If a video generation fails (error, no usable output, or a run lost while queued/processing), the credits used should be automatically refunded when applicable. You can review credit usage and refunds under Settings → Usage. Refresh the page and restart Venice to see your updated balance.
To avoid interruptions during video generation, ensure the Venice.ai page remains open while the process runs. Avoid refreshing the page or navigating away, and refrain from repeatedly clicking the "Generate" button after encountering an error. If using a mobile device, keep it awake until the generation completes.
If any of the failures did not get refunded, contact support with a screenshot from Settings → Usage of any failed generations and we will make it right. Please also send the Task ID, which can be copied by clicking the 3 dots next to any generation.
If you see an "insufficientBalance" error, no credits were charged — this error prevents a generation from starting, so nothing is deducted from your balance. It usually appears because your credits are split across separate buckets (DIEM credits, monthly plan credits, and purchased USD credits), and a single generation must be fully covered by one bucket. Venice cannot combine credits across buckets for a single task. For example, if a generation costs 195 credits and you have 100 monthly plan credits and 114 purchased USD credits (214 total), no single bucket covers the full cost, so the generation is blocked even though your combined balance is sufficient.
To resolve this, check which single bucket has the largest balance and ensure it holds enough to cover the full generation cost on its own. Combining credits across buckets for a single generation is a feature we are actively working on and is not yet available. For full details on credit types and spend order, see Which credits are used first?
If a video completed but the result missed what you wanted, that is an output-quality issue rather than a failure. Credits are not returned by default for completed results that missed the desired look. Use Prompt Enhancer or a web scraping model to improve the prompt, then try again deliberately rather than rapid retries.
Some third-party closed-source video models have their own content policies, so a blocked output can be enforced by the model provider rather than Venice. For less restrictive video generation, try Wan or Seedance where available.