Regenerate

When AI doesn’t meet your needs with its initial generation, you can instruct it to regenerate the output while keeping the same prompt and context. Common labels for this action include “Regenerate,” “Rerun,” or “Try again.”

By default, the system reruns your request through the model’s internal process, which is designed to allow variation. That’s why a new draft may use different words, reasoning, or details even though you didn’t change anything.

Some tools offer a “seed” setting that locks in part of this process. Using the same seed makes the model more likely to produce consistent results across regenerations. Changing or omitting the seed lets the system explore other possibilities, which can be useful for brainstorming or creative work, but can frustrate users expecting consistent results.

Unlike more intentional controls such as “Edit prompt” or “Refine,” regeneration is a blunt tool. It sacrifices precision for speed. Its design must be carefully balanced against adjacent patterns like branching, inline edits, or parameter controls, so that it remains helpful rather than overwhelming or destructive.

Regenerative modes

Product designers must decide how to support the relationship between the original generation and additional versions:

• Overwrite: The new output replaces the old one. This is common in chat tools where only the latest answer remains visible in-line. If inline-variants are supported, users can cycle through past variations.
• Branching: Each retry creates a separate version. This mode is common in video and text editors, and recently has had some adoption in conversational tools like ChatGPT. This mode supports comparison and exploration but can quickly fill up a workspace if not well organized.

Designers also must determine how much control users will have over the form of regenerations as they iterate:

• Guided: Instead of blindly rerunning, the user can adjust a parameter or style first. Tools like Jasper’s “Rephrase” or Notion AI’s “Try again with…” fall here. It gives users more precision without needing a full new prompt.
• Random seed: Users may also be allowed to regenerate without specifying changes to the prompt. This is common when the thing being regenerated is incidental, like a list of suggestions. However, this lack of control can lead to frustration and a lack of agency.

Finally, systems may be prompted to regenerate automatically after an error or timeout. This is common in coding tools, where the assistant silently retries a completion that fails. It reduces friction but must be transparent so users know what happened.