Image generation is live on Kyma

Kyma now generates images. The same API key that hits chat/completions also hits images/generations. Four models, per-image billing, one ledger.

Every image in this post was generated on Kyma. The four-muse hero, the editorial keysmith, the typography mark, the vector logo - all dogfood, no stock.

What's available

Pricing is per call. No subscription, no per-second compute charges. Your image-gen calls draw from the same balance as your text calls.

Showcase

Each of the three demos below was generated end-to-end on Kyma. The prompts and their outputs:

FLUX 1.1 Ultra - photoreal hero

Prompt: cinematic wide shot, an artisan crafting an ornate brass key on a workshop bench at golden hour, warm rim light pouring through dusty window, shallow depth of field, 35mm film aesthetic, photorealistic. Roughly 12 seconds, $0.081.

Look at the rim light, the depth of field, the way the apron catches the window glow. This is the photoreal frontier - editorial-grade output you'd put above the fold without retouching.

Ideogram V3 - typography that renders

Prompt: bold poster typography centered on dark navy background reading "IMAGE GENERATION" in large serif metallic gold letters, below it the word "on Kyma" in elegant smaller italic.

This is the model's actual differentiator. Most image models butcher text. Ideogram renders it - kerning intact, weight intact, no melted glyphs.

Recraft V3 - vector that ships

Prompt: flat vector illustration of an abstract psi symbol shaped like a wave, gold and deep navy color palette, modern minimalist tech brand logo on white background.

Vector-clean, brand-consistent, no painterly artifacts. The kind of output you'd actually ship as a product mark.

Why these four (and not twenty)

We picked four models that don't overlap. Each one does a thing the others can't:

• •FLUX 1.1 Ultra is the photoreal frontier. Editorial-grade fidelity, deep cinematic looks, the kind of image you'd put above the fold.
• •Ideogram V3 renders text correctly. If you're making posters, packaging, app store screenshots, or anything with words baked in, this is the one model that doesn't hallucinate the letters.
• •Recraft V3 is the vector specialist. Brand marks, illustration, flat geometric work. Style presets that hold across a series.
• •FLUX Kontext Pro edits an existing image. Mask-based inpainting, refinement, "make this version of the photo but change X."

Adding a fifth photoreal model wouldn't have given anyone a real choice. We'd rather route four sharp tools than twenty fuzzy ones. If something obviously different ships (video, audio-to-image, a true vector-from-scratch model with better quality), we'll add it.

How it works

Same OpenAI-compatible base URL. Same key. POST to /v1/images/generations, get a job_id back, poll GET /v1/jobs/{id} until succeeded. Most images finish in 10-30 seconds.

A few behaviors worth knowing:

• •It's async. Submit returns a job_id and a 202 immediately, so you don't hold a long HTTP connection.
• •Pricing is held at submit, charged at success. Kyma places a hold of the per-image rate at submit time. If the upstream fails, the hold is released and you never pay for a failed generation.
• •Output URLs are stable. We rehost the result on our own blob storage so the URL doesn't expire when the upstream cleans up its cache.
• •Kontext needs an image_url. It's the edit model. Send a base image plus your edit prompt - sending text-only triggers a refund, not a charge.

Full request shape, parameters, polling, and SDK examples are in the image generation API reference.

Use cases this unlocks

Concrete patterns we've seen builders ask for:

• •Marketing automation. Generate per-product hero images on demand instead of paying a designer per asset.
• •App-internal illustrations. Empty states, onboarding visuals, error pages - Recraft V3 keeps them on-brand.
• •Content pipelines. Blog cover images for every post. This post's hero is Exhibit A.
• •Logo and brand exploration. Ideogram and Recraft can turn a brief into a dozen mark variations in under a minute.
• •Photo editing at scale. Kontext for batch refinement, background swaps, style transfers.

Pricing in context

For perspective: a one-line of GPT-5.4 reasoning on a hard problem costs more than a FLUX Kontext edit. A typography poster from Ideogram costs less than a single completion call to Claude Opus 4.6 on a 30K-token context.

If you're already buying credits for text models, image generation slots in without a second budget. If you're starting fresh, the $0.50 free signup credit covers about six FLUX hero shots or nine Recraft vector pieces.

What's next

Phase 2 is on the roadmap: video generation, more granular aspect-ratio control, and a streaming-style "preview during generation" mode. We'll publish dates when we have provider parity verified.

For now, this is what's live. Same key, four models, per-image pricing, no markup mystery.

Links

• •Image generation API reference
• •Image models page
• •Pricing
• •Sign up - $0.50 free credit