Image generation on Okou
Describe the picture and get the file back. Several image models, no image account of your own, charged per image.
Generation · No setup · Managed credits per successful call
Ask for a picture in a message and a file comes back. Set the size and how much quality you want to pay for, ask for a transparent background when the image has to sit on something else, or hand it an image to work from instead of starting blank.
More than one model is available, because they are good at different things, and picking between them is part of the request rather than a separate integration. Okou holds the accounts and charges each image in credits, so nobody on your team keeps an image API key.
What Image generation is
Getting one usable image out of a model is easy. Getting fifty in the same style, on a schedule, into a page or a deck, is where the work actually is, and that part is a pipeline rather than a prompt.
This is the generation step of that pipeline as something an agent can call. A prompt goes in, a file comes out, and the agent can go on to put it in a report, a post or a hosted page without anyone downloading anything.
Which model draws the picture matters. Some are better at clean graphic work, some at photographic scenes, some at editing an image you already have, and some are simply cheaper for a draft. The choice is part of the request, and the quality setting decides what a picture costs.
It is the same thing people shop for as a text to image API. The difference is where the account sits: no signup, no key in an environment variable, no separate invoice, and access granted per agent rather than shared as a secret.
What Image generation can do
What an agent can do with it, and what comes back when it does.
Coverage and limits
What it does not do, said up front, so nobody plans a workflow around something that is out of scope.
Quality is a price decision
The same prompt costs very different amounts depending on the quality level and the model. A draft is a few credits and a finished asset can be a few hundred, so a workflow that generates in bulk should draft cheaply and finish selectively.
Text inside pictures is unreliable
Models have improved at rendering words, but none of them are typesetting. Anything where the wording has to be exact belongs in HTML or a design file with the image behind it.
Consistency takes a reference
Asking twice for the same style gives you two cousins, not twins. Holding a look across a set means giving the model a reference image or a written style to follow every time.
Content rules still apply
The models refuse what they refuse, and that includes likenesses of real people and copyrighted characters. A workflow that depends on getting one of those through will not run.
What Image generation costs
Web services are billed in credits per call, not per token. There is no separate vendor bill to reconcile.
OpenAI, Google and fal.ai modelszero generate imageCharged per image in credits, and the number depends on the model, the size and the quality level you asked for. A low-quality draft is a handful of credits; a large, high-quality image from a top model is a few hundred. There is no image API subscription underneath and no monthly minimum, so a team that makes twelve pictures a month pays for twelve pictures.
Product names and logos belong to their owners and appear here only to say what a service runs on. They do not imply any endorsement or partnership.
Setup and access
Nothing to set up
Nothing to connect. No image model account, no key to paste, no separate billing to reconcile. If you would rather run a particular provider on your own account, Okou also ships connectors for several of them, and the built-in service is what you use when you would rather not have the account at all.
Who can use it
Generating a file is a permission you grant per agent and per person. It is worth being deliberate about, because it is the permission that spends credits fastest when a workflow loops.
What teams use Image generation for
Art for things that are already automated
A blog post, a release note, a weekly report. The agent that writes it can draw the header instead of leaving a placeholder for someone to fill in on Monday.
Filling a set that has to look related
Empty states, category icons, section headers. Give the style once, generate the set, and regenerate the two that came out wrong rather than the whole batch.
Working from something you already have
A rough sketch, a screenshot, a product shot on the wrong background. Image to image is usually closer to what people actually need than starting from a blank prompt.
When not to use Image generation
Skip it when the wording inside the image has to be exact, and build that in HTML instead. Skip it when a real photograph is the point, because a generated one is a different claim. And skip it for high-volume generation without a quality decision, which is the fastest way to spend credits on pictures nobody looks at.
What this is called elsewhere
The same thing goes by several names in the market. If you have shopped for one of these, this is how it maps to what you get here.
Text to image API
A prompt in, an image file out, over an interface a program can call. That is this, with the account and the key on our side.
AI image generator without an account
What people usually mean when they search for one: no signup, no key, no separate plan. You ask in a message and the file comes back.
Image editing API
Image to image and mask-guided edits, which is the half of the job that starts from an asset you already own rather than from nothing.
Transparent background image API
Generating with no background so the result drops onto a page, a card or a deck. Supported by some of the models, and asked for per request.
Batch image generation
Making a set rather than a picture, which is where the agent earns its keep: same style, one loop, and a per-image charge you can see.
Image generation compared
Image generation vs an image model account of your own
Same models, different plumbing. On your own it is one account per provider, a key each, and a bill each, and switching models is an integration. Here it is one request with the model named in it, and one line of credits.
Image generation vs a design tool
A design tool is where a person makes something on purpose, with type, layout and a brand. This is where a workflow produces an image without a person in the loop. Most teams need both, and the mistake is asking this one to do layout.
Image generation vs a stock library
Stock is cheaper per picture, licensed and predictable, and it looks like stock. Generation is worth it when the picture has to be about your specific thing, or when nobody has time to search.
The short version
The way to get a picture into an automated workflow. Draft cheap, finish selectively, and keep the exact wording out of the image.
Frequently asked questions
Do I need an OpenAI or fal account?
No. Okou runs the generation on your behalf and charges it in credits. There is no key of yours anywhere in the workflow.
Which models can I use?
Several, across providers, chosen per request because they are good at different things. The agent picks a sensible default and you can name one instead.
What does an image cost?
Credits per image, depending on the model, the size and the quality level. Drafts are cheap enough to iterate with; large high-quality images are the ones worth being deliberate about.
Can it edit an image I already have?
Yes. Give it the image and describe the change, and use a mask when only part of the picture should move.
Can I get a transparent background?
Yes, on the models that support it. It is worth asking for whenever the image has to sit on a coloured surface.
Will two runs give me the same style?
Not on their own. Consistency comes from giving the model the same reference or the same written style each time, which is easy for an agent to do and easy for a person to forget.
Can I use the images commercially?
The images are yours to use, subject to the same content rules the models apply to what they will draw. Real people and copyrighted characters are the usual refusals.
How is this different from the image connectors?
A connector works against your own provider account when you already have one. The built-in service is ready to use with nothing to configure, which is what most teams want when they just need the picture.
How you ask for Image generation
You describe what you want in plain language. The agent works out which service it needs and makes the calls.
“Write the release note for this week and draw a header image for it in the same style as the last three.”
“Draw six category icons for our help centre, flat, one accent colour, all clearly from the same family.”
“Take this product photo, put it on a clean neutral background, and give me a square version for the store listing.”