FLUX Erase vs Flux Kontext Max: Same 10 Credits, Opposite Methods

FLUX Erase and Flux Kontext Max cost exactly the same on VdoBloom — 10 credits, $0.50 per run — so the choice between them is never about price: FLUX Erase deletes a region you paint and rebuilds the background behind it, while Flux Kontext Max rewrites an image from a natural-language instruction and is the stronger of the two whenever rendered text is involved. Same cost, opposite method, almost no overlap in what they are good at.

Identical price, different question

When two models sit at the same credit cost, the comparison stops being economic and becomes purely a question of method. FLUX Erase asks you where. Flux Kontext Max asks you what. Getting that backwards is the most common reason a Flux edit disappoints.

Both come from Black Forest Labs, the German lab founded by researchers behind the original Stable Diffusion work, and both carry a MODERATE content label on VdoBloom — standard provider filtering, neither strict nor unusually permissive. Beyond that shared lineage they behave nothing alike.

What FLUX Erase actually does

FLUX Erase is mask-driven inpainting. It takes a source image and a mask, and it reconstructs whatever the mask covers using the surrounding pixels. It does not accept a prompt. The underlying model is bfl:flux@erase, and the backend rejects any job that arrives without a mask.

On VdoBloom you never hand-build a mask file. Selecting FLUX Erase in the image editor swaps the usual upload grid for a two-step flow: upload the source, then paint directly over it with a brush. The painted area becomes the mask. There is no prompt box at all, and the generate button stays disabled until a mask exists.

The model is rated 4.2 by VdoBloom users. Its strength is that it cannot hallucinate an instruction, because there is no instruction. Its limit is the flip side of the same fact: it can only take away, never add. There is no way to ask it for a replacement object.

What Flux Kontext Max actually does

Flux Kontext Max is the top tier of the Kontext line, and it works in both directions — text-to-image from scratch, or upload-and-instruct editing on an image you already have. The Kontext mechanic is instruction editing that changes what you asked for and preserves what you did not. If ninety percent of a frame is already right, you upload it and describe the one change rather than re-prompting from zero.

It edits one image at a time, takes prompts up to 3,000 characters, and offers six aspect ratios: 21:9, 16:9, 4:3 (the editor default is 16:9), 1:1, 3:4 and 9:16. You can pick JPEG or PNG output. It is the higher-rated of the two Kontext tiers on VdoBloom at 4.5, and Black Forest Labs positioned the Max tier publicly around improved typography and prompt adherence — which is why it is the Flux model to reach for on poster comps, packaging mockups and anything where words inside the image have to survive close inspection. Full detail sits on the Flux Kontext Max model page.

Spec and cost comparison

Credit-to-dollar conversion below uses the Lite plan rate: $15 for 300 credits, so one credit is $0.05.

Spec	FLUX Erase	Flux Kontext Max
Credits per run	10	10
Cost at $0.05 per credit	$0.50	$0.50
Runs per 300-credit Lite month	30	30
Input method	Source image + painted mask	One image + written instruction
Prompt	None	Up to 3,000 characters
Text-to-image	No	Yes
Aspect ratios	Follows the source	21:9, 16:9, 4:3, 1:1, 3:4, 9:16
Output format choice	Provider default	JPEG or PNG
In-image typography	Not applicable	A headline strength of this tier
Content label	MODERATE	MODERATE
VdoBloom rating	4.2	4.5

The verdict by use case

Removing an object cleanly: FLUX Erase wins outright. Kontext Max can be asked to remove something, and often will, but it is re-rendering the frame through a language instruction — which means the rest of the image is in play. Erase touches only what you painted. For a power line, a bystander, a reflection, a stray prop, paint it and be done.

Anything involving rendered text: Flux Kontext Max, no contest. Erase has no text capability because it has no generative prompt. If you need a headline corrected on a packaging comp or a sign relabelled, Kontext Max is the Flux model built around that.

Starting from nothing: Flux Kontext Max. It is a generator as well as an editor. FLUX Erase requires an existing image by definition — the text-to-image lineup lives in the generation tab.

Reframing to a new aspect ratio: Flux Kontext Max. Six ratios from 21:9 ultrawide down to 9:16 vertical. Erase returns your source frame unchanged.

Predictability over creativity: FLUX Erase. When a client asset must not shift in any way except the one you selected, the model with no prompt field is the safer instrument.

A worked example: cleaning a location photo

You have an interior shot for a listing. There is a trash bin in the corner and the framed print on the wall has a legible brand name you cannot use.

Run one, FLUX Erase: upload the shot, paint over the bin, generate. The corner is reconstructed from the surrounding floor and wall. 10 credits, $0.50.

Run two, Flux Kontext Max: load the cleaned image, instruct it to replace the artwork in the frame with an abstract print in muted tones, keep the ratio at 16:9, output PNG. 10 credits, $0.50.

Total 20 credits, $1.00. Note which job went to which model: deletion to Erase, replacement to Kontext. Trying to route the bin removal through Kontext Max usually costs more in retries than the extra run costs outright, because a language instruction gives the model licence to re-render the whole corner rather than just the bin.

Practical notes before you run either

Paint your mask slightly wider than the object in FLUX Erase. A mask that hugs an edge too tightly leaves a fringe of the original object for the model to reconstruct around, and the result reads as a smear. A few extra pixels of margin gives it clean context to work from.

With Kontext Max, write the instruction as a change, not as a full scene description. Describing the entire image invites a full re-render; describing the one delta is what the Kontext design is tuned for.

If your source image contains a real person, you need that person’s consent before editing it, and the editor asks you to confirm this before it will run. New accounts get 10 free credits with no card — exactly one run of either model. After that, subscriptions start at $15 per month for 300 credits, annual billing brings Lite to $10 per month, and one-time packs start at $2.49 for 75 credits that never expire. Paid plans download watermark-free with commercial rights; the full breakdown is on the pricing page, and the mask-editing specs are on the FLUX Erase model page.

Frequently asked questions

Why do FLUX Erase and Flux Kontext Max cost the same 10 credits?

Credit costs on VdoBloom track what each model costs to run, not how many features it exposes. Both land at 10 credits per run, which is $0.50 at the Lite plan rate of $15 for 300 credits. Pick between them on method, not budget.

Can Flux Kontext Max remove objects like FLUX Erase does?

It can be instructed to, and it often succeeds. But it works by re-rendering the image from your instruction, so the areas you did not mention are still exposed to change. FLUX Erase only alters the pixels you painted, which makes it the more reliable removal tool at identical cost.

Which one is better for text inside an image?

Flux Kontext Max, decisively. Typography and prompt adherence are what the Max tier was positioned around. FLUX Erase has no prompt and therefore no ability to render text at all.

Does FLUX Erase need me to make a mask file myself?

No. The editor gives you a brush over your uploaded image and builds the mask from what you paint. White areas are the ones erased, and the run will not start until a mask exists.

Can Flux Kontext Max edit more than one image at a time?

No — it takes a single image per job. If you need several reference images considered together, Flux 2 Pro Image-to-Image handles one to eight inputs at 12 credits per run.