Stop retrying.
Start batching.
Runfra orchestrates distributed consumer GPUs to generate, score, and surface your best results in parallel. Then go further — expand prompts in the playground, upscale to 4× (or 8×) for print, and export. One platform, no infrastructure to manage.
"a tabby cat at a glowing workstation, multiple monitors, cinematic dramatic lighting, ultra-detailed fur, 4k photorealistic"
TOP PICK0.94One request. Five stages of automation.
POST /v1/jobs — prompt, batch_size, quality_mode. API key auth. No SDK. No infra to configure.
Job enters the queue. Available GPU workers each claim one task — distributed across consumer hardware nodes.
Workers generate independently and simultaneously. All batch_size tasks run at once.
Every output is scored. Results below the quality threshold are retried on available capacity before surfacing.
Highest-ranked output returned as best_result_url. Full batch available — all scored, all yours.
Generate, refine, enhance, and export.
The playground takes you from raw prompt to print-ready output in one workflow.
The playground's prompt expansion rewrites your prompt for better composition, lighting, and detail. Optional — one click before you generate.
Real-ESRGAN 4× upscaling on every result. Free for Runfra-generated images. 2 cr for user uploads. Output up to 4096 px.
Optional second-pass 2× upscale on top of the 4× result — 8192 px output. 300 DPI output suitable for large-format printing.
Export your enhanced image at print-ready DPI. Based on the 4× result. Free during Beta.
One endpoint. Any runtime.
No SDKs required. Ship to production from any environment.
POST /v1/jobs HTTP/1.1
X-API-Key: rfa_your_key_here
Content-Type: application/json
{
"prompt": "cinematic sunset over mountain lake,
ultra detailed, 4k",
"batch_size": 4,
"model_name": "stable-diffusion-xl-base-1.0",
"quality_mode": "strict"
}HTTP/1.1 200 OK
{
"job_id": "a1b2c3d4-e5f6-7890-abcd",
"status": "succeeded",
"best_result_url": "https://cdn.runfra.com/...",
"quality_score": 0.94,
"result_count": 4,
"execution_time_ms": 3820
}The Power of the Grid.
Consumer GPU workers, coordinated behind your request.
Runfra coordinates fragmented GPU capacity behind each request. No reserved instances, no manual provisioning. Pay for results, not infrastructure.
Each job is matched to hardware compatible with the requested model. SDXL, FLUX Schnell — assigned at dispatch. You specify the model; routing happens invisibly.
Every output is scored before it surfaces. Results that fall below the quality threshold are retried on available capacity. You see passing results, not raw attempts.
Don't want to wire the API? Use Runfra Studio.
Studio sits on top of the same Runfra API. Pick a product type, fill a brief, get finished mockups, size guides, and watermarked deliverables. No SDK, no code.
Batch image generation from ~$0.02/image.
Prepaid credits shared across Dashboard, API, n8n, and MCP. Buy more credits, pay less per credit. No subscription.
All packs include quality scoring, best pick return, full-batch return, and automatic refunds for failed jobs.