Skip to main content
Runflow
FLUX.2 DEV API

The most reliable FLUX.2 dev API,with $500 in matched credits.

32B parameters. 4 MP output. Up to 10 reference images per call. JSON prompting. The newest Black Forest Labs release, hosted on Runflow. Dollar-for-dollar credit match up to $500.

32B paramsNewest BFL release
10 referencesPer single call
Auto-failoverNo errors reach your users

SOUND FAMILIAR?

What breaks when you call FLUX.2 dev directly.

Same model, different reality once it's in production.

Failed generations

Provider returns 5xx. Generations fail mid-pipeline. Your users see errors and you pay for the retry.

Reference image chaos

Multi-reference workflows need orchestration. Without it, prompts drift, references collide, outputs feel stitched.

No fallback path

When your AI provider goes down, your pipeline goes down. No fallback.

JSON prompt glue

FLUX.2's JSON prompting needs schema validation. Every team rebuilds the same input contract.

HOW IT WORKS

Three steps. Zero failures.

Step 01

Send your request

Same params as any other provider. Swap the endpoint, keep your code.

Step 02

Auto-failover until it works

If a generation fails, we retry on our infrastructure. You never see the failure. You never pay for it.

Step 03

Sentinel scores the output

Reference adherence, text accuracy, color, artifacts. Scores land in your dashboard so you know exactly what shipped.

Runflow API
WAITING
0ms
POST /v1/generate
"Professional headshot, studio lighting"
+1.2s
Provider timeout
GPU overloaded. No charge.
+1.2s
Auto-retry initiated
Transparent to your code. Zero latency penalty.
+3.4s
200 OK
Image delivered: cdn.runflow.io/gen/img_9a8b.png
+3.8s
Sentinel evaluating...

Evaluate

See what you are actually shipping.

Every FLUX.2 dev generation through Runflow is evaluated by Sentinel. You get quality insights on every output: what went well, what went wrong, and where to improve your prompts.

black-forest-labs / flux-2-dev
Output qualitySentinel

Evaluations

2,924

last 30 days

The total number of generations scored by Sentinel in this period.

Pass rate

89.53%

of all evaluations

The percentage of evaluations where all judges passed. Higher is better.

Avg score

0.91

out of 1.00

The average weighted score from 0 to 1 across all evaluations. Each judge contributes based on its configured weight.

Gate failures

20

0.7% of evaluations

Evaluations terminated early by a pre-check, for example no face detected or not-safe-for-work content. These never reached the full judge panel.

Failed judges

1,133

0.4 per evaluation

The total individual judge failures across all evaluations. One generation can fail multiple judges.

Regenerate rate

9.68%

lower is better

The percentage of evaluations where Sentinel recommended regeneration. Lower is better.
Success rate by categoryclick to inspect
1,133failed judges
Input scoring

Catches bad inputs before you spend on a generation.

Output scoring

Every result graded against the checks that matter.

Trace on every call

One ID ties the input, the score, and the retry together.

ACT ON THE SCORE

Improve quality and user experience at the same cost.

Sentinel scores every output on reference adherence, text accuracy, color match, and artifacts. Pick the mode that fits your pipeline.

Discard bad generations automatically.

Adds a few seconds. If reference adherence drops, text breaks, or the score is low, the output is discarded and a fresh generation is triggered with the failure mode flagged.

Off-reference outputs, broken text, color drift caught before users see them
A few extra seconds, in exchange for guaranteed quality
Auto-retry with the failure mode passed back to the model
Guard evaluation
Nano Banana Pro
PASSED
94.2
Overall score
Face fidelity
96
Likeness match
93
Background clean
88
Artifact detection
97
Style consistency
91
img_3f2d58Face distortionRETRYING
img_3f2d91Retry successfulDELIVERED
WHY RUNFLOW

Built for FLUX.2 dev at scale.

10 reference images per call

FLUX.2 dev accepts up to 10 references in a single request. Use it for multi-product shots, style blends, or character consistency without chaining calls.

Automatic failover

When a generation fails, Runflow retries on our infrastructure. Your users never see an error.

JSON prompting with schema validation

FLUX.2's structured prompt format validated at the edge. Wrong shape returns a typed error instead of a wasted generation.

Sentinel analytics included

Every output scored for reference adherence, text accuracy, color match, and artifacts. See your quality trends in a dashboard.

Same price as Black Forest Labs

Per-megapixel pricing mirrors BFL's published rate. Sentinel, dashboards, and auto-failover are included at no extra cost.

Free credits to prove it

Free credits on signup. Enough to run your own reliability test.

MODEL VARIANTS

FLUX.2 dev, 32B parameters.

Every call includes auto-failover and Sentinel quality analytics.

ModelCategory1MP
FLUX.2 [dev] Text-to-image$0.012/MP

Enterprise-grade AI image generation API infrastructure.
Without the enterprise sales cycle.

Your images never touch our storage. Processing happens in isolated environments with automatic cleanup. Compliant by default.

SOC 2 Type II
ISO 27001· In progress (Q3 2026)
GDPR

SOC 2 Type II compliant.

Zero customer data retention

Images are processed and returned. We don’t store your customers’ data.

Isolated processing

Every job runs in its own sandboxed environment. No cross-contamination between customers.

99.9% uptime SLA

Multi-region infrastructure with automatic failover. GPU capacity is always available.

FAQ

FLUX.2 dev API: common questions

Limited credit match

Match your first $500 in credits.

Deposit any amount in your first 30 days and we add a matching credit within 24 hours, up to $500 in total. Put $500, you get $500 back.

Stop losing generations.
Start seeing quality.

Free credits on signup. Match your first $500 in credits, dollar for dollar.