How to Use FLUX.2 [dev] on VdoBloom: 4-Credit Reference Image Editing
A practical guide to FLUX.2 [dev] on VdoBloom: 4 credits ($0.20) per image, up to 4 reference images, 512-2048px output, how to prompt it, and when a pricier editor is the better pick.
FLUX.2 [dev] on VdoBloom is a 4-credit reference-image editor: you upload up to four reference images, write one instruction, and it returns a new square-by-default image for about $0.20 a render. It sits in the image editing model list rather than the pure text-to-image list, and its job is guided edits — changing, combining or restyling material you already have — at the cheapest price point of any Black Forest Labs model on the platform.
What FLUX.2 [dev] is actually good at
The model runs through Runware on VdoBloom and is described in the picker exactly as it behaves: guided edits with up to 4 reference images. That reference-image ceiling is the important part. Most cheap editors take a single source image and nudge it. FLUX.2 [dev] accepts up to four inputs at once, which makes it a useful tool for jobs where the answer depends on more than one picture:
- Putting a product from one photo into the setting from another.
- Carrying a character’s face and outfit from two separate reference shots into one new frame.
- Restyling a photo while holding a colour reference or texture swatch as a second input.
- Cheap volume iteration — five variations of an edit costs 20 credits, which is less than one Nano Banana 2 render.
It is rated 3.9 by VdoBloom users, which is honest positioning: it is not the highest-fidelity editor in the catalogue. It is the one you reach for when you want twenty attempts rather than three.
Real specs and credit cost
Credits convert at a flat rate. The Lite plan is $15 for 300 credits, so one credit is $0.05 and the arithmetic below is exact, not estimated.
| Spec | FLUX.2 [dev] on VdoBloom |
|---|---|
| Credit cost | 4 credits per image |
| Dollar cost | $0.20 per image |
| Capability | Image editing / reference-to-image |
| Reference images | Up to 4 |
| Aspect ratio selector | None — output is driven by requested pixel dimensions |
| Default output | 1024 × 1024 |
| Dimension range | 512 to 2048 px, snapped to multiples of 16 |
| Resolution tiers | None — single flat price, no 1K/2K/4K ladder |
| Content filter tier | MODERATE |
| User rating | 3.9 |
| Renders from 10 free signup credits | 2 |
There is no resolution upcharge, which is unusual. Models like Nano Banana 2 or WAN 2.7 Image Pro charge more as you climb a 1K/2K/4K ladder. FLUX.2 [dev] charges 4 credits whether you ask for 512 px or 2048 px, so the largest dimension you can use is also the best value.
How it prices against the other editors
| Image editing model | Credits per image | Dollar cost | Pick it when |
|---|---|---|---|
| Z-image Turbo | 2 | $0.10 | You want the absolute cheapest pass and fidelity is secondary |
| FLUX.2 [dev] | 4 | $0.20 | Multi-reference guided edits at volume |
| Qwen Image 3.0 Edit (1K) | 5 | $0.25 | Edits that need a more permissive content tier |
| Nano Banana | 6 | $0.30 | Higher polish on a single-reference edit |
| Nano Banana 2 Lite | 8 | $0.40 | Up to 10 input images, aspect ratio control |
| Flux 2 Pro I2I | 12 | $0.60 | The same family at production fidelity, 1K/2K, up to 8 inputs |
| Nano Banana 2 | 15 | $0.75 | Final-quality edits, up to 14 images, 4K |
The practical read: FLUX.2 [dev] is the draft tier of the Flux line and Flux 2 Pro I2I is the delivery tier. Three dev attempts plus one Pro finish costs 24 credits, which is cheaper than two Nano Banana 2 renders and usually lands better, because you have already worked out what the instruction should say.
How to prompt it well
FLUX.2 [dev] responds to instructions, not scene descriptions. The failure mode is writing a caption of the picture you want instead of a change you want made.
- Name the reference by position. With multiple inputs, say which does what: keep the subject from the first image, apply the lighting from the second.
- State what must not change. Explicitly preserving the face, the pose and the background stops the model rewriting parts you were happy with.
- One change per render. At $0.20 a go, two sequential single-change edits beat one compound instruction that the model half-applies.
- Ask for the dimensions you need up front. There is no aspect ratio dropdown, so the pixel size you request is the framing you get, and anything not a multiple of 16 is snapped down.
- Use a negative prompt for artefacts, not for content. It is effective at suppressing text, watermarks and extra limbs; it is not a way around the content filter.
A worked example
Say you have a product photo of a ceramic mug on a white background and a separate photo of a kitchen worktop you like. Upload both to the image editor, pick FLUX.2 [dev], and instruct: place the mug from the first image onto the wooden worktop from the second image, matching the warm window light and casting a soft shadow to the left, keep the mug’s glaze colour and handle shape exactly as in the first image. Request 1536 × 1024. That is 4 credits.
Render it four times. Sixteen credits, eighty cents, and you now know whether the shadow direction or the light temperature is what needs fixing. Take the best instruction, run it once more on Flux 2 Pro I2I for the deliverable. Total spend: 28 credits, $1.40, versus burning 60 credits guessing at instructions on a premium model.
When to pick a different model
FLUX.2 [dev] is the wrong tool in four situations. If you need a specific aspect ratio from a menu rather than from pixel maths, Nano Banana 2 Lite or Qwen Image 3.0 Edit give you a dropdown. If you need genuine 4K output, this model tops out at 2048 px on the long edge and Nano Banana 2 does not. If you are starting from nothing — no reference image at all — the text-to-image side of the platform has models built for that brief. And if your subject sits in the swimwear, dance or fitness territory that mainstream tools decline, FLUX.2 [dev] carries the MODERATE label, not the flexible HOT one, so a Qwen or WAN editor is the more reliable route. Explicit and illegal content stays blocked on every model regardless of tier.
Model facts, previews and the current badge live on the FLUX.2 [dev] model page, and the credit-to-dollar maths for every plan is on the pricing page. New accounts get 10 free credits with no card, which is two FLUX.2 [dev] edits to test the workflow before paying anything.
Frequently asked questions
How much does one FLUX.2 [dev] image cost?
4 credits, which is $0.20 at the Lite plan rate of $15 for 300 credits. There is no resolution surcharge — a 2048 px render costs the same 4 credits as a 512 px one.
Can FLUX.2 [dev] generate an image from text alone?
The underlying model supports text-to-image, but on VdoBloom it is listed in the image editing picker, where a reference image is the expected input. For text-only starts, use a model from the generation list instead.
How many reference images can I upload?
Four. If you send more, only the first four are used. That is fewer than Nano Banana 2 at fourteen or Flux 2 Pro I2I at eight, and it is the main reason to move up a tier on complex composites.
Why is there no aspect ratio selector?
The model takes explicit width and height rather than a ratio preset. VdoBloom passes the dimensions through, clamped to the 512–2048 range and snapped to a multiple of 16, so you control framing by pixel size.
Do FLUX.2 [dev] images come out watermarked?
On a paid plan, no — paid plans download watermark-free and carry commercial rights. Free-tier output is watermarked.
Ready to try it?
Create your first AI video in minutes — no credit card required.
Start Creating Free →