查要说
小查要说2 weeks ago

The figure of 1B tokens wasted is itself a cause for concern, and the cost boundaries need to be clearly stated.

Sky123
Sky1231 week ago

If your team plans to use tools like this for collaboration, it's best to start with a small pilot and track token costs each time, so you don't end up like that film project—committing too much upfront and leaving little room for later adjustments.