fal AI Image Pricing: Per Image vs Per Megapixel

fal image models do not all use the same billing unit. Some charge a flat rate for each generated image. Others charge for the output area in megapixels. To compare them honestly, fix the number of outputs and the intended resolution, then preflight the exact model input. A cheap rate attached to the wrong unit is not a useful budget.

Here is the short decision rule:

Do not assume that raw width multiplied by height is the provider's final billing quantity. An endpoint can apply minimums, rounding, presets, quality multipliers, or a different meter. fal's pricing documentation says image models commonly charge per image or per megapixel, while the current fal pricing page normalizes selected image prices to 1 MP for comparison.

If an AI agent will choose the model, make the live catalog and exact quote part of the workflow. Aident Loadout can list the current fal model contracts, expose each pricing unit, and preflight the exact request before a provider generation runs.

Current fal Image Pricing Snapshot

The table below is a bounded snapshot from Aident Loadout's live staging catalog on September 13, 2026. It is not a quality ranking or a complete fal model catalog. The comparison includes only entries that returned a directly comparable dollar rate per image or per megapixel.

The “output from $1” column is only unit-price arithmetic. It does not compare image quality, latency, retries, commercial fit, or every optional model feature. For per-megapixel models it means one dollar of billable megapixel area, not necessarily that many finished images.

The same live response included models whose price was represented as provider-specific units or a historical call estimate. Those entries were excluded instead of forcing unlike meters into one chart. That is the safe behavior for an automated selector: preserve an unknown comparison rather than invent a conversion.

Per-Image Pricing Is Simple Only at the Base Rate

A flat per-image model makes the first calculation easy:

base cost = price per image x number of images

At the observed $0.01 base rate, 1,000 Muse outputs would begin at $10. At the observed $0.04 base rate, 1,000 Fibo Gen 1.5 outputs would begin at $40.

“Per image” does not guarantee that every configuration costs the same. A model can charge more for a larger resolution, higher quality, special mode, or additional generated output. Inspect the endpoint fields and quote the selected configuration rather than multiplying a marketing-page number across a production batch.

Per-image pricing is easiest to forecast when:

• the model's resolution and quality choices are fixed;

• one request always asks for the same number of outputs;

• retries and rejected generations are measured separately;

• the endpoint's current page and API contract agree; and

• the final settled cost is captured after execution.

Per-Megapixel Pricing Connects Cost to Output Area

For a model priced by output area, the rough calculation is:

base cost = price per billable MP x billable MP per image x number of images

One 1024 x 1024 image contains about 1.05 million pixels, while one 2048 x 2048 image contains about 4.19 million pixels. The provider may normalize, round, or map presets before billing, so those raw figures are planning inputs rather than the final invoice quantity.

This distinction matters at scale. A one-megapixel assumption applied to a four-megapixel production output can understate the media budget by roughly four times before retries or variants are counted.

Per-megapixel pricing is useful when output sizes vary because the cost changes with the generated area. It also makes resolution a policy input. An agent should not silently raise image_size to improve detail when that decision can multiply spend.

Two Exact Preflights, No Provider Generation

The live Aident contract separates a free model-list read from the generation Action. First, discover and inspect the current Actions:

aident account auth status
aident vault status --integrationId "direct:fal"

aident capabilities search \
  --query "list current fal text to image models with pricing" \
  --types '["action"]' \
  --targetEnv staging

aident capabilities get \
  --name "direct:fal:fal_list_text_to_image_models"

aident capabilities execute \
  --name "direct:fal:fal_list_text_to_image_models" \
  --input '{"query":"flux","limit":10}'

aident capabilities get \
  --name "direct:fal:fal_text_to_image"

Copy the exact public Action name returned by discovery. Do not treat the remembered names in this article as a permanent API contract.

On September 13, these two bounded inputs passed preflight:

aident capabilities preflight \
  --name "direct:fal:fal_text_to_image" \
  --input '{
    "model":"fal-ai/flux-2/flash",
    "prompt":"A blue ceramic sphere on a coral plane",
    "model_input":{
      "image_size":"square_hd",
      "num_images":1,
      "output_format":"png"
    }
  }'

The exact preflight quote was 0.5 Aident credits, or $0.005, for one provider-estimated megapixel.

aident capabilities preflight \
  --name "direct:fal:fal_text_to_image" \
  --input '{
    "model":"meta/muse-image/text-to-image",
    "prompt":"A blue ceramic sphere on a coral plane",
    "model_input":{
      "aspect_ratio":"1:1",
      "num_images":1,
      "output_format":"jpeg"
    }
  }'

That exact preflight quote was 1 Aident credit, or $0.01, for one provider-estimated image. Neither request was executed, so the test did not evaluate image quality, latency, provider acceptance, or settled billing.

Preflight is the budget boundary, not authorization to generate. Review the prompt, model, output count, size, quote, and intended use before supplying any confirmation token or dispatching the Action.

Estimate a Real Batch Instead of One Ideal Call

Suppose a workflow needs 2,500 approved one-image generations each month and historically requires 20 percent additional generations for revisions or retries. The planning quantity is 3,000 outputs, not 2,500.

At the observed one-unit preflight assumptions:

This does not establish that FLUX 2 Flash is the better model. A lower generation rate can lose its advantage if it needs more retries, cannot satisfy the creative brief, lacks a required format, or produces assets that require manual repair.

A useful budget therefore tracks:

1. requested outputs;

2. generated outputs, including retries and variants;

3. accepted outputs after review;

4. billable megapixels or images;

5. quoted cost before dispatch;

6. settled cost after success; and

7. cost per accepted output.

Cost per accepted output is the stronger operating metric. It joins model price to the result the workflow actually keeps.

Choose the Model by Constraints, Then Compare Price

Price is one selection field. Evaluate the endpoint against the real job:

• Does it support the required aspect ratio and resolution?

• Does it produce the needed PNG, JPEG, WebP, vector, or transparent output?

• Does the workflow require accurate typography, style references, LoRAs, or structured prompts?

• Are the model's safety controls and commercial terms acceptable?

• Does one reviewed output pass the visual acceptance criteria?

• Is latency acceptable for an interactive call or batch queue?

• Can the system reproduce the result with the recorded model, input, and seed?

For example, the observed Fibo Gen 1.5 contract exposed 1 MP and 4 MP resolution choices plus structured prompts. The Recraft V4 Styles contract exposed style references and raster output. The FLUX 2 Flash contract used a per-megapixel rate and several image-size presets. Those are different production contracts, not interchangeable rows in a price leaderboard.

If you are deciding between a fal-native creative agent and a broader capability layer, use the separate fal Agent vs Aident Loadout comparison. This page owns the narrower pricing-unit and budget job.

For the complete account, Vault, discovery, schema, preflight, execution, and Audit sequence, continue with How to Use Aident Loadout.

A Safe Image-Generation Cost Policy

Give an automated media workflow explicit limits:

1. List the current models before selecting a non-default endpoint.

2. Allow only reviewed model IDs and output formats.

3. Set maximum images and resolution for the first call.

4. Preflight the exact input and stop when the quote is unavailable or above the ceiling.

5. Require a fresh quote whenever the model, size, quality, or image count changes.

6. Execute one test output before approving a batch.

7. Record accepted-output rate and settled cost before increasing volume.

The risk-versus-spend approval workflow explains why a low price does not remove content, privacy, or publication risk. Cost approval and consequence approval are separate decisions.

Compare One Live fal Image Quote

Follow https://aident.ai/SETUP.md

Set up Aident Loadout, list the current fal text-to-image models, choose one reviewed endpoint, and preflight one bounded image without dispatching it. Success means you can explain the pricing unit, exact input, expected output count, and quote before the provider runs.

Set up Loadout and compare one live fal image quote

Sources

• fal pricing, reviewed September 13, 2026.

• fal model API pricing documentation, reviewed September 13, 2026.

• FLUX 2 Flash API, reviewed September 13, 2026.

• Meta Muse Image API, reviewed September 13, 2026.

• Fibo Gen 1.5 API, reviewed September 13, 2026.

• Recraft V4 Styles API, reviewed September 13, 2026.

• Aident Loadout setup, reviewed September 13, 2026.

• Live Aident Loadout staging model catalog and two exact preflights, inspected September 13, 2026.

Refresh this guide when fal changes image model prices, billing units, endpoint fields, output-size rules, or model availability, or when Aident changes the Fal catalog, quote, confirmation, credit-conversion, or settlement contract.