Image model · deployable on request
Qwen-Image — generation with strong legible in-image text, plus instruction editing — validated on AxForge hardware and deployable on a dedicated European system for your traffic only. Not on the shared API today: there, generation runs on ERNIE Image Turbo and editing on FLUX.2 Klein.
Specifications
| Model | Qwen-Image — generation + instruction editing (Qwen-Image-Edit), strong in-image text |
|---|---|
| License | Apache 2.0 (open weights) |
| Hardware | Your own NVIDIA DGX Spark (GB10, 128 GB unified memory), monthly rental |
| Endpoint | OpenAI-compatible /v1/images/generations and /v1/images/edits on your machine |
| Region | EU — Sweden today, more regions in deployment |
| Pricing | On request — talk to an engineer |
Validated in production on our own nodes through 2026-08 — this is the model that served the shared API's generation until our seed-matched bake-off replaced it. It remains an excellent dedicated-deployment choice when you want generation and editing from one model family on one machine.
Data & privacy
On a dedicated system, prompts and images are processed in memory in the EU — not written to disk, not logged, not retained, and never used to train anything. The full policy is at axforge.ai/privacy.
FAQ
Not anymore. On the shared API, generation runs on ERNIE Image Turbo and editing on FLUX.2 Klein 4B — both picked in seed-matched bake-offs on our hardware. Qwen-Image remains validated and deployable on request on a dedicated system.
Strong legible in-image text and high-quality generation, plus instruction editing via Qwen-Image-Edit. A dedicated deployment gives you both on your own machine.
A DGX Spark (GB10, 128 GB unified memory) in an EU region running Qwen-Image for only your traffic, on a monthly rental with an OpenAI-compatible endpoint. Pricing on request.
No. Prompts and images are processed in memory in the EU, never persisted or used for training — see the privacy policy.