Image catalog refresh: Recraft V4, FLUX.2 Pro, and native SVG

Kyma's image lineup just doubled. The four-model launch from late April is now nine SKUs deep, with three things that weren't possible on the API a week ago: a default model that beats Midjourney V8 in blind preference, a flagship that blends up to ten reference images at once, and a vector path that ships actual editable SVG files.

Every image you see in this post was generated end-to-end on Kyma.

What's new

The four models from the original launch (flux-kontext-pro, ideogram-v3, plus the now-legacy flux-1.1-ultra and recraft-v3) all keep working — existing integrations don't break. New projects should pick from the V4 / 2 Pro lineup above.

The three bigger stories

1. Recraft V4 is now the default

Recraft V4 launched in February 2026 and currently sits at the top of the HuggingFace Text-to-Image Arena leaderboard. It beats Midjourney V8, DALL-E 3, FLUX 1.1, and the rest in side-by-side blind preference voting. The interesting part for Kyma users: the price is unchanged from V3 — a quality jump without a cost jump.

Old recraft-v3 is still there if your code references it explicitly. The system prompt for new code should be model: "recraft-v4".

2. FLUX.2 Pro and the multi-reference blend

Black Forest Labs released FLUX.2 in November 2025. It's a 32-billion parameter model — three times the size of FLUX 1.1 — and it wins about two-thirds of blind comparisons against the previous generation. Two things make it specifically interesting on Kyma:

Multi-reference blending. FLUX.2 Pro accepts an array of up to 10 source images that it composes into a single output. The first six get high-fidelity attention; the rest act as supplementary composition guides. This is brand new on the Kyma API:

curl -X POST https://kymaapi.com/v1/images/generations \
  -H "Authorization: Bearer $KYMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "flux-2-pro",
    "prompt": "blend these references into one cohesive editorial scene",
    "image_urls": [
      "https://example.com/style-a.jpg",
      "https://example.com/composition-b.jpg",
      "https://example.com/lighting-c.jpg"
    ],
    "size": "16:9"
  }'

If you've been hand-stitching reference images into a single grid before sending to a generator, that workflow goes away. One request, three references, one output.

Unified gen + edit. Same model handles text-to-image and image-to-image. Send only prompt for T2I; add image_url (or image_urls) for editing or blending. No model switching.

Per-megapixel pricing. FLUX.2 Pro bills a flat price for the first megapixel plus a per-extra-MP rate, rounded to the nearest whole MP:

At 1MP this is cheaper than the previous flux-1.1-ultra flat $0.081 — and the output is better. The hold at submit time uses the size you requested; the final settle uses the actual output dimensions returned by the provider, so the math always matches reality.

3. Native SVG — only on Kyma

Recraft V4 Vector and Vector Pro return real SVG files. Not raster images saved with a .svg extension; actual paths, layers, and clean geometry that open directly in Figma, Illustrator, or any vector editor. Every other generation model on the market — Midjourney, DALL-E, Imagen, Nano Banana, FLUX — is raster only. This SKU is unique to Kyma's catalog.

curl -X POST https://kymaapi.com/v1/images/generations \
  -H "Authorization: Bearer $KYMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "recraft-v4-vector",
    "prompt": "minimalist Greek psi symbol shaped like a wave, gold on navy"
  }'

The response output.url returns an SVG file with Content-Type: image/svg+xml. Drop it into a website with <img src="..."> and it scales infinitely without pixelation. Open it in Figma and every shape is editable. For logos, icons, illustration sets, and brand systems — this is the missing piece.

Backward compatibility

If you're already calling flux-1.1-ultra or recraft-v3, nothing breaks today. Both stay available, marked as legacy in the docs. Migration is a one-word change:

- "model": "flux-1.1-ultra"
+ "model": "flux-2-pro"

- "model": "recraft-v3"
+ "model": "recraft-v4"

For Recraft V3 → V4: same price, better output. For Flux 1.1 Ultra → 2 Pro: cheaper at 1MP, better output, plus you unlock multi-reference and image edit on the same SKU.

What's next

The image catalog is now nine deep. Next on the audio side: ElevenLabs TTS and Music are scoped, gated only on subscription tier upgrade as Kyma's audio output volume scales. After that, more Google models (Lyria, Imagen 4, Nano Banana family).

If you want to play with multi-reference today, the Kyma Canvas workspace exposes it visually — pick FLUX.2 Pro, select two or more sources from your canvas, and the output blends them in one shot. Same API under the hood.