Flux 2 Image Generator: Pro, Flex and Dev Compared

Run Flux 2 Pro, Flex and Dev for text-to-image and multi-image edits, with the credit cost of every render shown before it starts and no plan required at all.

About Flux 2

Flux 2 is Black Forest Labs' image family, and this page runs all three builds: Pro for fidelity, Flex for a speed and quality balance, and Dev, the open-weight version. Each one accepts a written prompt or an edit built from reference images, and the credit total updates before you run.

The first decision on this page is the build, and it matters. Flux 2 Pro is the quality build and accepts up to eight references in a single edit. Flex is described in the workspace as a speed and quality balance, and it is the more expensive of the two per plain run. Dev is the open-weight build, and its fast sampling path makes it the cheapest way to explore a direction. A plain text-to-image run costs 14 credits on Pro, 27 on Flex and 6 on Dev at one megapixel. Attach reference images and the price rises per image, so the total shown beside the generate button is the one that gets charged.

Image to image is where the family earns its place. Upload up to eight pictures in one request and the model reads them as a set: several angles of a product, a colour reference, a texture, a face. The prompt should describe the change, not the whole scene, because the pixels already carry most of the answer. Text to image is the opposite discipline. There is nothing to preserve, so the prompt has to name the subject, the light and the material. Both modes render at one megapixel and both write a single file per run, PNG by default or JPEG when you need something smaller to hand over.

Compared with the other pages on this site, Flux 2 is the one built for multi-reference work. Imagen 4 takes a prompt and nothing else. Ideogram works one image at a time, through a mask or a remix, and is organised around lettering. If your job is to keep a product, a palette or a face consistent across several scenes, the reference set here does that in one run instead of a chain of separate edits. Aspect ratios include square, portrait, landscape and 21:9, and a failed or cancelled prediction returns its credits automatically.

Auto is the ratio to leave selected in image to image. The selector describes it as matching the input when one is provided, so a portrait source stays portrait and a wide plate stays wide without you translating the shape into a number. Choose a fixed ratio only when the output has a destination that overrides the source, such as a square avatar or a story frame. In text to image, Auto has no input to match and settles on a square, so pick the ratio you actually need.

Dev is where the wording gets settled, and it is the only build with the go_fast switch; Pro and Flex expose no equivalent, so speed cannot be traded for quality on those builds. The reference set also deserves a trim before the final pass. Treat every file as having a job: a second angle that contradicts the first, or a texture that fights the product, gives the model conflicting instructions and the edit drifts. Remove anything that is only related, keep the set that agrees, and move the same prompt to the build that will be published.

How Flux 2 works

1
Pick the build
Pro for the image you intend to publish, Flex when you want its stated speed and quality balance, Dev for cheap exploration with fast sampling. All three share the same workspace and the same reference uploader.
2
Write the prompt for the mode
In text to image, describe subject, light and material. In image to image, upload the references first, then name only the change and where it applies. The uploader accepts up to eight images per request.
3
Set the frame and run
Choose a native ratio from 1:1 through 21:9, choose PNG or JPEG, and read the credit total next to the generate button. That is the amount charged when the run starts.

What Flux 2 is best for

Consistent product sets
Upload several angles of the same item and rewrite only the setting. The reference set holds the product still while the scenes change.
Brand palette control
Add a colour reference to the uploads and ask for it to be carried through. Pro is the quality build, so keep the final pass there.
Wide banners and posters
21:9 and 9:16 render natively, so a cinemascope banner or a vertical story keeps the composition instead of being cropped from a square.
Cheap drafts, paid finals
Explore directions on Dev at 6 credits, then rebuild the chosen frame on Pro once the prompt is settled. The same prompt structure carries across.

Flux 2 at a glance

Model: Black Forest Labs Flux 2 Pro, Flux 2 Flex and Flux 2 Dev
Credit cost: 14 credits per plain text run on Pro, 27 on Flex and 6 on Dev; reference images add credits and the total is shown before you run
Modes: Text to image, image to image
Reference images: Up to 8 per request, PNG, JPG, JPEG or WEBP, 10 MB each
Aspect ratios: Auto, 1:1, 3:4, 2:3, 4:5, 4:3, 3:2, 5:4, 16:9, 9:16, 21:9
Output: One PNG or JPEG per run, 1 megapixel

Flux 2 FAQ

What does a Flux 2 image cost?: A plain text-to-image run is 14 credits on Flux 2 Pro, 27 on Flex and 6 on Dev. Each reference image adds credits, so an edit with several uploads costs more; the generate button shows the exact total first. New accounts get 300 credits on signup and can claim 50 more once a day, both expiring after 30 days. Starter is $4.90 a month and includes 300 credits.
Which of the three builds should I use?: Pro when the image will be published, Flex when you want its speed and quality balance, Dev when you are still exploring. Pro is the quality build and tops out at eight references; all three run in the same workspace.
Can I edit a photo instead of generating from scratch?: Yes. Switch to image to image, upload up to eight references, and describe the change. Keep the instruction narrow, such as swapping the surface, changing the light or restyling one garment, because the uploads already define everything you do not mention.
How is this different from Imagen 4 or Ideogram?: Imagen 4 is text to image only, with no uploads. Ideogram is built around readable lettering and works one image at a time, through a mask or a remix. Flux 2 is the page that takes a set of references and keeps them consistent, which is the case for product catalogues, character sets and brand assets.
What happens if a run fails?: Credits are refunded automatically when the prediction fails or is cancelled. You pay for delivered images, not for attempts.
Should I use Auto or fix the aspect ratio?: In image to image, leave Auto selected and the render follows the frame you uploaded, which keeps a portrait portrait and a wide shot wide. Fix a ratio when the output has a destination that overrides the source, such as a square profile image or a vertical story. In text to image there is no input to follow, so Auto falls back to a square and you should pick the ratio yourself.

About Flux 2

How Flux 2 works

Pick the build

Pro for the image you intend to publish, Flex when you want its stated speed and quality balance, Dev for cheap exploration with fast sampling. All three share the same workspace and the same reference uploader.

Write the prompt for the mode

In text to image, describe subject, light and material. In image to image, upload the references first, then name only the change and where it applies. The uploader accepts up to eight images per request.

Set the frame and run

Choose a native ratio from 1:1 through 21:9, choose PNG or JPEG, and read the credit total next to the generate button. That is the amount charged when the run starts.

What Flux 2 is best for

Consistent product sets

Upload several angles of the same item and rewrite only the setting. The reference set holds the product still while the scenes change.

Brand palette control

Add a colour reference to the uploads and ask for it to be carried through. Pro is the quality build, so keep the final pass there.

Wide banners and posters

21:9 and 9:16 render natively, so a cinemascope banner or a vertical story keeps the composition instead of being cropped from a square.

Cheap drafts, paid finals

Explore directions on Dev at 6 credits, then rebuild the chosen frame on Pro once the prompt is settled. The same prompt structure carries across.

Flux 2 at a glance

Model

Black Forest Labs Flux 2 Pro, Flux 2 Flex and Flux 2 Dev

Credit cost

14 credits per plain text run on Pro, 27 on Flex and 6 on Dev; reference images add credits and the total is shown before you run

Modes

Text to image, image to image

Reference images

Up to 8 per request, PNG, JPG, JPEG or WEBP, 10 MB each

Aspect ratios

Auto, 1:1, 3:4, 2:3, 4:5, 4:3, 3:2, 5:4, 16:9, 9:16, 21:9

Output

One PNG or JPEG per run, 1 megapixel

Flux 2 FAQ

What does a Flux 2 image cost?

A plain text-to-image run is 14 credits on Flux 2 Pro, 27 on Flex and 6 on Dev. Each reference image adds credits, so an edit with several uploads costs more; the generate button shows the exact total first. New accounts get 300 credits on signup and can claim 50 more once a day, both expiring after 30 days. Starter is $4.90 a month and includes 300 credits.

Which of the three builds should I use?

Pro when the image will be published, Flex when you want its speed and quality balance, Dev when you are still exploring. Pro is the quality build and tops out at eight references; all three run in the same workspace.

Can I edit a photo instead of generating from scratch?

Yes. Switch to image to image, upload up to eight references, and describe the change. Keep the instruction narrow, such as swapping the surface, changing the light or restyling one garment, because the uploads already define everything you do not mention.

How is this different from Imagen 4 or Ideogram?

Imagen 4 is text to image only, with no uploads. Ideogram is built around readable lettering and works one image at a time, through a mask or a remix. Flux 2 is the page that takes a set of references and keeps them consistent, which is the case for product catalogues, character sets and brand assets.

What happens if a run fails?

Credits are refunded automatically when the prediction fails or is cancelled. You pay for delivered images, not for attempts.

Should I use Auto or fix the aspect ratio?

In image to image, leave Auto selected and the render follows the frame you uploaded, which keeps a portrait portrait and a wide shot wide. Fix a ratio when the output has a destination that overrides the source, such as a square profile image or a vertical story. In text to image there is no input to follow, so Auto falls back to a square and you should pick the ratio yourself.