#1 in generative media apis · the default, and the only one with a media catalogue this deep
91/100
verdictthe most complete media catalogue with per-output pricing printed on every model page — you pay a visible aggregator markup for it, and your credits expire.
fal vs Novita AI
Novita AI
fal
price
$0.018 / image for flux.1 kontext dev in fast mode
$0.025 / megapixel for flux.1 [dev]
free tier
no
no
flux.1 [dev]
not carried
$0.025 / megapixel
billing
per token, per image, per second
per output; per gpu-hour for deployments
catalogue
200+ models
1,000+ media models
llms too
yes
no — media only
mcp server
none published
none published
switch forteams shipping an image or video feature who want the widest model choice and the least thinking
pros
+over a thousand media models, and new releases land fast
+per-output price printed on every individual model page
+matches black forest labs direct pricing with no markup
+the de facto reference other platforms are compared against
cons
−roughly 12-25% markup on google models versus going direct
−media only — no text or reasoning models
−purchased credits expire after 365 days, promotional after 90
−422 errors can still be billed when gpu time was spent, and one model page states policy refusals are charged
#2 in generative media apis · the only one that tells you what a job costs before you run it
84/100
verdictthe best alternative to fal if billing mechanics matter more to you than catalogue size — it prices every job up front and bills only completed runs, on a catalogue a third the size with almost no community.
Wiro vs Novita AI
Novita AI
Wiro
price
$0.018 / image for flux.1 kontext dev in fast mode
$0.006 / request for flux-2-dev
free tier
no
no
flux.1 [dev]
not carried
$0.006 / request (flux-2-dev)
billing
per token, per image, per second
six methods, declared per model
catalogue
200+ models
538 models
llms too
yes
yes — 93 llm and chat
mcp server
none published
yes — mcp.wiro.ai/v1
switch foragents and pipelines that need to check the cost of a job before committing to it
pros
+dynamicprice and approximatelycost expose a job's cost before you run it
+billed only for completed runs — server errors and cancelled tasks are free
+media, llm, chat and realtime behind one api, unlike fal
+flat per-request pricing on its own endpoints — $0.006 for flux-2-dev at any resolution
+official mcp server, so an agent can drive the catalogue directly
cons
−538 models against fal's 1,000+ and Replicate's community library
−concurrency is capped at 10% of your account balance below $250
−no public pricing page — wiro.ai/pricing returns a 404
−small community, so undocumented edges mean a support ticket
#3 in generative media apis · the community library, and the best story for your own model
83/100
verdictthe widest long tail and the only first-class path for packaging your own model, with private-deployment billing that charges for time your model spends doing nothing.
Replicate vs Novita AI
Novita AI
Replicate
price
$0.018 / image for flux.1 kontext dev in fast mode
$0.025 / image for flux.1 [dev]
free tier
no
no
flux.1 [dev]
not carried
$0.025 / image
billing
per token, per image, per second
per output; per second for private deploys
catalogue
200+ models
thousands, community-contributed
llms too
yes
yes
mcp server
none published
none published
switch forteams running a fine-tuned or niche community model rather than a frontier one
pros
+the largest community model library in the category
+Cog makes deploying your own model a first-class path
+public models billed per output at $0.025/image, comparable with fal
+covers text models as well as media
cons
−private deployments bill idle and setup time separately
−no exact catalogue count published, only 'thousands of models'
−community models vary wildly in maintenance and documentation
#4 in generative media apis · the cheapest headline number in the category, and you cannot check it
75/100
verdictgenuinely aggressive pricing on a large catalogue, undercut by the fact that you cannot confirm what any specific model costs without signing up.
Runware vs Novita AI
Novita AI
Runware
price
$0.018 / image for flux.1 kontext dev in fast mode
$0.0006 / image at the floor
free tier
no
yes
flux.1 [dev]
not carried
not publicly priced
billing
per token, per image, per second
per output, varies by model and settings
catalogue
200+ models
303 image, 38 llm listed
llms too
yes
yes — 38 models
mcp server
none published
yes — runware mcp
switch forhigh-volume image generation where you can benchmark your own model choice first
pros
+lowest published price floor in the category at $0.0006/image
+own inference stack rather than resold gpu capacity
+$2 in free credits on signup
+native mcp server referenced on the pricing page
cons
−no public per-model price — flux.1 [dev] could not be priced from three separate pages
−the quoted range spans the entire catalogue, so it prices nothing specific
−confirming any real cost appears to require an account and the playground
#5 in generative media apis · half fal's price on the anchor, a fraction of the catalogue
73/100
verdictless than half what fal charges on the one model we can compare directly, on a catalogue small enough that you should check your models are on it first.
WaveSpeed AI vs Novita AI
Novita AI
WaveSpeed AI
price
$0.018 / image for flux.1 kontext dev in fast mode
$0.012 / image for flux.1 [dev]
free tier
no
yes
flux.1 [dev]
not carried
$0.012 / image
billing
per token, per image, per second
per output, pay-as-you-go
catalogue
200+ models
40+ listed on pricing
llms too
yes
yes
mcp server
none published
none published
switch fora known, stable set of models where the per-image saving compounds
pros
+$0.012 per flux.1 [dev] image — less than half fal or Replicate
+covers image, video and text models
+$1 free credit on signup, no card required
+pay-as-you-go with no monthly commitment
cons
−catalogue is roughly forty listed models with no published total
−custom deployment is enterprise-tier only
−no cold-start or latency figure we could verify independently
every tool on this page went through the same test as Novita AI — same tasks, same order, scored the same way. the comparison tables are the figures from that testing, not vendor spec sheets.