Generative ads are probabilistic: the same Generate click can produce a strong variant or a miss where the bottle shape, colorway, or promo text drifts from your brief. That is not always a platform failure—it is often a signal to tighten inputs. Rhinobird gives you levers—product imagery, model choice, duration, language, and Creative Requirements—so the next run stays on-brand. Use this workflow when previews in History show the wrong SKU, illegal claims, or off-language copy.
Separate “wrong creative” from “broken job”
A broken job never lands in History or fails with system error; credit returns may apply under membership rules. A wrong creative delivers files that do not match your brief but completed successfully—fix the brief and regenerate rather than expecting automatic refunds. Always preview before download so you do not ship mismatched assets to ad platforms.
Audit the product image first
- Hero product fills enough of the frame that labels are legible.
- Color and packaging match the SKU you describe in text fields.
- Background clutter is removed—consider AI matting before re-upload.
- Resolution is adequate—consider AI upscale when marketplace zoom would expose blur.
If the model never saw your real label, it will hallucinate substitutes. No amount of regeneration fixes a thumbnail-sized product photo.
Tighten Creative Requirements
Creative Requirements is where you state mandatory copy, forbidden phrases, target audience, and tone. Treat it as a contract, not decoration:
- Paste the exact headline and subhead legal approved—do not paraphrase.
- List forbidden claims your compliance team rejects.
- Specify language and market when the tool exposes country or language controls.
- Call out required visual elements (“show cap color red”, “include gift box from photo”).
After editing requirements, run a small proof—shorter duration or fewer scenes—before a full credit-heavy batch. Confirm the new cost on Generate each time settings change.
Adjust model, duration, and format settings
Wrong pacing or aspect ratio can make a correct product feel off-message. Revisit video duration and aspect ratio fields before blaming the brief alone. For clone-style workflows on Video Clone, ensure the reference motion fits the Seedance model limit you selected; a mismatched reference can steer body language away from your intent even when the product image is correct.
Use History to iterate deliberately
Open History after each run. Compare previews from successive attempts instead of downloading every batch. When one variant is close, note which settings produced it and iterate from that baseline rather than resetting the entire form. Download only the run you intend to upload to ads; discard mismatched files locally so teammates do not grab the wrong export.
Regenerate vs manual fix
Regenerate when the error is semantic—wrong language, missing promo line, extra product variant. Manual fix in an editor may be faster for a single typo on a static image, but video copy drift usually requires a new generation with stricter requirements. Keep a change log (settings + brief diff) so your team learns which constraints actually work for each SKU category.
Credits and support boundaries
Review credit transaction details when a job fails technically; mismatched but completed output generally follows normal consumption. For repeated anomalies—same brief and photo yielding unrelated products across models—contact 13378452864 or support@rhinobird.ai with History timestamps. Read Service Agreement for AI output disclaimers and Privacy Policy for how inputs are handled.
Your brief and assets steer the model more than luck. Tighten Creative Requirements, fix the hero image, proof in History, then regenerate with purpose—not endless identical clicks.