Regenerating or Tweaking an Image That Didn't Come Out Right
Two different ways to fix an image
When a generated image isn't quite what you wanted, SoulrisAI gives you two different paths forward, and picking the right one saves you time and cost.
- Regenerate — run the same (or a revised) prompt again as a fresh generation.
- Edit — start from the image you already have (one you generated, or one you uploaded) and ask for changes to it specifically.
They're not interchangeable. A straight regeneration gives the model a clean slate, so it's unpredictable — rerunning an identical prompt won't reliably give you the same image twice. Editing instead passes your existing image in as a reference, so the result builds on what's already there rather than starting over.
When to just regenerate
If the result missed the mark broadly — wrong subject, wrong composition entirely, style completely off — it's usually faster to revise your prompt and generate again from scratch rather than try to edit your way there. See writing effective image prompts for how to adjust a prompt that isn't landing.
A useful habit: change one element of the prompt at a time (just the lighting, just the style word) between regenerations, so you can tell which part of the wording was responsible for what changed.
When to edit instead
Editing makes more sense when the image is largely right and you want to adjust something specific — a color, an object, a detail — without redoing the whole composition. To edit, you supply the existing image as a source alongside your instructions for what to change, rather than writing a prompt from zero.
You can use either a previously AI-generated image or an image you've uploaded yourself as the source for an edit — both are treated the same way as a starting point.
A note on cost for edits
An edit includes the cost of processing your source image as a reference, on top of the generation itself — so an edit isn't necessarily cheaper than a fresh generation, even though it can feel like a "smaller" request. See choosing an image quality tier and what it costs for how generation cost is structured generally.
If editing isn't available
Not every model in every quality pool supports image editing. If you try to edit and it's not available for your current quality tier, that's a temporary or tier-specific limitation rather than something broken on your end — try a different quality tier, or fall back to a fresh regeneration with a more detailed prompt describing the change you wanted.
If the result still isn't right after a few tries
If you've regenerated or edited a few times and you're still not getting close, the most likely culprit is the prompt rather than the tier or the editing flow — step back and check it against the elements in writing effective image prompts rather than continuing to regenerate blind. If generation is actually erroring out rather than just producing a result you don't like, see common reasons an image generation fails.
Try it yourself
Next time a result is 80% right, try editing it with a specific instruction instead of rewriting your whole prompt — it's often the faster path to the image you actually wanted.