How to Use FLUX Erase: Mask-Based Object Removal for 10 Credits

FLUX Erase on VdoBloom removes an unwanted object from a photo for a flat 10 credits per run, and it is driven entirely by a mask you paint over the image — not by a text prompt. It is Black Forest Labs’ dedicated erase model, wired into the VdoBloom image editor as FLUX Erase on the Image Edit page. You upload one source image, brush over whatever should disappear, and the model reconstructs the painted region from the surrounding pixels. There is no describing, no negotiating, no re-prompting — which is exactly why it succeeds at removal jobs that conversational editors quietly fumble.

What FLUX Erase is actually good at

Most AI image editors are instruction models: you type “remove the trash can” and hope the model finds the trash can, understands you mean that one, and does not redecorate the rest of the frame on its way out. FLUX Erase is a different class of tool. It takes two inputs — the source image and a black-and-white mask where white marks the area to destroy — and rebuilds only that area. Nothing outside the mask is regenerated, so skin tone, grain, product colour and background texture stay byte-for-byte where they were.

That makes it the right model for a narrow, common set of jobs: a photobomber at the edge of a portrait, a power line across a skyline, a logo on a mug you cannot legally show, a price sticker on packaging, a stray cable in a product shot, a date stamp burned into the corner of an old photo. It is a retoucher, not an illustrator.

The honest limitation: because it reconstructs from context, it needs context to reconstruct from. Erasing a person standing in front of a plain wall works beautifully. Erasing a person standing in front of a detailed bookshelf, where the model has to invent book spines that were never photographed, is a coin flip. The larger and more structured the masked region, the more you should expect to re-roll.

Real specs and credit cost

VdoBloom bills every model from one credit balance, so the maths is simple. On the Lite plan, $15 buys 300 credits, which means one credit costs about $0.05. Annual billing lowers that: the Pro annual tier is $49/month for 2,000 credits, or roughly $0.0245 per credit. Here is how FLUX Erase compares to the neighbouring editing models on the platform.

Model	What it does	Inputs	Credits per image	Cost at $0.05/credit
FLUX Erase	Mask-driven removal and fill	Source image + painted mask (2 images max)	10	$0.50
FLUX Outpainting	Expands the canvas outward	1 source image	12	$0.60
P-Image-Edit	Fast instruction editing	Up to 10 images	6	$0.30
Seedream 4.5 Edit	Instruction editing, multi-image	Up to 10 images	7	$0.35
Flux Kontext Max	Premium single-image instruction edit	Exactly 1 image	10	$0.50
Nano Banana Pro (1K/2K)	Heavyweight editing with resolution control	Up to 8 images	18	$0.90

A few specifics worth knowing before your first run. FLUX Erase accepts a maximum of two images — the source and the mask — and the mask is mandatory; a run without one is rejected before any credits move. The in-browser mask painter caps the working canvas at 2,048 pixels on its longest side, so extremely large originals are scaled to that ceiling. There are no aspect-ratio, resolution or output-format controls on this model, because it returns the source image with a region rewritten rather than generating a new frame. The model carries a 4.2 user rating on VdoBloom.

The exact steps

1. Open the Image Edit page and pick FLUX Erase from the model list. It sits in the Flux Edit group alongside FLUX Outpainting.
2. Tick the upload consent box, then upload your source image. The panel will read “Upload Source Image — then paint over what to erase.”
3. The mask painter loads with your photo underneath. Drag the brush-size slider — it starts at 30 pixels — down for fiddly edges and up for big regions.
4. Paint over the object. The red overlay is the mask preview; everything you cover in red is what the model will erase and fill. Undo steps back one stroke, and clear wipes the mask if you want to start over.
5. Paint slightly past the object’s edge. Leaving a halo of the original silhouette unmasked is the single most common cause of a ghosted, smeared result.
6. Generate. Ten credits are deducted, the mask is uploaded as the second image, and the result comes back as a full-size edited image you can download or push into a video tab.

If you are erasing something from a photograph that contains a real, identifiable person, you must have that person’s consent to upload and edit their image.

How to get a clean result first try

There is no prompt to tune here, so all of your control lives in the mask. Three habits do most of the work.

Mask generously, not tightly. Shadows and reflections belong to the object. If you erase a bollard but leave its shadow on the pavement, the model will often reconstruct a new bollard to justify the shadow. Cover the object plus its cast shadow and any reflection in glass or water.

Erase one thing at a time. Two separate 10-credit runs on two objects beat one run with a scattered mask across the frame. Sequential passes also let you keep the better of two attempts instead of re-rolling everything.

Pick the easy background. Where you have several frames, erase from the one where the object sits against sky, wall, grass or blurred bokeh.

When to pick a different model

FLUX Erase only subtracts. The moment your edit adds, replaces or restyles something, you want an instruction model instead. For a premium single-image change with strong typography handling, Flux Kontext Max at 10 credits is the direct counterpart. For cheap, high-volume instruction edits, P-Image-Edit at 6 credits is the value pick. If your problem is that the frame is too tight rather than too busy, FLUX Outpainting at 12 credits extends the canvas instead of rewriting inside it. And if you want to browse the whole editing line-up with ratings and inputs side by side, the FLUX Erase model page links out to every sibling.

A worked example

Say you have a product photo of a ceramic mug on a wooden table, and a competitor-branded bottle sits half in frame on the right, plus a crumb of masking tape on the table edge. Two runs.

Run one: upload, brush size around 60, paint the bottle and the soft shadow it throws onto the wood, extending the mask about ten pixels past the glass edge. Generate. Ten credits. The wood grain rebuilds cleanly because the surrounding table gives the model plenty of reference.

Run two: re-upload the result, drop the brush to about 12 pixels, paint the tape crumb, generate. Another ten credits. Total spend: 20 credits, roughly $1.00 on a Lite plan or about $0.49 at the annual Pro rate. See the tiers on the pricing page; one-time credit packs start at $2.49 for 75 credits and never expire, which suits people who batch retouching a few times a year.

Frequently asked questions

Does FLUX Erase use a text prompt?

No. The model is purely mask-driven — the prompt field is ignored for this model, and the reconstruction comes entirely from the pixels surrounding your painted region. If you want to describe a change in words, use an instruction editor such as Flux Kontext Max or Seedream 4.5 Edit instead.

How many credits does one FLUX Erase run cost?

Ten credits, flat, regardless of image size or how much of the frame you mask. At the Lite rate of about $0.05 per credit that is $0.50 per erase, or roughly $0.245 on the annual Pro tier.

Do I have to make the mask myself in another app?

No. VdoBloom includes an in-browser mask painter: upload the source image and brush directly over it, with adjustable brush size, undo and clear. The mask is generated and submitted for you as the second image.

Why did my erase leave a blurry patch?

Almost always an under-sized mask or an over-ambitious one. Re-run with the mask extended a little past the object outline, including its shadow. If the region was large and sat against detailed structure such as text, tiles or shelving, split it into smaller sequential erases.

Can I remove a person from a photo with it?

Technically yes, and it works best when they are against a simple background. You still need that person’s consent to upload their photo, and the model will not process explicit or otherwise prohibited content.

Is the output watermarked?

Free accounts get watermarked downloads. Paid plans download watermark-free with commercial rights, which matters if the retouched image is going on a product listing or into a client deliverable. For broader photo-editing workflows, the AI image editor page walks through the rest of the edit models.