Watermark Remove on the deep-imager AI image generator cleans overlay artifacts on a photo you upload — it does not invent a new scene.
Upload, mask or full-frame, Nano Banana only. Then stop — do not restyle the SKU.
What does Watermark Remove actually do?
It takes an uploaded photo, targets watermark or logo overlays, and reconstructs those pixels while keeping the rest of the frame.
Open the workspace tab labeled Watermark Remove (or /?mode=watermark_remove). You must upload a source image. Paint the overlay so only that region is editable; if you skip the mask, submit uses full-frame cleanup instead.
This is not image to image restyle. Image to image will change background, color, or props. Watermark Remove is written to leave subject, lighting, and packaging text in place.
Record what worked: model name, ratio, and the exact prompt line that survived review. That notebook becomes your reusable playbook across ecommerce, social, and concept batches.
On deep-imager, treat the AI image generator as a credit-aware studio: lock aspect ratio, pick the model, then write the prompt so text to image spends once on a usable frame.
Which model can you use for Watermark Remove?
Only Nano Banana — the API rejects other models for this mode.
Keep the selector on Nano Banana. GPT-4o image and Midjourney-style paths are for new stills, not this cleanup lane.
A prompt is optional. If you type one, it is appended as an extra request; the locked instructions still say: remove overlay traces, do not invent new logos, preserve real labels.
Record what worked: model name, ratio, and the exact prompt line that survived review. That notebook becomes your reusable playbook across ecommerce, social, and concept batches.
On deep-imager, treat the AI image generator as a credit-aware studio: lock aspect ratio, pick the model, then write the prompt so text to image spends once on a usable frame.
- Upload a photo you have the right to edit.
- Paint the watermark, grid, or logo — or close the editor for full-frame.
- Do not ask the model to restyle the product; that is a different tab.
When should you generate a new still instead?
Use text to image when you need a new scene; use image to image when you need a new set around the same object.
AI stills from this generator have no output watermark; paid plans add HD originals. Cleaning an old listing photo is Watermark Remove. Inventing a lifestyle shot is text to image. Swapping marble for seamless color is image to image, covered on the product photo use case.
When a still is close, prefer image to image to keep composition while you change lighting, background, or crop intent. Regenerating from a blank text to image prompt usually costs more exploratory credits.
Record what worked: model name, ratio, and the exact prompt line that survived review. That notebook becomes your reusable playbook across ecommerce, social, and concept batches.
On deep-imager, treat the AI image generator as a credit-aware studio: lock aspect ratio, pick the model, then write the prompt so text to image spends once on a usable frame.
How many credits does Watermark Remove spend?
Each submit spends the same credit as other Nano Banana generates; the tab itself is free to open.
New accounts have 5 free credits. If you need a batch of listing cleanups, pricing from $1/month lists Basic at 10 credits, Advanced at $9.99 for 120, and Professional at $19.99 for 300.
Failed jobs that the product marks failed return the credit. Typical still time is 30 seconds–2 minutes. Read how credits work before you loop the same overlay.
Record what worked: model name, ratio, and the exact prompt line that survived review. That notebook becomes your reusable playbook across ecommerce, social, and concept batches.
On deep-imager, treat the AI image generator as a credit-aware studio: lock aspect ratio, pick the model, then write the prompt so text to image spends once on a usable frame.
| Job | Tab | Model |
|---|---|---|
| Clean overlay on an upload | Watermark Remove | Nano Banana only |
| New scene, no file | Text to image | Any still model |
| Keep object, change set | Image to image | Img2img-capable models |
How do you keep quality high without wasting credits?
Lock ratio first, run one strong text to image pass, then refine with image to image instead of endless fresh prompts.
Before you click Generate, confirm the credit cost on the button and the aspect ratio for the channel you will actually publish to.
If the model returns a safety or policy error, rewrite the prompt without protected names or sensitive requests rather than retrying the same blocked wording.
For team handoff, save the winner still and the prompt in your dashboard history so the next editor can Make Same or continue with image to image.
On deep-imager, treat the AI image generator as a credit-aware studio: lock aspect ratio, pick the model, then write the prompt so text to image spends once on a usable frame.